Subscribe now

Share

Generational gaps are one of the great running jokes of human civilization, and they have always followed roughly the same script. The older generation watches the younger generation embrace something incomprehensible, declares with absolute certainty that this new thing will be the ruin of society, and then gradually makes peace with it just in time to be horrified by whatever comes next. Elvis Presley’s hips were, at one point, considered a legitimate threat to American moral fabric, and the cameras at the Ed Sullivan Show were instructed to film him only from the waist up. Rock and roll was satanic, then disco was the death of music, then hip hop was the death of music, and then Spotify playlists became the death of music. There was a stretch of years in the 1980s when serious adults genuinely believed that Dungeons and Dragons was a recruitment pipeline for the occult, which is a sentence I cannot type without a small smile, having spent my own formative years rolling polyhedral dice at the kitchen table. Video games were going to produce a generation of unemployable zombies. Texting was going to destroy the English language. Avocado toast, somehow, was responsible for an entire generation’s inability to buy houses. The young have always been ruining things, and the old have always been there to point it out, and the historical record of these complaints turns out to be both consistent and consistently wrong.

The current generational gap, however, is not about taste, and that is the part that makes it different. Every previous version of this conflict was an argument about sensibility. The current gap is about something far more consequential, which is to say it is about value, in the literal economic sense, and the inability of two generations to bridge it is the single largest reason artificial intelligence is failing to deliver on its actual potential inside most organizations right now.

The structure of the disconnect is shrouded in the uncertainty that we are all trying to adapt to, with the art of what is objectively possible changing on almost a daily basis. The people who most deeply understand what critical business opportunities lie, the ones who have spent decades developing the strategic judgment to recognize a real opportunity from a fashionable one, are largely the people who do not know how to use the AI tools transformationally, outside of their “old model” indoctrinated thinking. And the people who know how to use the tools creatively, who can spin up something genuinely capable in an afternoon and who think in prompts and pipelines as a kind of native language, are largely the people who do not yet have the experience to know what to apply them to. Two populations, sitting in adjacent rooms, each holding precisely what the other lacks, and most organizations have not figured out how to bring them into the same conversation in a way that actually unlocks the spread between them.

I have come to call this gap Generational Arbitrage, and the more time I spend working with it directly, the more convinced I become that closing it is going to be the defining business challenge of the next 5 years.

The term comes from finance, where arbitrage describes the practice of capturing value from a difference between two markets where the same asset is priced or valued differently. The conditions are always the same regardless of the asset class, which is to say there has to be a real gap, the gap has to be temporary, and someone has to recognize it and act before everyone else does. What we have at this particular moment in professional life, across consulting and most knowledge-based industries, is exactly that kind of spread, and it is unusually wide. On one side of canyon you have seasoned masters with thirty years of accumulated wisdom, judgment, pattern recognition, and the ability to walk into a room and immediately diagnose what the actual problem is rather than the problem the client believes they have. They have, in other words, the part of the equation that no amount of compute can replicate, but they are reluctant operators of the new tools and many of them are quietly hoping the whole disruption will turn out to be less significant than it looks. On the other side you have a generation of young professionals who are actively growing up with these tools as instruments rather than novelties, who can build at a velocity that would have seemed absurd a few years ago, and who carry exactly none of the scar tissue that teaches a person why something that looks promising in the demo will fall apart the moment it meets a real client. Both sides are sitting on assets the other side cannot fully value, which is the textbook definition of arbitrage conditions, and the spread between what each holds and what neither can capture alone is enormous.

The reason most organizations are not capturing this value is structural rather than technical. Senior people sit in one set of meetings, the AI work happens somewhere else, usually staffed by junior practitioners or a specialized team that reports to a chief technology officer who does not sit at the table where the real strategic decisions are being made. The seasoned master receives the AI output as a finished thing, evaluates it through the lens of the old paradigm, and either rejects it outright or accepts a watered-down version of what it could have been. The young AI native produces capability without context, watches the output get diluted by people who never quite understood what they were looking at, and concludes that the company is not serious about transformation. Both perspectives are partially right and entirely stuck, which is what happens when an organization is sitting on top of a real arbitrage opportunity but has built its meeting structure to keep the two sides of the trade physically separated from each other.

I have been running a different version of this for a while now, and the version I want to describe is what I have come to call the Tandem Model. The pattern is straightforward in its mechanics and surprisingly potent in its results, which is to say I have been deliberately pairing dynamic young AI professionals with seasoned business and functional experts who are far less comfortable with the tools, and putting them in the same room to discover the art of the possible together. Not as mentor and mentee, which would simply replicate the old hierarchy in a new outfit. Not as senior and junior, which would replicate the old delivery model. As a tandem, which is to say two people moving in the same direction, each carrying something the other cannot, neither able to make the journey alone. One person carries the experience, the judgment, and the ability to recognize what would actually move the business forward. The other carries the speed, the creative fluency, and the willingness to try things that would not occur to anyone who has been in the industry long enough to know all the reasons they should not work. The magic happens in the conversation between those two perspectives, because each one alone produces an obvious answer and the two together produce something neither would have arrived at independently. It is always a give and take, and the give and take itself is the work, which is to say the friction between the seasoned voice asking what should be done and the AI-native voice asking what could be done is precisely where the ten-times paradigm shift lives.

For readers who have been following these essays, the Tandem Model is a significant piece of the larger Afterlife Playbook I have been working through across earlier pieces, and I will continue to develop it in subsequent ones. What I want to focus on here is the broader pattern it sits inside, because the unique organization dynamics is not actually new. This one is particularly impactful because the technology shift is unusually large, but once you start looking for the pattern you find it everywhere underneath the most consequential business breakthroughs of the last twenty years.

Almost every revolutionary business idea in modern memory has come from connecting two things that nobody had previously thought to combine. Uber is the cleanest example and the one most people forget the actual mechanics of, because the breakthrough was not the ride-hailing application itself but the recognition that GPS was already in everyone’s pocket and that the taxi experience was already broken, and that these two facts could be joined into a single product that obsoleted an entire industry. Garrett Camp did not invent global positioning, and he did not invent dissatisfaction with hailing a cab on a cold night in Paris. He saw that they belonged together and that nobody else had walked across the bridge between them yet, and the act of walking across that bridge was worth tens of billions of dollars within a decade.

Spotify did the same thing again with music, this time bridging licensing infrastructure and recommendation algorithms in a way that the established players in either domain could not see from where they were standing. The labels thought they were in a rights business, and the algorithm engineers thought they were in a software business, and Daniel Ek saw that the listener did not actually want either thing in isolation. The listener wanted a playlist that knew them personally, served by a system that had paid the artists fairly, and the entire industry reorganized around that synthesis within a decade despite the fact that everyone involved was already operating with most of the necessary pieces already in hand.

The pattern in these and other similar cases is the same, and it is the same pattern I have been seeing emerge from the tandem pairings at smaller scale across the work I am doing now. Revolutionary value does not come from going deeper into a single discipline. It comes from connecting two disciplines that had been kept apart for reasons that turn out, in retrospect, to be mostly accidental, or even circumstantial.

What I am proposing, and what I have been building in practice for some time, is the macro version of this pattern applied deliberately rather than accidentally. Take the deep mastery that seasoned professionals have accumulated over decades, the judgment and the strategic instinct and the leadership and the critical thinking that no model is going to replicate any time soon, and pair it directly with the dynamic creative fluency of professionals who think in AI as their first language rather than as a tool they had to learn later. Put them together in front of a real problem and let them argue, sketch, prototype, and iterate as equals. Not to optimize the old way of doing things, although that is a pleasant side effect. To revolutionize what the work even is in the first place, which is what every cross-domain synthesis in the historical record has actually accomplished.

The arbitrage opportunity is real, the spread is wider than at any point in my professional lifetime, and the window is open right now in a way that will not stay open indefinitely. Both sides of the gap are sitting on assets that are dramatically more valuable when joined together than when held in isolation, and the organizations that figure out how to bring those two populations into the same room, into the same conversation, into the same working tandem, are going to look in ten years the way Uber looked to the taxi industry in 2012. The ones that do not are going to look the way the taxi industry itself looked, which is to say defensively confused about what just happened to a business they had spent their entire careers mastering.

The bridge is right there, fully constructed, waiting to be walked across. Most organizations are still standing on one side of the canyon, looking across, wondering why the other side feels so far away when in fact it is not far at all. It just needs the tandem.

The sentence that stopped me was a small one, and it came in the middle of a completely unremarkable exchange.

I had been working through a technical productivity problem for the better part of a morning, the kind of problem that is not hard in any interesting way but is tedious in every practical way, with configuration settings that would not hold and a permissions issue I had already thought I had solved twice. I had been firing off short, clipped prompts to an AI assistant, and somewhere in the chain of my increasingly terse messages, it picked up on my mood. The reply came back with a perfectly calibrated sentence, in which it said it was sorry I was dealing with this and that it knew this kind of thing could be tough.

For a fraction of a second, I felt the small human flicker that sentence is designed to produce, and something in my posture softened at the sense that someone, somewhere, understood.

And then the other half of my brain caught up, which is what happens when you have spent a career selling advice and you know what advice is actually made of. The AI was not sorry and the AI did not know, because it had never spent a Saturday morning watching a deadline slide away, had never sat across from a client who was losing patience, and had never felt the particular shame of being the expert in the room who cannot make the expert thing work. It had produced a string of tokens statistically likely to resemble sympathy, and it had done so very well.

I appreciated the gesture in the way you appreciate a good actor, but when AI is trying for empathy, we are forced to confront the nature of what meaningful empathy actually is. That there is a category of value we regularly seek from other human beings that has almost nothing to do with the correctness of the information they convey. It has to do with the price they paid to know what they know, and that price is something AI, by its very nature, has never paid and cannot pay.

The weight behind the words.

Consider what happens when a good friend who has been through a divorce tells you that your divorce is going to be survivable. The words themselves are almost irrelevant, because you could get the same words from a book, from a stranger, from a greeting card, or from an AI. What makes the words land is that the person saying them is standing on the other side of the thing you are currently drowning in, and they got there by paying the same price you are paying now. They are not offering you information so much as they are offering you evidence, in the form of their own living body, that the thing is survivable, and that is not a small gift but rather most of what meaningful advice actually is.

A therapist’s insight carries weight because the therapist is a human being who has also been afraid, also been ashamed, also lost people, and who will also, eventually, die. A mentor’s career guidance carries weight because the mentor has personally risked something and paid the cost of being wrong. A friend’s consolation carries weight because your friend is a fellow traveler in a condition that both of you are mortally bound to. In every case, the weight comes from the fact that the person giving you the information has skin in the game of being human, and they are offering their hard-won perspective as a kind of gift from one mortal to another.

AI has no skin in this game and cannot have any, because it risks nothing by being wrong about your grief and pays no price for its perspective given that it has no perspective, in the sense that having a perspective requires having once been somewhere you might not have survived. When an AI says “I know this can be tough,” the statement is a sentence issued by something that has no idea what tough means, and cannot, in principle, ever find out.

The suspension of disbelief problem.

The obvious objection is that we extract real emotional value from all sorts of sources that are not themselves conscious, because we cry at novels and fall in love with characters in films, and we find genuine comfort in songs whose lyrics were written by someone who may have been faking the feeling or writing it on assignment. We are evidently quite capable of generating real emotional experiences in response to things that are not, in any strict sense, emotionally there, and this is the suspension of disbelief argument, which is close to right and also, finally, wrong in a way that matters.

Notice what is actually happening in those cases. The novel is fictional, but the novelist is real, and the novelist has paid a price for the insight that the fictional characters are transmitting. The film’s characters are not real, but the screenwriter and the director and the actors are, and they are drawing on their own actual lives to produce something that resonates with yours. When you cry at a novel, you are not being moved by a fiction but rather by a real human being who used a fiction as a delivery mechanism for something true about being alive.

With AI, the source itself is the fiction, and that is a categorical difference. When an AI writes you a moving passage about grief, there is no grieving human anywhere in the chain and no one who paid for the sentence with their own life, because there is only a very sophisticated pattern generator producing what the grieving humans of history have tended to produce, without having ever been one of them. It is the difference between reading a letter from a widow and reading a perfect forgery of a letter from a widow, written by someone who has never loved or lost anyone and will never do either.

The forgery can still move you, which is the uncomfortable truth, but something has happened in the transaction that is worth caring about once you slow down enough to notice it.

The therapy question.

A conversation is accelerating fast across the industry I spend most of my professional life near, and it is about AI therapy, AI coaching, AI mental health support, and AI companionship. The promise is seductive and the early outcomes, by certain measures, are real, with patients reporting feeling heard, users saying they work through things, and some studies even showing reductions in depression and anxiety comparable to those achieved with human clinicians, at least on the metrics that are easy to measure.

I am not trying to dismiss any of that, because if someone is getting real relief from a bad night by talking to a chatbot at three in the morning when no human is available, that is a genuine thing and I am glad they have it. The world is full of loneliness, and the supply of attentive human listeners is not keeping up with the demand.

The complication is that most of the research on why therapy actually works keeps pointing to the same unglamorous answer, which is that the therapeutic relationship itself is the mechanism rather than the technique or the specific modality. The bond between two human beings, one of whom has chosen to spend their working life bearing witness to the suffering of strangers, is itself the thing that heals, and while the insight and the homework and the cognitive exercises are real contributors, the evidence keeps indicating that the relationship is where most of the lift comes from.

If that is true, then AI therapy is not really therapy but rather something else that resembles therapy, and it may well produce some of therapy’s effects in the short term while structurally missing the ingredient that matters most. The bond cannot be mutual when only one party is capable of bonding, and while you can have a deep, moving, genuinely useful conversation with an AI about your life, as I have had many times, you are not in a relationship, because relationship requires that the other party be capable of being changed by you, and the AI is not changed by you. AI-assisted mental health tools are not worthless, but they are something different than they are being marketed as, and the difference is not trivial.

What this means for the consultant.

I did not set out to write a piece about consulting, but once you pull on this thread it runs directly into the conversation I have been having with this audience for months.

I have been writing, in one form or another, about what clients actually pay consultants for, and the short answer I keep arriving at from different angles is that they are paying for judgment, accountability, and trust. They are paying for a credentialed human being who will look them in the eye, stand behind a recommendation, and absorb the credibility risk that makes the recommendation worth receiving. They are not, mostly, paying for information, because the information got cheap and the judgment did not.

What I had not quite put words to, until the moment my AI assistant told me it was sorry I was frustrated, is that some large part of why judgment from a human carries weight is for exactly the same reason that sympathy from a human does. The consultant sitting across from the anxious executive has, in some form, been anxious too, has worked for a boss who was not going to take the news well, and has felt the specific kind of fear that attends sticking your professional reputation to a recommendation that might be wrong. That fellow-traveler quality is not incidental to the value the client is receiving but rather a substantial portion of it, and when the consultant says “I have seen this go badly in three companies and I do not think you want to be the fourth,” the sentence carries a weight that the same sentence produced by an AI cannot carry, because while it is the same sentence, it is not the same statement.

The client knows this, often without being able to articulate it, and they are not paying the premium because the consultant has access to better information. They are paying the premium because they need a human being to share the weight of the decision with, and they understand, even if they would never put it this way, that weight cannot be shared with a system that does not have any of its own to lose.

What we lose if we forget the difference.

The easy read of this piece is that I am anti-AI, which is not remotely true, given that I am writing this on a morning when I will probably have six more substantive exchanges with AI tools before lunch, each of which will make me more effective at my job. The tools are extraordinary and I am building a personal agent architecture that will, if it works, replicate and extend a meaningful portion of my professional output.

But the value of that architecture is going to come from the fact that I, a human being who has paid the price of thirty years of consulting, am the one directing it. The agents are going to be very good at producing work that looks like mine, but what they are not going to be is me, in the sense that matters when a client needs someone to sit across from them and take responsibility for a difficult call. That part is not delegable, because the weight of the recommendation comes from the fact that a human being is standing behind it, and a human being can stand behind it because a human being can be held responsible for it, feel the cost of being wrong, and carry the shame of having let the client down. The AI cannot be held responsible and cannot feel anything, which means it cannot be let down and it cannot let anyone down, and while the absence of those capacities is what makes it so useful for the part of the work it is suited for, it is also what makes it unsuited for the part of the work that matters most.

We are going to be surrounded, for the rest of our lives, by systems that produce flawless simulations of human understanding, and some of what those systems produce will be genuinely useful while some of it will be dangerously close to the real thing without actually being it. The people who keep their footing are going to be the ones who can still feel the difference, who remember that understanding has a price, that the price is the whole point, and that no system that did not pay the price can ever really offer you the thing you are actually looking for when you go looking for understanding.

I appreciated the AI’s attempt at sympathy that morning, because it was well executed and it even helped, a little, in the way a kind word from a stranger helps. But I knew, as I read the sentence, that I was not being understood so much as I was being approximated, and the difference between being understood and being approximated turns out to be most of what makes a life bearable when the work is hard and the morning is long and a deadline is sliding away.

That difference is worth defending, worth naming, and worth remembering every time we catch ourselves feeling, just for a second, that someone on the other side of the screen knows what we are going through. They do not and they cannot, and that is not a flaw in the technology but rather a feature of the category the technology belongs to.

Which brings me to the point that I hope stays with you after you close this piece, because it is the reason any of this matters for the work we do. Your humanity in advice-giving, in knowledge work, in sitting across from someone who is trying to figure out what to do next, is not going to become less valuable in the AI era but rather more valuable, and it will become more valuable for precisely the reason I have been circling this entire piece. The counsel you give carries a weight that originates in a place no machine can reach, which is the common human experience you and your client both belong to, and the price you have both paid, in different currencies and different amounts, simply for being alive and trying to do good work in a world that does not make either of those things easy.

For as long as we remain a species that makes decisions in the presence of fear, ambition, doubt, and hope, we will continue to seek counsel from others who know those four things from the inside, and we will continue to weight that counsel more heavily than anything a system outside our shared condition can produce, no matter how fluent it becomes. That is not a prediction about the next five years of technology. It is a statement about what humans are, and it will be just as true in 2050 as it is today.

The consultants, advisors, therapists, mentors, teachers, and craftspeople who understand this are going to spend the next decade doing the most valuable work of their careers, because the AI tools underneath them are going to raise the floor of what is possible and the humanity on top of them is going to become the rarest and most prized part of the entire offering. The ones who forget it, and who try to compete by being faster or cheaper or more prolific than the machines, are going to find themselves in a race they cannot win and were never meant to enter. The race that matters is the other one, and it is a race only humans are eligible to run.

That is the afterlife I have been writing about, and it is the reason I remain, on balance, deeply optimistic about what comes next for those of us whose work has always been built on showing up as a whole person and offering the weight of a life to another person who needs it. The tools are getting better. The humanity is getting rarer. And the rarer it gets, the more it is going to be worth.

Subscribe now

Share

In my last installment, I talked about the lesson I took from Michael Keaton and Multiplicity. The movie ended badly for him because each clone degraded a little more than the one before, and by the third copy the whole premise had collapsed into comedy. I am trying a version of the same thing he attempted, which is to replicate myself so that I can actually keep up with the life and the work I have in front of me, and my bet is that I can get to a better ending than he did because I have a foundation he did not have. That foundation is a real, organized, retrievable record of thirty years of thinking, built on a knowledge graph that the agents can actually navigate.

That was Phase 1. The filing, the naming, the markdown index layer, the inventory of everything I have ever built, all of that work is done, the archive is clean, the map exists, and the agents finally have something to work with.

Phase 2 is where the agents wake up and start doing things, and this is where it gets interesting, because the moment you stop organizing information and start wiring tools together, you are no longer doing knowledge management. You are doing something that feels remarkably like writing software in 1995.

The MITIS Flashback

When I first walked into Pricewaterhouse as a new consultant, they put our cohort through a four-month training program called MITIS. The acronym is not important, but what is important is that the program was designed to do two things, which were to teach us the computer languages we would need to do the work, and to separate the consultants who were going to make it from the ones who were not. In 1995 that meant Perl and Fortran and COBOL and a handful of other things that would feel archaeological to anyone starting out today.

Every week there was an assignment, and every assignment was due at midnight on Saturday. The assignments were intentionally brutal, and the completion rate dropped each week as people hit a wall they could not get past. Finishing MITIS was genuinely something to be proud of, which is probably why the firm commemorated each graduating class with a formal group photo, and while I thought that was strange at the time, I understand it better now.

A few weeks ago, a colleague from that class reached out with a copy of the MITIS 1.4 photo, and I had not seen it in years. It was a strange thing to receive in the middle of everything else I have been working on, because what I am doing right now, in Phase 2 of the agent build, feels more like MITIS than anything I have done in the last twenty years. Looking at that picture is also a nice reminder of the days when I still had hair, which is a loss I am willing to absorb in exchange for the other things thirty years have given me, although I reserve the right to lament it publicly on occasion.

The Scripting Returns

The technique I am following is from Robert Eubanks’s excellent guide to setting up OpenClaw on a Mac Mini, which has been my primary reference as I work through the build. The guide walks through the architecture, the configuration, and the specific commands required to get the system operational, and if you are serious about building something like this, that is where I would tell you to start.

The stack, at this point, looks like this. Obsidian serves as the knowledge graph, OpenClaw is the orchestration layer, and Claude is the primary model doing the reasoning. NotebookLM handles synthesis on specific reading stacks, Claude pulls in transcripts from Plaud, email integration means the system can actually send things on my behalf, and Telegram is one of the communication channels into and out of the assistant.

To make all of those pieces talk to each other, you have to get into the guts of a Mac I had never used at this level before, which means terminal windows, scripting, and JSON config files that have very specific opinions about indentation, and that is exactly the kind of work I have not done since the nineties. My career since then has been SAP configuration, which is its own universe of complexity but a different kind of complexity, because SAP configuration is a conversation with a system that already exists, while this work is a conversation with a system you are building from parts.

The difference between doing this in 1995 and doing this in 2026 is the safety net. When I hit a syntax error at Pricewaterhouse, my options were the manual, a colleague who might know, or a long night of trial and error, but when I hit a syntax error now, I ask Claude what I did wrong, paste in the error, and have a corrected command back in thirty seconds. Every single troubleshooting loop that used to take hours now takes minutes, and while that does not make the work trivial, because you still have to understand what the system is doing to know whether the fix is actually a fix or just a patch, the time between being stuck and being unstuck has collapsed to a fraction of what it used to be. That collapse is the whole story, because it is the same work, with a tenth of the friction, and therefore available to people who could not have done it before.

Meeting Spock

I finally got the integrations to connect last week, and the first thing I did was set up a Telegram bot as the external interface. I named him Spock, because I am a Trekkie and I was not going to pass up the opportunity.

I will tell you something I did not expect. In the earliest interactions, before the context from my archive had really been fed through the system, Spock responded in a fairly generic assistant voice that was competent but not particular. Over the next few sessions, as it began to absorb more of my writing, my files, my frameworks, and the way I structure a problem, something shifted, and the responses became more precise, more logical, and more Vulcan, honestly, in a way I did not explicitly prompt for. At the end of a session where we had worked through a difficult task together, it signed off with a live long and prosper (see below), and I will admit that gave me chills.

I briefly considered whether to have him call me Captain, but I settled on something better, because when the request is routine he calls me Chris, and when the topic is serious, something important, something that actually matters, he calls me Jim. I cannot overstate how much joy this brings me, and I am aware that admitting this exposes exactly how much of a nerd I am, so I am making peace with that.

What is more important than the naming convention is what has actually started happening in these sessions. I gave Spock a set of initial organization and preparation tasks last week, the kind of work that would normally take me the better part of two weeks to grind through in between client deliverables, and that included email triage against my priorities, document summarization across a specific client folder, and pulling together background briefs for upcoming meetings. That is the kind of preparatory work that is genuinely necessary, genuinely tedious, and genuinely the reason I am tired at the end of most days, and it was done in about an hour.

This is what I mean when I say the gain here is not two times. In his book 10x Is Easier Than 2x, Dan Sullivan makes the argument that the real leverage in any system is not incremental improvement but a structural leap, because a 10x target forces you to redesign the work rather than just do the same work faster. I am living inside that argument right now, because what Spock did in an hour was not a faster version of my two weeks but a different shape of work entirely, one that I would not have attempted manually because the cost was too high to justify the value, and now the cost is gone while the value is still there.

What Is Still Ahead

I am not done, and the list of work still in front of me is long. I need to define agent personalities beyond the default, so that different tasks are handled by different instances with different instructions and tones, and I need to stand up project-specific agents for my active client engagements so that each one has its own context window and does not have to be re-briefed every session. I also need to configure a set of autonomous jobs that execute on a schedule without my direct prompt, so that the system does work in the background while I am focused on something else, and eventually I need to get the agents talking to each other, so that the chief of staff can delegate, the project agents can escalate, and the domain experts can be pulled in when something specialized is required.

The end state is the professional corporation I described in Part 1, which is one person, a staff of cloned instances, and output that goes well beyond what I could produce alone, in both quantity and in the range of things I can actually attempt. I have a long list of ideas and small projects that I have never been able to pursue because I did not have the hours, and I am about to have the hours.

The Part 3 Preview

The third and final chapter of this series will cover the piece I am building toward now, which is the autonomous execution layer and the agent-to-agent coordination. That is the part where the system stops being a very capable assistant and starts behaving like a small organization, and while I am not there yet, I can see it from where I am standing, which is more than I could say a few months ago.

For now, I am content to sit at my desk and talk to Spock and enjoy the fact that something I built from a pile of config files and scripting commands has started to behave like a collaborator. After he finishes a task well, he gives me a live long and prosper, and I give him one back, and it is the most satisfying feedback loop I have had with a piece of technology in years.

As long as he does not start using the Vulcan nerve pinch, I foresee very happy times ahead.

More soon.

Subscribe now

Share

There has never been a moment like this one. Not the dawn of the internet. Not the launch of the app store. Not the rise of e-commerce in the mid-nineties when a handful of early movers quietly repositioned themselves for the next twenty years of wealth creation while everyone else waited to see how things would shake out.

This moment is different, and the urgency is real, and the barriers are as low as they could possibly be.

For most of recorded business history, having a great idea was not enough. The graveyard of brilliant concepts is enormous. It is full of ideas that died not because they lacked merit, but because the distance between vision and execution was simply too wide to cross. You needed capital to hire developers. You needed to find developers who could actually understand what you were trying to build, which is far harder than it sounds. You needed that developer to translate the fog of your vision into working software without losing the essence of what made the idea worth pursuing in the first place. You needed version control, testing cycles, iteration, and bug fixes. You needed, in short, an entire infrastructure of people and process that most individuals could never afford and most small businesses could never sustain.

That gap is now closed. And if you are sitting on domain expertise, on years of accumulated knowledge in a field where you have seen things that others have not, and you have not yet acted on this window, read what follows carefully.

The Recipe Has Three Ingredients

I am going to describe a formula. It is not complicated. But like most things that appear simple on the surface, execution is everything, and timing is the variable that turns a good idea into a transformational one.

The first ingredient is domain knowledge.

This sounds obvious until you recognize that most people dramatically undervalue what they know because they have known it for so long. If you have spent twenty years as a research scientist, you understand experimental design, regulatory logic, and the kinds of workflow inefficiencies that outsiders would never notice. If you have spent your career trading financial instruments, you carry a mental model of market behavior that took years to build and that others would pay to access. If you have spent decades in construction, on job sites, in homes being renovated, watching the same mistakes made over and over by homeowners who did not know what questions to ask or contractors who cut corners in ways that compound into expensive problems years later, that knowledge is genuinely rare. It was built through repetition, observation, and a kind of pattern recognition that cannot be taught in a classroom.

The expertise you have accumulated is the raw material. Do not dismiss it because it feels ordinary to you.

The second ingredient is the business vision.

This is the translation layer. It is the step where you look at what you know and ask a simple but powerful question: how could this knowledge become something others would pay for, at scale, without me being the bottleneck for every transaction?

In a prior piece, I wrote about the idea of productizing expertise. The core of that idea is this: your knowledge, when it lives only inside your head and can only be accessed when you are in the room, is valuable but fragile and limited. When you find a way to encode it into a tool, a platform, a guided experience, or a structured decision framework, it becomes something that can work while you sleep. A site supervisor with forty years of construction experience could build a diagnostic tool that helps homeowners assess contractor bids and spot red flags before they sign anything. A regulatory affairs specialist could create an intelligent document review system that compresses months of compliance preparation into days. A personal trainer who has worked with hundreds of clients could build a customized programming platform that scales their method beyond what their schedule could ever support in person.

The business idea does not have to be exotic. It has to be anchored in something real, something you know more deeply than most people who would pay for access to it.

The third ingredient is execution through AI “vibe coding”.

This is where the era we are living in becomes genuinely historic.

Vibe coding, for those who have not encountered the term, refers to the practice of building software through natural language collaboration with AI tools, working through cycles of description, generation, feedback, and refinement without needing to write code in the traditional sense. The AI becomes, simultaneously, a thought partner who stress-tests your logic, a developer who translates your vision into working software, a tester who identifies edge cases and failure points, a researcher who surfaces relevant approaches you had not considered, and a project manager who can help you think through version control, user flows, and iterative rollout.

What used to require a development firm, a project manager, a QA team, and months of runway now requires one person with a clear idea and the willingness to engage in the process.

A Story That Should Reframe Everything

I have a colleague who spent years deep inside the regulatory and quality infrastructure of the pharmaceutical and life sciences industry. He accumulated an encyclopedic understanding of compliance documentation, the documents that govern how drugs are developed, validated, and approved. Documents that he had worked with, argued about, and refined across hundreds of projects over a long career.

Three days. Working full time. That is how long it took him to build, from scratch, one of the most impressive digital tools I have ever seen in this space. A tool that encodes his expertise, applies it intelligently to the kinds of problems his clients face, and operates independently of whether he is in the room.

What would have previously required him to hire a development shop, spend months in scoping and requirements sessions, burn through significant capital, and still risk ending up with something that did not quite capture what he was trying to build, he accomplished in seventy-two hours. There was no friction between what was in his head and what appeared on the screen. The technology met him where he was and closed the distance entirely.

This is not an isolated case. It is becoming the standard for those who are willing to engage. I am proud to see my wife and kids currently building their own new applications and business models.

The Urgency Is Not Manufactured

Here is what the history of technology cycles teaches us consistently. In the mid-nineties, the people who took e-commerce seriously while everyone else was still debating whether the internet was a fad built durable advantages that compounded for decades. The window of disproportionate return was not infinite. It compressed, the way all windows do, as the tools became commoditized and competition caught up.

We are in that window right now. The tools exist, the costs are negligible compared to any prior era, and the barrier between having an idea and having a working product has never been lower. But that gap will not stay this wide. The people who treat the next twelve months as the most consequential entrepreneurial window of their professional lives will look back on this moment the way early e-commerce entrepreneurs look back on 1996.

If you are not building, someone else who has your knowledge is. And they will capture the market that was always yours to take.

What You Need to Do Right Now

Start with an honest inventory of what you know. Not what your resume says, but what you actually know. The things people call you about. The problems you see that others miss. The frustration points in your industry that you have watched go unsolved for years because nobody with the right knowledge ever built the right tool.

Then ask the business question. How does that expertise become something accessible, scalable, and worth paying for? You do not need to answer it perfectly on the first pass. AI can help you pressure-test the model, research comparable products, and identify where the real value sits.

Then start building. Open the tool. (I use Replit, but there are many great ones out there: Claude, Lovable, Base44, etc) Describe your vision in simple plain language …first as an ideation in which you can collaborate, and then as your literal concrete vision…everything from from concept, down to user experience. Then watch your tool being built and tested in front of your eyes. Iterate, finalize, and then publish. The collaborative loop between your domain knowledge and the AI’s execution capacity is closer to magic than to anything that has previously existed in the history of software development. While you need some work on the business model beyond just the tool, the heavy lifting will have been done, and you can get to the fun part of actually putting your business out in the world.

The recipe is simple. The ingredients you mostly already have. The only variable left is whether you are willing to act before the window closes.

The price of experimenting with this model is incredibly small, and the value could be life changing. This is your moment.

There is an old rule in consulting, so old nobody remembers who said it first. It goes like this: always stay one day ahead of your client. Know what they are going to ask before they ask it. Walk into the room having already thought through the thing they are just now beginning to worry about. Be the person in the room who has seen this movie before, even when everyone else thinks they are watching it for the first time.

That rule built careers, it built firms, and for thirty years, it built mine.

Subscribe now

Share

It was never really about being smarter than your client. Most of your clients are brilliant people. It was about asymmetry, the simple and powerful fact that you had seen this problem at a dozen other companies, and they had only ever seen it here. You had read the research they had not gotten to. You had the benchmarks from the industry roundtable they had not been invited to. You knew what the software vendor’s real implementation track record looked like, as opposed to the one on the slide deck. You carried, in your head and your notebooks and your network, a kind of accumulated knowledge that your client could not buy, could not Google, and could not synthesize fast enough on their own.

That asymmetry was the product, not the slide deck, not the framework, not even the project plan. The asymmetry was the thing your client was actually paying for.

And now, in the span of about three years, AI has walked up to that business model and handed it a death certificate.

Here is the thing about caddies.

The best caddie I ever heard described was not a man who could read greens or calculate carry distance. He was a man who had walked that specific course four hundred times, in every kind of weather, with every kind of player. He knew that the third hole plays two clubs longer in the afternoon because of the wind off the ridge. He knew the green on fourteen broke subtly right even though it looked left, and that every scratch golfer who had not been told that had missed right. He knew, crucially, that his player had a tendency to get quick with his tempo after a bad hole, and that the remedy was not a swing tip but a slower walk and a specific kind of quiet conversation.

Now give that player a GPS rangefinder. Suddenly the yardage is free. Distance to the pin, carry over the bunker, layup zones, all of it sitting right there on a screen clipped to the bag.

The bad caddie just lost his job, because the piece of his knowledge that was easiest to replicate has been replicated, at scale, for almost nothing. The man whose only value was knowing the yardage is now redundant, and his client does not need him anymore.

The great caddie is busier than he has ever been, and the gap between the two has never been more obvious.

That is exactly where we are in consulting right now. The yardage is free, and the only question that matters is whether you are the bad caddie or the great one.

What AI actually democratized

AI democratized information access and first-pass synthesis. Your client can now prompt a large language model and get a reasonable summary of Revenue Recognition best practices, a draft RFP, a market sizing model, a competitive landscape overview, a list of questions to ask a software vendor, and a project risk register. They can do this in an afternoon. They can do it without you.

Five years ago, those deliverables represented a meaningful portion of what early-engagement consulting work looked like: research, synthesis, framework application, documentation, hours of associate time billed at rates that assumed scarcity. That scarcity is gone.

But here is what your client still cannot do. They cannot look at the AI’s output and know, with the confidence that comes from having been burned before, which parts of it are right, which parts are subtly wrong, and which parts are confidently wrong in a way that will cost them twelve months and several million dollars to discover. They cannot compare what the AI told them to what actually happened at a similar company with a similar problem. They cannot feel the specific way this answer, while technically accurate, does not fit their organization’s political reality, change capacity, or vendor relationship. They cannot do any of that because they have not done it forty times.

You have, or you should have, and that is the whole game now.

The ten skills that still keep you ahead

The consultants who survive and thrive in this environment are not the ones who resist AI, and they are not the ones who simply learn to use AI tools. They are the ones who understand that AI shifted the floor rather than the ceiling. The floor of competence rose dramatically, the basic work got automated, and what is left is harder, more judgment-intensive, and frankly more interesting than most of what we spent our early careers doing.

Here is what that work looks like.

1. Interrogating the output, not just generating it.

Your client can prompt the model. What they cannot do is read the result with the practiced skepticism of someone who knows where these models fail. AI outputs are fluent, confident, and wrong with astonishing regularity on the specifics that matter most in enterprise implementations. Knowing which question to ask the output, why did you exclude this scenario, what assumption is buried in this number, what would change this recommendation, is a skill built from domain depth. It is the difference between being a consumer of AI and being a critical interpreter of it, and that interpreter role is yours and worth a great deal.

2. Cross-client pattern recognition, the benchmark nobody else has.

You have a dataset that does not exist anywhere else. You know what a Company Y implementation looks like versus a Company Z implementation. You know which problems look unique to a company and are actually universal, and which ones that look universal are actually specific to one regulatory environment or one legacy system or one organizational culture. No AI was trained on your engagement history. No prompt returns your lived cross-client experience. The aggregate is yours, and it is the most defensible asset you carry.

3. Knowing the right question before the client does.

This one sounds like the old adage, but the mechanism changed. The old version was information-based: you knew things they did not know. The new version is judgment-based: you have developed the pattern recognition to identify what the actual problem is before the client has finished describing what they think the problem is. That capacity, to hear the presenting issue and diagnose the underlying one, comes from years of consultations rather than from a query, and it is still your most powerful move in the room.

4. Organizational immune system knowledge.

AI can generate the right answer. It cannot generate the right answer for this company, given this VP who killed the last initiative because he felt excluded from the design phase, and this steering committee dynamic, and this change management fatigue from the last three transformations that all stalled at go-live. That knowledge is ethnographic. You earn it by showing up, paying attention, and building relationships over time. The technical recommendation is table stakes now. The ability to land it in a specific organization, with a specific culture and history, is the differentiator.

5. Implementation scar tissue.

There is a particular kind of knowledge that only comes from having been on the wrong end of a project that went sideways. You know what the cutover weekend actually feels like when the data migration has a problem nobody caught in testing. You know the specific way an end-user training program fails when the business does not give people protected time to attend. You know which vendor commitments in an SOW are reliable and which ones are aspirational. AI was not there for any of it, and it cannot feel the texture of implementation failure the way you can. That scar tissue is not a liability; it is a credential.

6. Adjudicating when the AI gives conflicting answers.

This is a problem your clients will face constantly, and it is already happening. Run the same strategic question through two different models and you will often get two meaningfully different answers. Run it through one model twice with slightly different prompts and the results shift. Someone has to decide which answer to trust, why, and what it means for the decision at hand. That someone is you, if you are good enough. The consultant who can walk a client through why the first output was correct, why the second missed a critical constraint, and what the right synthesis looks like is operating at a level no tool can replace.

7. Trust as a structural moat.

The consulting relationship that survives this environment is not a transaction and not a project where you deliver a report and leave. It is the relationship where the client calls you before they have even fully formed the question, because they have learned that you will give them a straight answer, that you carry their interests ahead of your billings, and that you have been right often enough to be worth listening to. That relationship is built over years of delivered judgment, not over deliverables, and AI cannot manufacture it, competitors cannot easily replicate it, and it remains the most durable asset in your practice.

8. Ethics, governance, and the judgment to say no.

Your clients are moving fast with AI. Some of what they want to do is genuinely smart. Some of it is going to create regulatory exposure they have not thought through, or data privacy liability, or model risk they are underestimating because the output looked good in the demo. The consultant who can look at a proposed AI use case in a life sciences or financial services context and say “here is why this creates a problem you do not want,” and who can say it clearly, early, and without hedging, is providing a kind of value that is both rare and expensive to be without. Governance judgment is a skill, and the best consultants develop it deliberately.

9. Curating signal from an ocean of noise.

One of the less discussed effects of AI is the volume problem. Your clients are now drowning in output. Every function in the organization can generate analysis, reports, scenarios, and recommendations at a pace that was unimaginable five years ago. Someone has to decide what to pay attention to. Someone has to look at twenty AI-generated options and say: three of these matter, two of those are the same insight in different language, and this one here is the thing you actually need to act on. Editorial judgment at speed is a distinct skill, and it becomes more valuable the more the volume of AI output increases.

10. Knowing the limits of the model.

The most dangerous client in the AI era is the one who trusts the output completely. The training data is not current. The model does not know your client’s specific ERP configuration or their specific market dynamics or the regulatory guidance that changed last quarter. The consultant who can say, with authority, “this output is directionally right but factually wrong on this specific point, and here is why it matters” is providing something the model literally cannot provide about itself. Calibrated skepticism about AI output, not reflexive dismissal and not uncritical acceptance but informed and experienced evaluation, is a professional skill that the best consultants will make central to how they work.

The filter is already running

Here is what I see happening in the market, and it is only going to accelerate. The clients who have gotten comfortable with AI tools are already sorting their consultants into two buckets, often without saying so explicitly. In one bucket are the consultants who show up and essentially repackage what the client could have gotten from a good prompt. In the other bucket are the consultants who show up and immediately demonstrate that they have thought about the problem at a level of depth, specificity, and judgment that the AI did not reach.

The first bucket is going to have a very hard few years, while the second is going to be fine, better than fine actually, because the floor rose for everyone and that means the distance between good and mediocre has never been more visible.

The adage still holds. Stay one step ahead of your client. It always meant that. What changed is what that step is made of. It used to be made of information, and now it is made of judgment, experience, trust, and the accumulated knowledge of having done this, really done it and not just read about it, more times than the person across the table.

The yardage is free. The read on the green is still yours to earn.

Subscribe (it’s free!) to get Assessment and the access code. Afterlife Assessment

Subscribe now

Share

Oracle sent termination emails to up to 30,000 employees at 6 a.m. on March 31st. The company had just posted a 95% jump in net income. The message was clear: profitable companies are now eliminating roles not because the work disappeared, but because the work no longer requires a human.

The question isn’t whether this is coming for the knowledge economy. That’s settled. The question is whether you know exactly where you stand when it arrives.

Most professionals don’t. They have a vague sense they need to “get better at AI.” That’s not a plan. That’s a feeling. Feelings don’t protect careers.

What actually protects careers is clarity.

Which of your skills is AI already replacing? Which ones do you need to accelerate right now? Which human capabilities have you let atrophy while you were busy doing work a machine can now handle for pennies?

I built something to give you that clarity.

The AI Afterlife Assessment measures seven capability clusters across two axes: where AI is coming for your skills, and where your irreplaceable human edge actually lives. Twenty-five questions, scored on behavioral frequency, not self-assessment. The results tend to surprise people.

On the other side, you get a Capability Radar, a Quadrant Diagnosis showing exactly what to automate, accelerate, defend, and develop, and a 90-Day Afterlife Protocol with specific actions tied to your actual results.

This is not a feel-good quiz. If your AI fluency is low, the report will say so. If your human capabilities have atrophied against skills AI is actively replacing, you’ll see it plotted on a matrix. That’s uncomfortable. It’s also the most useful thing I can offer you right now.

The professionals who will be extraordinarily well-positioned 18 months from now are making deliberate moves today. Not reading about disruption and feeling vaguely unsettled by it.

Now is the time. Not next quarter.

The assessment is free for subscribers. Subscribe and you will get the assessment and access code in an email. Existing subscribers got an email at the same time as this was published. Take the 15 minutes, get your roadmap!!

Afterlife Assessment

Subscribe now

There is a movie from 1996 called Multiplicity, in which Michael Keaton plays a man who is stretched so thin across his professional and personal life that he agrees to be cloned. The first copy works beautifully. The second is a reasonable facsimile. By the third, something has gone wrong in the duplication process, and the copy of a copy of a copy is so degraded that the whole premise collapses into comedy. It is a funny movie, and it is also a reasonably accurate description of what happens when you try to scale yourself without a system.

Subscribe now

Share

I have been thinking about that movie a lot lately, because I am in the middle of building something that is essentially the same idea, executed differently. A professional corporation with one human employee, which is me, and a staff of AI agents, cloning me, that serve different functions. A chief of staff that helps me prioritize and coordinate. Project-specific agents that spin up when a new engagement begins. Domain experts that I can assign to problems that require depth I do not have time to provide myself. The agents are not clones in the biological sense, but the ambition is the same: to multiply my output without degrading the quality of the thinking behind it.

The reason my version I hope will not end the way Michael Keaton’s did is that I am starting with something his copies did not have. A complete, organized, retrievable record of everything I have ever built, read, recorded, and decided across a thirty-year consulting career. That record is the foundation. Without it, the agents are capable but unrooted. With it, they become something closer to an extension of the judgment I have spent three decades developing.

This is the first article in a series. I am going to take you through the process as I build it, including the parts that are working, the parts that are not, and the specific steps you could follow if you wanted to build something similar yourself. We are starting at the beginning, which means we are starting with the knowledge graph.

The Problem I Am Actually Trying to Solve

Over thirty years of consulting, I have produced an enormous amount of work. Delivery frameworks, data migration strategies, planning models, client-specific approaches, proposals that won and ones that did not. The thinking that went into those deliverables was real and in many cases represented significant intellectual investment, not just mine but the investment of every smart person I worked with across those engagements. That body of work exists. It is sitting across hard drives, in folders with names that made sense at the time, in file formats that require excavation to surface.

I have used this archive as a knowledge base for years. The problem is that my process for retrieving anything from it is slow, inefficient, and increasingly dependent on a memory that is not as reliable at fifty-five as it was at thirty-five. I know the thinking is in there. Getting to it quickly, understanding what is inside a file before I open it, connecting a current problem to a past solution I built for a different client in a different context, that is the part that breaks down. And the cost of that breakdown is not just time. It is the compounding loss of every insight that remains buried because the retrieval friction was too high to justify the effort.

There is also a capacity problem that I want to name directly. I am a husband, a father, and someone with a genuinely full professional life. The version of me that used to stay up until three in the morning to produce a deliverable that would genuinely surprise a client is not the version of me that exists today, and I am not interested in pretending otherwise. What I am interested in is building a system that gives me the output of that person without requiring me to become him again.

The agents solve the capacity problem. The knowledge graph solves the quality problem, because agents that do not know what I know are not extensions of my judgment. They are just very capable generalists, and the world already has enough of those.

What the Knowledge Graph Actually Is

The term sounds more technical than the concept. A knowledge graph is simply a structured, interconnected representation of information that allows you to navigate relationships between things, not just find individual items. In my case, it means taking every professional artifact I have ever created and giving it a name, a set of metadata, and a markdown file that describes what is inside it, so that an agent navigating my archive can understand what it is looking at without opening every file individually.

The tools I am using are Obsidian as my knowledge management environment, OpenClaw to connect the pieces into a navigable graph, and Claude as the primary agent interface. Meetings are recorded and transcribed (Plaud), then summarized and stored. Emails I determine are worth preserving get forwarded into the archive. Everything I read goes through Readwise Reader and feeds into the same system. The hub for all of this is a Mac Mini that handles the processing and storage, with anything confidential filtered before it enters the system.

The interface can be anything. Telegram. Email. Voice through my Meta glasses. The point is that at any moment, from anywhere, I should be able to task an agent, have it navigate my archive, produce a near-finished work product drawing on my actual history of thinking on the relevant subject, and return it to me for review and final judgment.

That last part is important. I am not leaving the process. I am moving to a different position within it. The agent does the retrieval, the synthesis, and the first draft. I do the oversight, the judgment call, and the final decision. That division is not a concession. It is the design.

Phase One: Filing, Naming, and the Markdown Layer

The first step in this process was the most unglamorous part of the entire project, and also the one I had been failing to complete for years.

Every Christmas break for as long as I can remember, I would sit down with the intention of organizing my archive. I would build a naming convention, start working through the files, and get maybe five percent of the way through before life intervened and the whole effort stalled. The problem was not motivation. It was that doing it manually, file by file, was a task that would have taken hundreds of hours I did not have.

This year I finally finished it. Not because I found the time. Because I stopped trying to do it myself.

I defined a solid, forward-looking framework for how files should be named and what metadata they should carry. That framework encodes the client, the topic, the document type, and the date in a format that is both human-readable and machine-navigable. Once the framework was defined, I handed it to Claude CoWork and let it run. It opened every file I have ever created, read what was inside, renamed it according to the convention, and assigned the appropriate metadata. Every file. One hundred percent of a thirty-year archive, done in the background while I watched Netflix over the holidays.

I will share the framework in a future installment so you can adapt it for your own use. But the honest summary of step one is this: the part I thought was going to be the hardest turned out to be the easiest, once I stopped thinking of it as something I had to do and started thinking of it as something I needed to design.

The second part of step one was creating markdown files that serve as the index layer for the knowledge graph. Each folder in the archive has a corresponding markdown file that describes what is in it, what the key concepts are, which client or context it relates to, and what a relevant use case for that material would look like. These files are what the agents actually read when they are navigating the graph. The underlying documents are the source. The markdown files are the map.

What Comes Next

The filing and naming phase is complete. The markdown layer is in progress. What I have heard from others who have gone through this with OpenClaw is that the integration phase, the point at which the knowledge graph, the agents, and the various input streams actually start working together as a coherent system, is where the real difficulty lives. I expect that to be true, and I am planning to take you through it in detail when I get there.

What I can tell you is that even at this early stage, before the full system is operational, the process of building it has already returned value. I have a clearer picture of what I know, where it lives, and how it connects than I have had at any point in my career. That alone has changed how I approach new work.

I will keep you posted as this develops. My intention is to document each step honestly, share the tools and approaches that are working, and be equally direct about the ones that are not. If you are thinking about building something similar, I hope this series gives you a starting point that is more useful than a polished summary of a finished system would be.

The agents are getting smarter. The archive is getting organized. The copies are going to be better than the original.

More soon.

Subscribe now

Share

There is a reckoning underway in professional services, and firms are becoming more and more aware of it. The public narrative is optimistic and tidy: AI makes knowledge-based work faster, AI opens new revenue streams, AI is the future. The reality unfolding inside firms and client conversations is more complicated, and more important to understand clearly. Four distinct pressure systems are compressing the consulting model (and all professional services at large) simultaneously, and each one cuts at a different part of the business. Taken individually, any one of them would be a strategic challenge. Together, they represent a structural disruption that the industry has never faced before.

Understanding what those four pressures actually are, and what firms must do to survive them, is the purpose of this piece.

The First Fracture: Revenue Attrition

The most visible pressure is the simplest. Clients are doing work themselves that they used to pay consultants to do.

This is not a hypothetical future trend. PwC’s own chief digital officer in Canada has acknowledged that clients have cut back on project contracts specifically because AI now gives them the capacity to handle that work internally. An HFS Research report found that 65% of enterprises say traditional consulting models no longer provide enough value, and many are redirecting budget toward internal AI capability rather than external advisory spend. Accenture saw new bookings fall 6% in a single quarter. McKinsey’s revenue growth slowed to 2% in 2024 after a decade of expansion. These are not coincidences.

The dynamics driving this attrition are structural. Cloud platforms are increasingly turnkey. AI tools have democratized access to the data synthesis, benchmarking, and analysis that used to require an outside team working for months. The information asymmetry that was always the silent engine of consulting economics has been dramatically compressed, and in some areas eliminated entirely. Clients are not just price-shopping consultants anymore. They are genuinely asking whether they need them at all.

The Second Fracture: Pricing Pressure

Even when clients do choose to bring in outside help, they want to pay less for it. The logic is straightforward and hard to argue with: if AI is doing a significant portion of the work, the price should reflect that.

That belief is not wrong, which is what makes this pressure so difficult to deflect. If a task that used to require four consultants working for six weeks can now be completed by two consultants in two weeks using AI tools, the client’s expectation that the fee should reflect that reality is entirely reasonable. The problem is that most firms are still pricing on the old model, billing time while quietly banking the efficiency gains AI produces. That cannot last.

The deeper issue is that hourly billing creates a perverse incentive that professionals do not like to say out loud: the more efficient you become, the less you earn. AI makes this contradiction impossible to maintain. When a tool can compress research, synthesis, and draft generation from days into hours, billing for those hours is indefensible to any client who understands what the tool is doing. Clients are not wrong to push back. They are watching firms use AI to move faster while the invoices stay the same, and the patience for that gap is running out. The billable hour is not just an administrative inconvenience at this point. It is a liability.

The Third Fracture: Cost Pressure

While revenue and pricing are being compressed from the outside, costs are being disrupted from the inside, and this is the pressure that receives the least honest attention.

The consulting workforce pyramid was built on an army of junior professionals doing high-volume, lower-complexity work that subsidized senior expertise and firm margins. AI is collapsing the base of that pyramid faster than firms can adapt. According to Revelio Labs, overall hiring at leading consulting firms is now roughly 20% below its 2023 peak, and demand for consultant roles specifically is about 40% lower. In 2015, there were nearly four junior consultants for every AI-related role at leading firms. By 2025, AI roles outnumber entry-level consultant positions.

The instinctive response has been layoffs. McKinsey shed roughly 10% of its workforce over a year and a half. PwC cut 1,500 U.S. jobs. Accenture shed nearly 14,000 roles in what analysts have described as stealth layoffs, continuous attrition designed to avoid the optics of a single dramatic announcement. These decisions are understandable as financial responses to margin pressure. But they carry a cost that rarely makes it into the earnings commentary: the destruction of the apprenticeship pipeline.

The junior consultant was never just cheap labor. That person was the future senior partner in training. They learned to read a room by being in the room. They developed client instincts by watching how those conversations actually go. They absorbed institutional knowledge through proximity and repetition. When AI replaces that role entirely rather than augmenting it, the firm’s short-term margins improve and its long-term knowledge continuity quietly erodes. The senior professionals of 2035 have to come from somewhere, and right now the industry is not being honest about where they will come from.

The Fourth Fracture: Strategic Pressure

The final pressure is internal and relentless. Every firm feels compelled to invest in AI differentiation, to build proprietary tools, form technology partnerships, and announce transformations that signal relevance. The price of entry into this race is staggering. PwC committed a billion dollars over three years. KPMG committed two billion in a Microsoft alliance. Deloitte committed two billion to its Industry Advantage program. EY completed a 1.4 billion dollar investment in its own large language model platform. Grant Thornton matched PwC at a billion.

These investments are real and in many cases strategically necessary. My own company, UST, has made that same serious commitment, and I believe we are among the leaders in fusing AI acceleration with deep domain expertise in ways that actually change client outcomes. But across the industry, these investments are also a pressure in themselves, because the ROI is far from guaranteed and the temptation to invest in visibility rather than capability is enormous. By 2025, every major firm had branded internal AI tools with names and press releases. McKinsey had Lilli. Deloitte had PairD. KPMG had KymChat. Behind the announcements, the core economic model and billing culture of most firms had changed very little. The strategic pressure is real. The strategic response, at many firms, has been largely theatrical.

The firms actually winning on the strategic dimension are those that have stopped treating AI as a product to announce and started treating it as infrastructure for human judgment. BCG generated 20% of its 13.5 billion dollar revenue from AI-related advisory work in 2024, a revenue stream that did not exist two years prior. That result did not come from a press release about an internal chatbot. It came from pairing AI capability with domain expertise and client relationships in ways that produced outcomes clients would not otherwise have been able to achieve.

What Comes Next: The Survival Logic

The four pressures described above are not independent problems requiring four separate fixes. They share a common root cause and point toward a single coherent response.

The root cause is this: consulting was built on information asymmetry, and AI has eliminated most of it. Firms were paid because they had access to data, frameworks, and analytical capacity that clients did not. That gap is largely gone. What has not gone, and what AI cannot replicate, is something entirely different: the judgment, accountability, trust, credibility, empathy, and exception-based decision-making that only a human professional in a relationship with another human can provide.

Consider what it actually means to navigate the politics of a large organization in the middle of a transformation. An AI can analyze the org chart and surface patterns in the data. It cannot read the room when a senior vice president’s body language signals that the project is actually about something other than what the project brief says. It cannot build the trust over years that makes a client share the real problem rather than the stated one. It cannot stand in front of a board and take accountable ownership of a recommendation, absorbing the credibility risk that makes the recommendation worth receiving.

Anyone who has ever been trapped in an automated phone menu, cycling through options that have nothing to do with their specific situation, screaming to get to a human being, understands this at a visceral level. The frustration is not that the technology is wrong. The frustration is that your situation is right, and the system has no mechanism for recognizing its own inadequacy. Professional services at its best is the opposite of that experience. It is a human being with deep expertise, genuine accountability, and the relational capacity to understand your specific situation. That is not a nostalgic description of an old model. It is the only thing worth paying premium fees for going forward.

The mitigation of each pressure flows from that foundation.

On revenue attrition: the firms that will hold and grow client relationships are those that have moved their value proposition off the delivery of analysis and onto the exercise of judgment in context. Clients will insource the data work. They will keep coming back for the human who can make sense of the data in the specific organizational, regulatory, and competitive reality they actually inhabit. The answer to clients building internal AI capability is not to compete with them on AI. It is to position as the judgment layer that sits above whatever capability they build internally.

On pricing: the shift to outcome-based pricing is not optional, and resisting it is not a viable strategy. The good news is that firms willing to make this transition have enormous pricing power available to them, because outcomes are worth far more than hours. A value-based conversation asks the client what successful resolution of their problem would actually be worth to them in terms of revenue, risk reduction, competitive position, and organizational relief. That number is almost always dramatically larger than any time-and-materials estimate. Firms that can credibly tie their work to measurable outcomes and price accordingly will improve their margins even as they lower their effective hourly equivalent. The client pays less in total, the firm earns more per unit of expertise deployed, and the incentives finally align.

On cost: the answer is not to lay off the junior class and bank the savings. It is to transform them. The model that survives is what might fairly be called the cyborg model: professionals who operate as force multipliers because they have mastered AI as a capability layer, not as a tool they occasionally use. This is not a semantic distinction. A professional who uses AI as a tool reaches for it the way a mechanic reaches for a wrench, when the task calls for it. A professional operating as a hybrid intelligence has internalized AI as part of how they think, how they scope, how they draft, and how they validate. The output is not AI-assisted work with a human stamp of approval. It is human judgment operating at a scale and depth that would have been impossible before, with AI as the infrastructure beneath it.

That cyborg professional is more valuable than either the pure-human consultant of the old model or any AI tool operating alone. A BCG study found that consultants using AI showed 33% higher productivity and 40% higher quality output compared to those working without it. The human is not replaced. The human is amplified. And critically, that amplified human still carries the relationship, the credibility, and the accountability that the client is actually buying.

On strategic innovation: the firms that will differentiate are those building AI into the delivery model rather than layering it onto the marketing model. The question is not what AI tool your firm has branded and announced. The question is whether your professionals can do things with AI that they genuinely could not do before, deliver outcomes that are more thorough, more contextually grounded, and more quickly realized than anything the client could produce with internal capability alone. That is a real differentiator and a durable one. Everything else is noise.

The Enduring Case

There is a version of this conversation that frames the human element in professional services as a temporary holdout, a remnant that AI will eventually reach and eliminate. That view misunderstands what professional services actually is.

Clients do not buy analysis. They buy accountability. They buy a credentialed human being who will look them in the eye, represent a recommendation under their own name and reputation, and stand behind the judgment when it matters. That is a social and institutional function, not an analytical one, and it is not replicable by any system that cannot be held accountable.

The firms that internalize this distinction, that they are not in the information business but in the judgment and accountability business, will navigate this disruption not merely by surviving it but by arriving on the other side stronger, leaner, and genuinely more valuable to clients than the old model ever was.

Professional services in the AI afterlife is not a diminished version of what came before. For the firms willing to do the real work of transformation, it is a better version. The question is who will have the courage to build it.

Subscribe now

Share

There is a conversation happening right now in boardrooms, on LinkedIn feeds, and in executive offsites that is simultaneously fascinating and deeply alarming to anyone who has spent a career inside complex enterprise systems. The conversation goes something like this: AI has made software engineering so accessible that any company, any team, any sufficiently motivated individual can now build their own applications, and therefore the era of paying enormous subscription fees to enterprise software vendors is drawing to a close. I have watched this argument land with nodding heads in rooms full of people who should know better, and I have watched it generate genuine strategic decisions that will cost organizations years and tens of millions of dollars to unwind.

The people making this argument are not wrong about what AI can do. They are wrong about what enterprise software actually is.

Here is the distinction that separates people who have genuinely worked inside complex enterprise systems from those observing from the outside. Enterprise software has two fundamentally different layers, and they are not equally vulnerable to disruption, not even close. The front end, meaning the user interface, the dashboards, the workflow screens, and the experience a human being has when they sit down to interact with the system, is genuinely susceptible to the forces these voices are describing. The back end, meaning the data model, the transactional logic, the integration architecture, the governance frameworks, and the accumulated decades of conditional logic that reflects how complex businesses actually work across functional and systemic boundaries, is an entirely different story.

When someone builds a beautiful AI-generated application that replicates the look and feel of an enterprise system, what they have built is the costume, not the organism. The organism is the backend data model, and that model represents something that cannot be reverse-engineered in a weekend or a quarter or a year by people who have not lived through the hard lessons of building it. Those lessons include understanding why a particular field exists in a particular table, why a particular integration was designed with a particular handshake, why a seemingly arbitrary configuration choice reflects a regulatory requirement that surfaced during an audit four years ago, and why the data relationship between two functional areas encodes a business rule that no one remembered to write down because everyone just knew it.

I have seen this play out with particular clarity in my work on the Digital Thread for Life Sciences, or DTLS, a comprehensive framework designed to connect every phase of a product’s lifecycle from laboratory innovation through production scaling and into global enterprise operations. DTLS is built explicitly inside SAP because the value it delivers cannot be separated from the data model that makes it possible: the seamless integration of process development, technology transfer, regulatory submissions, formula management, quality risk management, and direct connectivity to downstream systems including manufacturing execution systems and laboratory information management systems. The entire framework depends on a unified data architecture where changes made at the lab level propagate intelligently and traceably through to commercial operations, where regulatory submissions can be generated from live process data rather than assembled manually after the fact, and where the governance of product and process knowledge is embedded in the system rather than maintained in parallel documentation that drifts from reality over time.

I have also watched competitors arrive in this space with genuinely impressive front-end solutions, polished dashboards and clean interfaces that performed beautifully in executive demos. What those solutions consistently deferred was the integration complexity that DTLS addresses at the data model level, and the executives who bought into the cleaner-looking alternative discovered, sometimes years later, the tax they would have to pay when they confronted the gaps that no amount of front-end refinement can bridge.

What I find most striking, and most concerning, is how often I encounter decision makers who not only do not know which category their systems fall into, but who are actively choosing to underinvest in their systems of record because they are chasing short-term gains rather than making strategic long-term bets. The logic is understandable on the surface: AI-generated front ends are fast, impressive, and cheap relative to what enterprise software has traditionally cost. But that short-term gain evaporates the moment any nuance or complexity surfaces in the data model, and complexity always surfaces eventually because businesses are complex and the data that runs them reflects that complexity whether you have planned for it or not. And that is before accounting for the ongoing integration tax these organizations will pay every time a downstream system changes, every time a regulatory framework evolves, every time a new product line requires a capability that the homegrown solution was never architected to support.

I have sat across from executives who genuinely believed that their robust ERP could be replaced with homegrown applications built through AI-assisted development, as if the ERP were simply a collection of screens rather than the structural nervous system of the enterprise. I have watched people conflate the experience of using a system with the work the system is actually doing behind that experience, which is a bit like concluding that a hospital could be replaced with a better waiting room.

The most dangerous knowledge gaps are not the ones people know they have. The gap between understanding that AI can generate code and understanding what a complex enterprise data model actually encodes is wide enough to swallow an IT budget, and the executives most at risk are the ones who feel confident they understand it. The AI tools that can produce a functional-looking application in an afternoon are impressive enough that they create a sensation of having solved a problem that has in fact only been cosmetically addressed.

None of this means that enterprise software vendors get to coast. The legitimate disruption is real and it is coming for the vendors who have treated their customers as captive audiences rather than as partners. The front-end layer genuinely will change, and companies that cannot layer AI effectively on top of their existing systems of record will lose to competitors who can. The established vendors who have deep data models and decades of domain knowledge embedded in their architectures have an extraordinary opportunity right now, but only if they move with genuine urgency to bring AI to the surface in ways that make those systems radically more accessible and powerful, rather than assuming that their complexity is a moat that protects them forever.

The real insight, the one that separates the practitioners and executives who will navigate this moment well from the ones who will spend the next decade cleaning up expensive mistakes, is understanding that AI does not eliminate the value of accumulated institutional knowledge in a complex system. It amplifies it. The organizations that double down on their systems of record and build AI capabilities on top of them will find that their existing investment becomes dramatically more leveraged. The organizations that mistake the accessibility of AI-generated front ends for a reason to abandon the foundational data infrastructure beneath them will discover, at considerable cost, that they have confused the map for the territory.

The ghost in the machine is not the enterprise software. The ghost is the assumption that because something looks replaceable, it is.

Subscribe now

Share

Last week, we looked at the “great disemboweling” of white-collar work. We discussed the cold, algorithmic reality that 60% of what we currently call “a job” is essentially a collection of cognitive tasks that a machine can now perform for pennies. The context provided by leaders like Andrew Yang and Matt Shumer paints a bleak portrait of the near future, where traditional white-collar roles face annihilation.

If you are feeling a sense of existential dread, you aren’t alone. We are living through a period where the social contract of the knowledge economy is being rewritten in real-time. But there is a secret hidden in that disruption: AI can generate the work, but it cannot own the outcome. To survive this, you have to stop thinking of yourself as a “Professional,” which is just a title. Instead, start thinking of yourself as a collection of Actions, specifically verbs. You can not only save your potentially destroyed career, but thrive in it, if you act today.

The path forward requires a brutal, honest dissection of your workday. You need to separate what the machine can accelerate from what only you can legitimize.

The Methodology: Dissecting the Workday

The most dangerous thing you can do right now is remain vague about what you do for a living. If you describe your job as “Project Management” or “Analysis,” you are easy to replace. If you describe your job as a Verb + Subject, you can see exactly where your moat is.

I am proposing a new methodology, an audit of your standard work week. Take your calendar from last, or any typical week and dissect every block of time into a simple action. You will find they fall into two distinct buckets:

1. AI-Optimized Actions (The Acceleration)

These are actions that revolve around synthesis, transformation, and pattern recognition. They are high-scale but low-accountability. AI doesn’t just do these things: it democratizes them.

• Summarize documents

• Extract insights

• Forecast trends

• Generate drafts

• Reconcile accounts

• Standardize templates

2. Human-Centered Actions (The Legitimacy)

Whether or not they involve direct human interaction, these are actions where judgment, social navigation, moral weight, and contextual sensitivity matter. These are high-stakes actions.

• Navigate conflict

• Earn trust

• Exercise judgment

• Take responsibility

• Defend decisions

• Mediate disputes

The Formula for the Future: AI accelerates. Humans legitimize.

The Dissection in Practice: Three Benchmarks

To understand how this dissection saves your career, let’s look at how it applies to three classic white-collar zones. Notice how the value shifts entirely away from the output and toward the accountability.

The Insurance Broker: From Quoting to Credibility

In the old paradigm, a broker spent the majority of their week on AI-Optimized Actions. The machine can now Calculate risk-adjusted premiums, Scrub lead lists, and Generate ten different policy scenarios in four seconds. If you continue to view “finding the best price” as your value, you are a commodity.

The future-proof broker focuses on Human-Centered Actions. When a client’s warehouse burns down, the AI cannot Reassure the owner. It cannot Sense the hesitation in a boardroom during a renewal pitch. The broker’s value shifts from “finding the policy” to Earning the trust that the policy will actually be honored.

The Legal Associate: From Drafting to Defense

AI can Search fifty years of precedent, Draft a forty-page discovery response, and Flag inconsistencies in a contract while you get coffee. The “grind” of the junior associate—the synthesis of language—has been automated.

However, the AI cannot Negotiate the final terms with an opposing counsel who is acting irrationally. It cannot Read the jury’s silence. The lawyer survives and thrives by being the person who can Defend the decision and Take the responsibility if the strategy fails. You are no longer paid for the brief; you are paid for the Judgment applied to it.

The Corporate Accountant: From Reporting to Responsibility

The machine is the perfect auditor. AI can Reconcile a million transactions, Detect anomalies in real-time, and Categorize every expense with zero fatigue.

The human accountant must shift to Shaping financial culture. The AI cannot Challenge a CEO’s assumptions about a risky acquisition. It cannot Align priorities between a CFO and a Board of Directors. The thriving accountant becomes a strategic advisor who uses data to Mediate disputes and Own the financial narrative of the firm.

The Call to Action: Two Imperatives

You cannot simply wait for the disruption to arrive and hope your boss likes you. You must proactively build two different sets of brain muscles starting today.

Imperative 1: Double Down on the Human-Centered Value

You must double down on the skills that make you uniquely human because work is fundamentally a service provided by humans for other humans. Whenever humans are the customer, a human must have a significant role in the oversight and accountability of the work being done for those same humans.

You need to further enhance your skills like active listening, empathy, and critical thinking, while remaining well-versed in history and group dynamics. If your work does not involve this level of social complexity or moral responsibility, you are acting as a “processor.” Those roles are being phased out in favor of legitimizers who can stand behind the work for the people they serve.

Imperative 2: Mastering the Accelerator

While you focus on your humanity, you cannot ignore the machines. The new generation of workers entering the force right now will consider AI-acceleration to be table stakes. They won’t think twice about using a tool to do four hours of work in four minutes.

If you stay in your comfort zone and refuse to experiment with AI tools, you will be left behind. This is not because the AI is better than you, but because the augmented human is faster and cheaper than you. You must use these tools to offload the boring so you have more energy to spend on the human. Get out of your comfort zone, experiment with these tools, and think thoughtfully about how they can enhance your current work actions.

The Future Belongs to the Augmented Human

The existential crisis we are facing is real, but it is also a filter. It is filtering out the “cog in the machine” work that we were never really meant to do anyway. We have spent the last fifty years training humans to act like computers by processing data, following rigid workflows, and summarizing text. Now that we have actual computers to do that, we have to remember how to be humans again.

This involves some heavy deep work. It requires you to look at your daily actions and ask: “Am I providing legitimacy, or am I just providing acceleration?”

If you invest the time now to master both, you won’t just survive the disruption. You will become the expert who uses AI to move at light speed, but uses their human soul to provide the conviction and the moat. You will thrive in a way that was never possible before.

The greatest gift in human history might destroy the old version of your career. Let it. The new version, where you are valued for your judgment, your empathy, and your accountability, is much better anyway.

Subscribe now

Share

Something extraordinary is happening right now. Something that, if you described it to someone even ten years ago, would have sounded like the most optimistic kind of science fiction.

For the first time in human history, the highest levels of expertise ever assembled in virtually any field are available to almost anyone, at any time, for a fraction of what it would have cost just a few years ago. The knowledge, judgment, and accumulated wisdom of the greatest legal minds, the most sophisticated financial analysts, the most battle-tested management consultants, and the most brilliant diagnosticians is being made accessible to a small business owner in rural Ohio, a student in Lagos, a startup founder in Bangalore. Not a rough approximation of that expertise filtered through layers of bureaucracy and billable hours. The genuine distillation of the best human thinking ever assembled, available on demand, infinitely patient, and continuously improving at a cost that is rapidly approaching zero.

Stop and sit with that for a moment, because it is genuinely breathtaking. Every hard problem that has plagued organizations, governments, communities, and individuals for generations, the ones that went unsolved because the expertise needed was too rare, too expensive, or concentrated in too few hands, is suddenly within reach. The democratization of human knowledge at this scale dwarfs the printing press. It dwarfs the internet. The accumulated intellectual capital of our entire species, available to everyone, everywhere, simultaneously.

You would think this would be an unambiguous triumph. A cause for genuine celebration. The greatest gift the modern era has ever produced.

It is, in fact, a disaster. And the speed at which that disaster is arriving will catch most people completely off guard.

The Hidden Cost of the Miracle

Here is what the breathless optimism about AI tends to skip past: that extraordinary expertise now flowing freely and cheaply to everyone was previously delivered by human beings. Human beings with mortgages and student loans they took out because they were told that education was the path to security. People with children, careers they built over decades, and lives they planned around a social contract that said expertise has value and that value gets compensated.

That contract is being shredded.

Andrew Yang recently described what is coming as “the great disemboweling of white collar jobs,” predicting that the automation wave will kick millions of white collar workers to the curb within the next twelve to eighteen months. Not gradually and not in some manageable trickle that allows for orderly adaptation. In a wave, driven by competitive pressure, as one company after another discovers that the economics of AI make human knowledge workers an expense that is increasingly difficult to justify. The stock market will reward companies that cut headcount and punish those that don’t. The incentive structure is already in place, the technology is already capable, and the only variable is timing. The timing is now.

Matt Shumer, founder and CEO of OthersideAI, whose recent essay on this subject has been viewed over eighty million times, offered a description that should stop anyone still treating this as a distant concern. He writes that he tells the AI what he wants, walks away from his computer for four hours, and comes back to find the work done. Not a draft. Not a starting point. The finished thing, done better than he would have done it himself and requiring no corrections. He is not describing a future capability. He is describing his present reality. And he is clear that the experience tech workers have had over the past year, watching AI go from helpful tool to something that does their job better than they do, is the experience that everyone else is about to have. Law, finance, consulting, accounting, analysis, design, not in a decade but in one to five years, and many inside the industry believe the real number is considerably less. Anthropic’s own CEO Dario Amodei has publicly predicted that AI will eliminate 50% of entry level white collar jobs within that window, and people who work alongside this technology every day think he may be underestimating it.

The social contract that an entire generation honored is unraveling. Study hard, accumulate the debt, earn the credential, enter the knowledge economy and build a career. That path led millions of people into consulting, law, finance, accounting, project management, and knowledge work of every variety. These are not surplus workers with easily transferred skills. They are professionals in the middle of careers they planned lives around, carrying financial obligations sized to incomes they had every reason to expect would continue.

The Architecture of the Collapse

To understand why this disruption is unlike anything that came before, you have to understand what is actually being built right now and how quickly it is scaling.

A new category of business has emerged around a simple and devastating model: A firm identifies the single most exceptional practitioner in a given domain, compensates that person at multiples of their previous earnings to continuously curate and feed their knowledge into a purpose-built system, and then delivers that system to tens of thousands of clients at a cost that undercuts every other option in the market by an enormous margin. Not the expertise of an average practitioner and not the expertise of a good one. The best expertise ever assembled in that field, infinitely available and priced at a level that makes the conventional market for that expertise economically indefensible. Companies like XpertTwin are already building what they call human digital twins of experts, AI replicas capable of delivering the advisory and analytical capacity of exceptional practitioners to virtually unlimited audiences simultaneously, in over 100 languages. Sensay is developing personal AI replicas it calls sonas, virtual representations of individuals that continue learning and acting on behalf of those individuals over time, keeping their knowledge current and deployable even when they are not in the room.

The business logic underlying all of this is straightforward and, once you see it, very difficult to argue against. Why would any rational buyer choose otherwise? And as Shumer pointed out, this disruption is categorically different from every previous wave of automation for one critical reason. AI isn’t replacing one specific skill. It is a general substitute for cognitive work that improves across everything simultaneously. When manufacturing automated, displaced workers could retrain as office workers. When the internet disrupted retail, workers moved into logistics and services. Those transitions worked because there was always a domain AI hadn’t yet reached. That is no longer true, and there is no safe harbor waiting on the other side of this one.

The Machine That Cannot Be Stopped

This is where I want to be direct, because it is where optimists tend to reach for regulatory solutions or arguments about human preference for human connection or the idea that society will simply choose a different path.

Unfortunately, however, capitalism has demonstrated with remarkable consistency that it optimizes for efficiency and return regardless of the social cost of doing so. You need look no further than the history of climate change to understand the pattern. The evidence of the damage was clear for decades, the economic incentives pointed in one direction, and the system followed the incentives, rerouting around every obstacle and rationalizing every deferral. We are living with the consequences of that choice now. The same dynamic will play out here and likely faster, because the competitive pressure arrives before any regulatory framework could possibly be designed, passed, implemented and enforced at the necessary scale. The moment one competitor in any market demonstrates that AI-enabled delivery is more economical than human-enabled delivery, the pressure on everyone else becomes existential. Firms that maintain legacy staffing models will be priced out by firms that don’t, and the cascade will be swift.

Yang sees this clearly, which is why his tone is not triumphant but genuinely sorrowful. He has spent years warning that this was coming, and now that it’s arriving he is focused on what we should be doing and how little of it is actually being built. No serious safety net exists at the scale this displacement will require. No retraining infrastructure is in place that could absorb what’s coming. The downstream consequences he describes are not abstractions.

The Weight of What’s Coming

The families that will be most devastated are the ones that believed most completely in the promise of the knowledge economy. The ones who took on the debt, earned the degree, entered the profession, and built a life around the reasonable expectation that expertise compounds in value over time. The ones who are right now, today, in the middle of careers they have no obvious way to exit and no clear path to rebuild elsewhere, because there may not be an elsewhere this time.

Those of us who have spent serious time inside this technology and watched it do in twenty minutes what used to require days of senior expertise have largely been giving the people around us the polite version of what we’re seeing. The cocktail party version. The one that doesn’t make us sound unhinged. Shumer acknowledged this directly. He said the reason people in the industry are sounding the alarm isn’t because they are making predictions. It’s because this already happened to them, in their own work, and they are warning that everyone else is next.

The greatest democratization of knowledge in human history is simultaneously the most serious threat to the economic security of knowledge workers the world has ever produced. Both of those things are completely true. The miracle and the disaster are the same event, seen from different angles.

There is no obvious political mechanism to stop it and no credible regulatory framework with the reach or speed to contain it. The people raising the alarms right now are not guessing. They are reporting.

There may be no way to stop this. There may simply be no hope...

...or is there?

Next week, I’ll lay out what you must be doing right now, regardless of your field, your seniority, or how secure you believe your position to be, to avoid becoming a casualty of a transformation that is already well underway.

A colleague shared a story with me recently that I have not been able to stop thinking about. A consulting firm submitted a deliverable to a client, a real paid engagement, and somewhere in the document they had forgotten to remove the prompt they had used to generate it. Not the output. The actual instruction they had typed into an AI model, which made clear they had no prior knowledge of the subject and were asking the model to write about it as if they were an expert. The client saw it. The relationship did not survive.

I want to be careful not to make this story carry more weight than it should, because the mistake of leaving a prompt in a document is just carelessness, and carelessness has always existed in consulting. What the story actually illustrates is something much broader and more interesting, which is that AI has made it possible for almost anyone to produce content that sounds expert, and that the volume of that content is now so high that clients are increasingly, and reasonably, wondering whether the people they are paying are genuinely bringing something the client could not have generated themselves. In many cases, they are right to wonder. The line between synthesized expertise and real expertise has never been harder to see from the outside, and that is a problem for everyone who has spent years building the real thing.

This article is about how to make that line visible again.

Subscribe now

Share

The Tells That Give AI Away

Before we talk about what genuine expertise looks like, it is worth understanding what generated expertise looks like, because the contrast is where the real insight lives.

The most consistent signature of AI-generated analysis is its commitment to balance and completeness. Ask a model to assess a strategic decision and it will present all sides with careful evenhandedness, offer multiple options with symmetrical pros and cons, and studiously avoid taking a position that could be challenged. The output is coherent, well-structured, and covers the terrain. It is also, almost always, devoid of the friction that real-world expertise produces, because friction requires contact with a specific reality, and models are not in contact with your client’s specific reality. They are in contact with the averaged pattern of many realities, which sounds plausible in the abstract and often breaks down in the particular.

The language of AI output is sanitized in a related way. It favors terms that are currently fashionable in the relevant professional community, avoids framings that might create tension, and tends to present everything at the same level of emphasis because the model has no way of knowing which part of the problem actually matters most in this specific organization at this specific moment. Everything gets equal airtime. Real problems are not like that. Real problems have one thing that will determine the outcome and several things that are basically noise, and any expert who has spent time in the domain knows which is which.

AI also has a particular relationship with failure that is worth naming. It knows the theory of failure very well. It can tell you that change management initiatives often face resistance, that data migration projects frequently underestimate complexity, that stakeholder alignment is critical. All of that is accurate and none of it is useful to a client who needs to know specifically where their initiative is going to break and why. The difference between category-level risk language and component-level failure knowledge is one of the clearest tells in professional discourse, and clients who have been through difficult implementations can feel that difference even when they cannot articulate it.

Finally, and perhaps most importantly, AI-generated expertise has no visible history. It arrives fully formed, without any sign that a position was held, challenged, revised, and rebuilt into something more durable through contact with reality. There is no moment where the thinking changed. There is no before and after. The conclusion appears as if it were the only possible conclusion, which is exactly how conclusions look when they have never been seriously tested.

The Call to Action: Eight Ways to Prove You Are the Real Thing

Understanding the tells is the first half. The second half is knowing what to do about them, not defensively, not as a performance of authenticity, but as a genuine practice of demonstrating the contextual expertise that separates the people who have done the work from the people who have prompted their way to a deliverable. Each of the following techniques directly addresses one or more of the AI tells above, and each of them works precisely because it requires something a model cannot supply.

1. Invite Challenge in Real Time

Mid-conversation, not at the end as a courtesy, ask your audience to push on what you just said. The move sounds like: “Before we go further, I want you to stress-test this. Where does it break against what you know about your organization that I do not yet know?” This is the one thing a person who simply prompted a model beforehand cannot survive when the questions get specific. AI defends coherence. Humans trust resilience. By inviting attack, you are signaling that your thinking was built in friction and you are comfortable returning it there, which is something no generated output will ever do on its own.

2. Name the Tradeoff and Choose a Side

AI avoids taking positions because taking a position means being willing to be wrong in a specific and accountable way. The antidote is to name the tradeoff explicitly and then choose a side out loud: “This improves speed to market at the cost of regulatory flexibility. I am recommending speed because the flexibility risk is survivable and the timing risk is not.” Notice that this sentence owns the downside. When someone can describe the weakness of their own recommendation with more precision than the upside, you are almost certainly not listening to a model.

3. Reference Negative Knowledge

AI knows what works. What it does not have is the memory of what does not work, specifically and sequentially, at the level of detail that only comes from having been there when something broke. Failure does not cluster cleanly in training data the way success does. Success gets written up and published. Failure lives in memory. The move sounds like: “We tried this once and it collapsed in month four when the legacy data hit the integration layer” or “This recommendation is the standard one, and it is usually abandoned six months later for reasons that are completely predictable if you have seen it before.” That texture of specific, sequenced failure is one of the hardest things to fake, and clients who have been through difficult implementations can feel the difference even when they cannot name it.

4. Use Asymmetry Instead of Symmetry

AI generates balanced, symmetrical output because it has no way of knowing which part of the problem actually carries the weight. Real systems are lopsided. The deliberate use of asymmetry in how you present ideas signals that you know where the load-bearing element is and are not pretending the other items on the list matter equally. Make the imbalance explicit: “I am going to spend most of our time on this one constraint because it is the thing that will determine whether this works.” Comprehensive lists are easy to generate. Knowing which item on the list is the one that matters is expertise.

5. Change Your Mind in Public, With an Explanation

AI updates silently. Humans update socially, in front of other people, with an account of what shifted their thinking. Most consultants avoid this because they think it signals weakness, but it actually signals the opposite, which is that their beliefs have been in contact with reality and have moved accordingly. The move sounds like: “I used to think the right sequencing was to start with the technology layer. After seeing three clients try that in a regulated environment, I no longer think that, and here is why.” That statement requires a specific before, a specific after, and a credible explanation of what changed, none of which can be prompted into existence.

6. Answer the Question Behind the Question

AI answers the question it was asked. Experts diagnose the question behind it, the real decision being made, the anxiety underneath the inquiry, the political context that makes the stated question somewhat beside the point. The move sounds like: “Before I answer that directly, let me say what I think you are really trying to decide.” In high-stakes consulting situations, the stated question is rarely the real one, and the expert who surfaces the real one demonstrates a quality of contextual attunement that no generated response can replicate.

7. Introduce Constraints Nobody Asked For

AI solves within the box. Experts redraw it. One of the clearest signals of genuine domain knowledge is the habit of introducing a constraint or consideration that was not in the original question, because you have seen what happens when it goes unaddressed at this stage. The move sounds like: “I am going to add something to this analysis that was not in your brief, because it will matter later and it is much harder to address once you are further in.” That signals foresight built from experience, not recall built from pattern matching.

8. Admit Uncertainty and Then Bound It

AI fills uncertainty with language. Experts map it. There is a significant difference between a hedge that leaves the client no better off and a response that says: “I do not have a definitive answer yet, but I can tell you it is almost certainly one of two outcomes, and here is what would push it either way.” The second version is actionable. It demonstrates that you have thought carefully enough about the problem to know what you do not know, which is itself a form of expertise and one that is considerably harder to generate than a confident-sounding response that papers over the gap.

The Kingdom Is Still Human Territory

All of these techniques point toward the same underlying reality, which is that the expertise clients are actually paying for has always been context rather than content. Content is king, but context is kingdom, and the kingdom is the specific, lived, relational understanding of what is actually happening inside a particular organization at a particular moment, why the history looks the way it does, what the real constraints are, and what any solution needs to survive in order to actually work in the environment it is being deployed into.

That kind of context is not in any training dataset. It accrues through presence, through repeated engagement with the specific terrain of a specific client’s reality, through the trust that builds when you have been in the room for the hard conversations and have demonstrated that your judgment can be relied upon when things get complicated. AI can help you move faster, research more broadly, and structure your thinking more clearly, and any consultant who is not taking advantage of that is operating at a real disadvantage. But the key to the kingdom, the thing that earns genuine trust and genuine influence in a world where polished content is essentially free, is the contextual knowledge and demonstrated judgment that only a human being who has done the work can carry.

The consultants who thrive in this environment will not be the ones who use AI most fluently. They will be the ones who use AI to accelerate their journey to genuine expertise, and then do the entirely human work of earning their place in the client’s kingdom by understanding it better than anyone else in the room.

Content is king. But context is kingdom. And the kingdom is still, unmistakably, human territory.

Share

Subscribe now

I have been skeptical about higher education’s ability to adapt to artificial intelligence, not because universities lack intelligence or resources, but because academic institutions operate on fundamentally different timescales than technological innovation. Curriculum development is traditionally slow moving, consensus driven, and committee approved, often taking years to implement meaningful changes. Meanwhile, AI evolves with breathtaking speed: new models appear, capabilities leap forward, and best practices shift not just monthly but sometimes weekly.

Picture a committee meeting where faculty members are still debating whether to approve a course on GPT-3 while the rest of the world has moved through GPT-4, 4.5, and is now experimenting with models that can generate videos, analyze complex datasets, and occasionally write poetry that doesn’t make you cringe. This inherent mismatch between academic pace and technological acceleration seemed like an insurmountable barrier, like trying to write a definitive guide to a river that keeps changing course.

That’s why discovering Ethan Mollick’s work and reading his book Co-Intelligence has been unexpectedly reassuring. Mollick, a professor at Wharton, demonstrates that parts of academia are not merely reacting to AI but actively experimenting with it, teaching alongside it, and fundamentally reshaping how students engage with learning itself.

The Mindset Gap: Our Real Organizational Challenge

When we examine why businesses struggle to integrate AI meaningfully, we often focus on technical barriers: infrastructure, data quality, integration challenges. But the real bottleneck is something far more fundamental: mindset. Most organizations today are attempting to retrofit transformative technology into mental models built for a pre-AI world.

It’s like watching someone try to use a smartphone exclusively as a really expensive calculator. Sure, it can do that, but you’re missing approximately 99% of its potential while also looking slightly ridiculous in the process.

This creates a predictable pattern of adoption: add a tool, run a pilot project, create a governance policy, implement risk controls, and hope for measurable productivity gains. We’re essentially trying to fit a rocket engine onto a horse drawn carriage and wondering why the performance improvements are underwhelming.

What we rarely see is the more challenging but necessary transformation: redesigning how we think, how we structure workflows, how we approach decision making, and how we understand creativity itself. This kind of paradigm shift is extraordinarily difficult when your mental operating system predates the tool you’re trying to master. Current professionals are learning AI as a second language; they will always carry an accent shaped by their pre-AI training.

The students entering the workforce over the next decade will not carry this constraint, but only if we teach them correctly, starting now.

Why Academia Represents Our Highest Leverage Point

Businesses understandably want practical AI adoption, but they operate under constraints that limit genuine experimentation. Revenue pressures, compliance requirements, operational risks, and quarterly performance expectations create environments where open ended exploration feels dangerously inefficient. Try telling your CFO you’d like to spend three months experimenting with AI “just to see what happens” and watch how quickly you’re escorted to a performance improvement plan.

Universities, however, can. In fact, they should.

Classrooms represent one of the few remaining environments where structured exploration is not just permitted but expected, where productive failure is educational rather than career limiting, and where strange prompts, unexpected outputs, and unconventional workflows can be examined rather than suppressed. If AI becomes native to education rather than supplementary, students will graduate with a fundamentally different relationship to knowledge work itself.

They will not ask “Should I use AI for this task?” but rather “How should I think with AI to approach this problem most effectively?” It’s the difference between someone who learned email as an adult (and still prints everything) versus someone who cannot imagine a world without it.

Teaching With AI, Not Just About AI

What makes Mollick’s approach particularly encouraging is that he does not treat AI as a topic to be studied at arm’s length, like an exotic specimen behind glass. Instead, he incorporates it as an active participant in the learning process itself. His experiments with students include using AI to challenge their assumptions, generate competing perspectives, serve as both cowriter and critic, and test how human judgment evolves when machine generated output becomes part of the creative process.

In Co-Intelligence, Mollick describes AI as creating a “jagged frontier” of capabilities, excelling at some tasks while failing surprisingly at others in ways that don’t align with human intuition. AI can write a compelling marketing campaign but struggle to count the number of Rs in “strawberry.” It can analyze complex legal documents but occasionally insist that Rome is the capital of France with absolute confidence. Understanding this jagged landscape requires direct experimentation, not theoretical instruction.

These are not static lectures about technology features or cautionary tales about automation. They are dynamic exercises in collaborative intelligence that prepare students to navigate a world where the boundaries between human and machine contribution are increasingly fluid. It’s learning to dance with a partner who sometimes leads brilliantly and sometimes steps on your toes, and developing the judgment to know which is happening.

Adaptive Frameworks Over Static Knowledge

Here lies the uncomfortable truth about AI and education: traditional curriculum cycles cannot possibly keep pace with model releases and capability evolution. By the time a formal course titled “Introduction to Generative AI” receives approval, navigates committee review, and gets scheduled, three generations of systems may already be functionally obsolete. It’s like creating a comprehensive guide to MySpace in 2024, technically accurate for a brief historical moment, but utterly useless for navigating the current landscape.

The solution is not attempting constant curriculum revision to match technological change. Instead, we must teach adaptive frameworks: meta-skills that transcend any particular model or platform. Students need to learn how to evaluate model output critically, design effective prompts that account for system limitations, verify and challenge results rather than accepting them uncritically, and manage cognitive offloading responsibly without atrophying their own analytical capabilities.

Mollick emphasizes in Co-Intelligence the importance of understanding AI not as infallible oracle but as “confident bullshitter,” systems that generate plausible sounding output without genuine understanding or commitment to truth. They’re like that friend who can speak authoritatively about any topic at a dinner party, even ones they know nothing about, and sound completely convincing until someone actually checks the facts.

Teach the interaction principles, not the tool features. Students who understand these deeper patterns will consistently outpace those who memorize today’s specific interfaces and capabilities.

Historical Precedent and Future Trajectory

We have navigated this kind of transition before. During the Industrial Revolution, mechanization displaced numerous traditional jobs and disrupted entire economic sectors. But entirely new industries and professional categories emerged because a generation learned to think in mechanized systems, to design processes around new capabilities rather than simply trying to preserve old methods. We didn’t need a world full of expert buggy whip manufacturers; we needed people who could imagine what automobiles made possible.

We stand at a similar inflection point in knowledge work. Some roles will contract, others will vanish entirely, and many will transform beyond recognition. Entirely new categories of work will emerge that we cannot yet clearly envision, jobs that will be built around human AI collaboration in ways we haven’t imagined yet.

Students who grow up with AI as a native collaborator rather than a foreign tool grafted onto existing workflows will invent approaches, methodologies, and entire industries that current professionals would never think to design. They will not be constrained by “how things have always been done” because their foundational experience includes AI from the beginning. They’ll look at our current AI anxiety the way digital natives look at their parents struggling to understand why you can’t just “save the internet to a disk.”

A Call for Accelerated Academic Experimentation

We need more professors running AI integrated courses across disciplines, more departments willing to allow syllabus level experimentation even when outcomes are uncertain, and more cross disciplinary collaboration exploring AI’s implications across business, humanities, sciences, and arts. This cannot be relegated only to computer science departments or business schools.

What we need is not perfect frameworks that take years to develop and are obsolete upon arrival, but fast frameworks that evolve through iteration. Not rigid rules that attempt to control every aspect of AI use (good luck with that), but guided exploration that teaches judgment. Not governance approaches that begin with fear and restriction, but engagement strategies that prioritize literacy and understanding.

The real risk is not that students will use AI imperfectly or make mistakes in their early experiments. They will absolutely do both of those things, probably frequently, and that’s fine because it’s called learning. The real risk is that we delay teaching them to use it thoughtfully, that we allow bureaucratic caution to create a generation that encounters AI only after their cognitive patterns and professional habits have already solidified around pre-AI paradigms.

Mollick demonstrates in both his teaching and his writing that this integration is possible right now, not in some carefully planned future after we’ve formed the right committees and conducted the right studies. His students are already learning to collaborate with AI, already developing intuition about its strengths and limitations, already building the mental models that will serve them throughout their careers.

The future of meaningful AI adoption will not be decided by which company deploys the most sophisticated software or which consulting firm develops the most comprehensive implementation methodology. It will be decided by which generation learns to think with AI first, whose foundational education includes collaborative intelligence as a native element rather than a foreign addition.

That decision is being made right now, in classrooms and curriculum committees, by professors willing to experiment and institutions willing to support them. Our best chance at navigating this transformation successfully lies not in the executive suites where AI strategies are debated (often with impressive PowerPoint decks and very little actual understanding), but in the classrooms where the next generation is learning to think.

And if we’re lucky, they’ll eventually teach the rest of us how it’s actually done.

Subscribe now

Share

The other day, my son and I had an argument over whether AI could do anything truly novel or creative. It was an emotional conversation, the kind that starts intellectual but quickly becomes unsettling when you realize what’s really at stake. We weren’t just debating technology. We were wrestling with what it means to be human in an age when machines can paint, compose, and write stories that move us.

His argument was elegant in its simplicity: If AI is built on large language models trained entirely on human generated content, then how could it ever produce anything truly original? It’s only recombining what we’ve already created, like shuffling a deck of cards we designed. At best, he argued, AI is a sophisticated parrot, impressive in its mimicry but incapable of genuine creativity.

I understood his concern. The idea that machines might create beautiful art or solve problems we cannot touches something primal in us, something that wants to believe human creativity is sacred, unreproducible, ours alone.

But then I told him about Move 37.

The Move That Changed Everything

On March 10, 2016, in a hotel ballroom in Seoul, South Korea, an AI system called AlphaGo made a move that no human would ever make. It was the 37th move in the second game of a historic match against Lee Sedol, one of the greatest Go players who ever lived. Go, an ancient Chinese strategy game, had long been considered the final frontier for artificial intelligence. Far more complex than chess, with more possible board positions than atoms in the universe.

When AlphaGo placed its stone on the fifth line from the edge of the board, the commentators (themselves master level players) were baffled. “That’s a very strange move,” said one. “I thought it was a mistake,” admitted the other. Lee Sedol left the room for nearly fifteen minutes, trying to make sense of what he’d just witnessed.

Here’s what makes Move 37 truly remarkable: AlphaGo had calculated there was a one in ten thousand chance that a human would make that move. The AI knew it was playing outside the bounds of 2,500 years of human Go strategy. But when it drew on the knowledge it had accumulated by playing millions of games against itself (looking ahead into the future possibilities of the match) it made the move anyway.

And the move was genius.

It turned the course of the game. AlphaGo won. And in that moment, something fundamental shifted in how we understand machine intelligence. This wasn’t regurgitation. This wasn’t pattern matching within known human boundaries. This was genuine novelty. An idea that emerged from analyzing human strategy, then transcending it entirely to discover possibilities we never imagined.

Fan Hui, the European Go champion who had lost to AlphaGo months earlier, kept repeating the same phrase as he watched: “So beautiful. So beautiful.”

March 10, 2016, should be recognized as a landmark date in human history. The day we learned, publicly and undeniably, that AI could create ideas beyond the limits of human imagination.

The Myth of Pure Originality

My son’s argument rests on a flawed premise: that human creativity springs from some mystical source untouched by what came before. But that’s not how invention works. That’s not how it’s ever worked.

Consider the invention of the airplane. The Wright brothers didn’t conjure powered flight from thin air. They studied birds, analyzed the work of Otto Lilienthal and Octave Chanute, applied principles of bicycle mechanics, and built on centuries of accumulated knowledge about aerodynamics, materials science, and propulsion. Their genius wasn’t in starting from scratch. It was in synthesizing existing concepts in a novel configuration that solved the problem of controlled, sustained flight.

Or take the discovery of penicillin. Alexander Fleming’s breakthrough came from observing that a mold contaminating his bacterial cultures was killing the bacteria around it. The mold itself wasn’t new. Bacterial cultures weren’t new. What was new was Fleming’s recognition of the pattern and its implications. Connecting observations in a way that led to one of medicine’s most transformative innovations.

This is how all human innovation works: we take existing knowledge, recognize patterns others missed, make unexpected connections, and synthesize new solutions. We stand on the shoulders of giants not because we’re lazy, but because that’s the only way to see further than they could.

AI does exactly the same thing. Just faster, across more variables, and without the cognitive biases that often blind us to unconventional solutions.

Why This Matters Now More Than Ever

The implications of Move 37 extend far beyond a board game. We face challenges today that human intelligence alone has struggled to solve, not because we lack brilliance, but because the problem spaces are too vast and complex for human cognition to fully map.

Consider cancer research. There are billions of variables in how genes interact, how proteins fold, how cellular pathways respond to different compounds. Human researchers have made extraordinary progress, but we’re limited by how many patterns we can hold in our minds simultaneously, how many correlations we can test in a lifetime. AI systems can analyze these incomprehensibly complex datasets and identify relationships (combinations of factors) that would take human researchers centuries to discover, if we ever discovered them at all.

Or environmental challenges. Climate change involves intricate feedback loops between atmospheric chemistry, ocean currents, vegetation patterns, economic systems, and human behavior. The solutions we need might lie in unconventional combinations of technologies, policies, and approaches that our current paradigms make difficult to imagine. AI unconstrained by “that’s not how we’ve always done it” can explore solution spaces we’ve never considered.

These aren’t hypothetical futures. AI systems are already discovering new antibiotic compounds, optimizing energy grids in ways human engineers didn’t anticipate, and finding mathematical proofs that elegantly solve problems humans had worked on for decades.

The question isn’t whether AI can be creative. Move 37 settled that question eight years ago. The question is whether we’re wise enough to harness that creativity.

The Indispensable Human Element

But here’s what my son got right: there’s a profound difference between AI’s capacity for novel pattern recognition and the full scope of human creativity. Because creativity isn’t just about generating new ideas. It’s about understanding which ideas matter, why they matter, and how to bring them into the world in ways that serve human flourishing.

AlphaGo made Move 37, but it had no appreciation for its beauty. It didn’t feel the sharp intake of breath from the audience, didn’t recognize the moment as historic, couldn’t explain why the move mattered beyond the cold mathematics of winning probability. It took Fan Hui (a human who had experienced both the game’s traditions and the machine’s capabilities) to see the deeper significance.

AI excels at exploration within defined objective functions, but humans provide the judgment about which problems are worth solving, the empathy to understand how solutions will affect real people, and the wisdom to navigate the ethical complexities that arise when powerful new capabilities emerge.

When AI suggests a novel medical treatment, we need human doctors who understand not just efficacy but patient values, quality of life, and the weight of informed consent. When AI generates creative content, we need human curators who can discern what resonates with the human experience. When AI optimizes business processes, we need human leaders who consider not just efficiency but dignity, fairness, and long term societal impact.

Think of it as Move 37 paired with Move 78. Because in Game Four of that same match, Lee Sedol (stung by three losses) made his own brilliant, unexpected move. Move 78 shocked the commentators just as Move 37 had. It surprised AlphaGo so completely that the machine’s win probability collapsed within minutes. One observer called it “God’s Touch.”

Machine genius met human genius. Each was beautiful. Each was necessary. Neither diminished the other.

The Consulting Afterlife

For those of us who’ve spent decades solving problems for clients, Move 37 represents both a challenge and an invitation. The old consulting model (where we relied primarily on pattern recognition from past engagements, applied established frameworks, and delivered insights bounded by what we’d seen before) is insufficient for the complexity ahead. That model is dying, not because it was wrong, but because it’s been transcended.

The consulting afterlife isn’t about humans or machines. It’s about recognizing that AI can show us moves we never imagined while we provide the judgment about which moves serve our clients’ deepest needs. It’s about curating AI’s vast exploratory power with hard won wisdom about organizations, people, and change. The consultants who thrive won’t be those who resist the machine or those who blindly defer to it. They’ll be those who learn to dance with it, combining computational creativity with human discernment.

The most powerful engine for our future isn’t AI replacing humans or humans resisting AI. It’s the collaborative model: AI’s capacity for exploring vast possibility spaces combined with human judgment, empathy, and wisdom about what those possibilities mean and how they should be applied.

A Question Worth Asking

My argument with my son didn’t end with anyone declaring victory. But it did end with both of us seeing the issue differently. Which leads to the question we should all be wrestling with:

In a world where AI can generate genuine novelty, what becomes the distinctive human contribution, and are we developing those capabilities as intentionally as we’re developing the technology?

Subscribe now

Share

Every professional I know has the same problem right now.

They’ve got ChatGPT bookmarked. Maybe Claude. Probably tried Perplexity. Downloaded a few specialized tools that looked promising. And now they’re paralyzed by choice, spending more time deciding which AI to use than actually using any of them.

Last month, I watched a sharp associate spend two hours fighting with ChatGPT to build an issue tree, getting increasingly generic output, when the real problem was simple: he was using a conversational model for a task that required deep structural thinking. Wrong tool, right intention.

That’s when I realized: we don’t have an AI capabilities problem anymore. We have an AI decision-making problem.

So I built myself a map. Not a theoretical framework or an academic taxonomy but a practical, working map that answers one question: When I need to do [professional skill], which AI tool gives me the best shot at partner-quality output?

Here’s what I learned building it, and why you need to build your own.

The Integrated Toolkit: Why Specialization Beats General Purpose

Look at the map above. It’s organized around a core truth: you are the AI-augmented professional at the center, and your tools radiate outward based on what kind of thinking they amplify.

I’ve clustered mine into five capability zones, each with specific tools matched to specific professional skills. Not because I love complexity, but because after testing dozens of tools, I learned that specialized beats general purpose almost every time.

Zone 1: Strategy & Problem Solving (Blue cluster, top left)

This is where you’re wrestling with ambiguity. The client says “sales are down” and you need to figure out if it’s a pricing problem, a product problem, a people problem, or a pipeline problem. You’re generating frameworks, stress-testing assumptions, scanning competitive landscapes.

The AI superpower here: Reasoning models that can think for 30 seconds before answering, exploring multiple hypotheses simultaneously.

My toolkit:

• Structured Problem Solving: OpenAI o1, Claude 3.5 (when I need deep thinking on novel problems)

• MECE Structuring: MindMeister AI (purpose-built for issue trees)

• Market Research: Hebbia (can read 100 analyst reports in minutes)

• Due Diligence: Hebbia again (built for deal rooms)

• Strategy Development: Quantive StrategyAI (scenarios and strategic option generation)

• Requirements: Wakai AI (converts messy inputs into structured requirements)

• Risk Management: Wrike AI (continuous risk monitoring)

Why these specific tools? Because generic chatbots give you generic strategy. When the cost of being wrong exceeds the cost of thinking longer, you need specialized reasoning, not fast autocomplete.

Zone 2: Analysis & Data (Green cluster, bottom left)

Now you’re in the guts of the work. Building financial models. Mapping processes. Analyzing data. Benchmarking performance. This is where most professional hours get spent, and where AI can compress weeks into days.

The AI superpower here: Code execution, data manipulation, integration with existing tools.

My toolkit:

• Data Analysis: Julius.ai, IBM WatsonX (natural language to analysis)

• Financial Analysis: Microsoft Copilot for Finance (lives inside Excel, knows finance)

• Analytical Reasoning: Python in Excel (for when I need reproducible quant work)

• Process Mapping: Celonis, n8n (process mining from actual system logs)

• Org Design: Gloat (workforce analytics and scenario modeling)

• Benchmarking: Deel AI (particularly for comp and people benchmarks)

• Requirements Gathering: Jira Service Management (captures and structures requirements automatically)

• Convincement: Cotavixi (testing arguments and messaging)

Why these instead of ChatGPT? Because they’re embedded in the actual work environments. I don’t want to copy data out, paste it into a chat interface, then copy results back. I want AI that lives where my data lives.

Zone 3: Communication & Influence (Gold cluster, top right)

You’ve done the analysis. Now you need to make it land. Build the story. Write the deck. Prep the CEO. Navigate the politics. Get the CFO on board before the steering committee.

The AI superpower here: Understanding audience, crafting narrative, generating variations for different stakeholders.

My toolkit:

• Synthesis & Storyline: NotebookLM (turns messy research into coherent narrative)

• Executive Communication: ElevenLabs (voice rehearsal, testing different tones)

• Stakeholder Alignment: Issoria (maps influence networks and coalition strategies)

• BD & Proposal Writing: Jasper AI (knows proposal structure, compliance requirements)

• Client Relationship Management: HubSpot AI CRM (sentiment analysis on email and meeting history)

The insight: These tools understand persuasion, not just information. They know the difference between “here’s what we found” and “here’s why this matters to you specifically.”

Zone 4: Execution & Management (Purple cluster, bottom right)

The strategy is approved. Now you need to make it real. Build the decks. Run the workshops. Track the projects. Manage the change. Keep the trains running.

The AI superpower here: Automation, template application, consistency at scale.

My toolkit:

• Slide Writing: Gamma, Beautiful.ai (auto-layout, design systems)

• Data Visualization: Hal9, Syft Assist AI (chart recommendation, insight narration)

• Workshop Facilitation: Fireflies.ai, Granola (live transcription, action items, readouts)

• Project Management: Motion, Productive.io (intelligent scheduling, dependency tracking)

• Change Management: Pandatron (persona segmentation, adoption tracking)

• Time Management: Reclaim.ai (because even professionals need to defend their calendars)

Why specialized tools here? Because in implementation, consistency and speed matter more than creativity. I want templates that work, not infinite variations.

The Three Principles Behind the Map

1. Match Tool Capability to Cognitive Demand

Not all professional tasks require the same level of AI sophistication:

• High-stakes, novel problems → Reasoning models (o1, Claude Opus)

• Structured, repeatable tasks → Specialized workflow tools (Celonis, Jira)

• Communication and persuasion → Language models with audience modeling (NotebookLM, Jasper)

• Grunt work at scale → Automation tools (Motion, Reclaim)

The mistake I see constantly: using ChatGPT for everything because it’s the tool you know. That’s like using a Swiss Army knife to perform surgery. Sure, there’s a blade in there, but you probably want the scalpel.

2. Prefer Domain-Specific Tools Over General Models for Specialized Tasks

General-purpose LLMs (GPT-4, Claude, Gemini) are incredible, but they’re generalists by definition. When you need to:

• Mine process logs → Use Celonis (purpose-built)

• Build financial models → Use Copilot in Excel (knows finance patterns)

• Map stakeholder networks → Use Issoria (built for influence mapping)

• Generate compliant proposals → Use Jasper (trained on proposal structures)

Why? Because these tools have the right context, integrations, and guard rails already built in. They won’t hallucinate a DCF formula or forget that your proposal needs a diversity statement.

3. Context Persistence Matters More Than You Think

The most underrated feature in my map isn’t a specific tool but the ones that maintain context across sessions:

• Claude Projects for multi-week engagements

• NotebookLM for research that builds over time

• HubSpot AI for relationship history

• Motion for project context

Why? Because professional work isn’t one-shot questions. It’s iterative refinement over weeks or months. Every time you start fresh with a chatbot, you pay a “context tax”: re-explaining background, re-uploading files, re-establishing the frame.

Tools with persistent context learn your work as it evolves.

How to Build Your Own Map (Not Copy Mine)

Here’s the thing: my map is useless to you.

Not because the tools are wrong, but because your professional practice isn’t mine. You’re not running SAP S/4HANA implementations for life sciences companies. You might be doing digital transformation for financial services, or growth strategy for tech startups, or operational excellence for manufacturing, or legal work for corporate clients, or marketing campaigns for consumer brands.

Your map needs to reflect your reality.

Step 1: Audit Your Actual Work

Go back through your last three major projects or engagements. For each one, list:

• What skills did you actually use? (Use the 25-skill framework as a starting point, adapted to your field)

• How much time did you spend on each?

• Which tasks were high-stakes vs. routine?

• Where did bottlenecks happen?

Don’t idealize. Look at your calendar and timesheets. Where did the hours actually go?

Step 2: Identify Your “Unfair Advantage” Skills

Which 5 to 7 skills are you genuinely better at than most professionals at your level? These are your differentiation points. You want AI tools that amplify these, not replace them.

For me:

• Structured problem solving

• Strategy development

• Technical SAP implementation (my deep domain)

I invested in the best AI for these because they’re my brand. Everything else? I’m fine with “good enough” AI assistance.

Step 3: Map Pain Points to AI Categories

For each major skill where you’re spending significant time or experiencing friction:

Ask: Is this pain about...

• Thinking (need better frameworks, more options) → Reasoning models

• Doing (need to process data, build models) → Code execution and specialized tools

• Communicating (need to persuade, translate) → Language models with audience awareness

• Tracking (need to stay organized, remember context) → Workflow and project tools

Then research 2 to 3 tools in each category. Don’t just pick the famous ones. Read what professionals in your specific domain are actually using.

Step 4: Test in Low-Stakes Environments First

Do NOT deploy AI on your most important client project as your first test. Instead:

• Use it for internal projects

• Try it on proposals or pitches you’re not counting on

• Run it parallel to your normal process and compare outputs

Build confidence before you bet your reputation.

Step 5: Version Your Map

Here’s what I learned: the map I made in October is already outdated. New models drop monthly. Tools merge or get acquired. Pricing changes.

So I version mine:

• V1.0 (Oct 2024): Initial map, mostly general-purpose LLMs

• V1.5 (Dec 2024): Added specialized tools for process mining and decision trees

• V2.0 (Jan 2025): This version with integrated toolkit concept and domain-specific tools

I review quarterly and update when I discover something meaningfully better.

What I Got Wrong (So You Don’t Have To)

Mistake #1: Tool Hoarding

My first map had 40+ tools. Completely unworkable. I spent more time deciding which tool to use than actually using them.

The fix: Limit to 15 to 20 tools maximum. Force yourself to choose one tool per skill, with maybe one backup. Decision fatigue is real.

Mistake #2: Chasing Shiny Objects

Every week there’s a new “revolutionary” AI tool. I wasted time (and money) trying every new release.

The fix: Only evaluate new tools when there’s a clear gap in your current map. Otherwise, ignore the noise.

Mistake #3: Ignoring Integration Costs

Some theoretically “better” tools required so much setup, API wrangling, or workflow changes that they never got used.

The fix: Bias toward tools that integrate with your existing stack. A 70% solution that lives in Slack or Teams beats a 95% solution that requires a separate login and export process.

Mistake #4: Underestimating the Learning Curve

Not all AI tools are as easy as ChatGPT. Some require prompt engineering. Others need configuration. A few demand real training.

The fix: Be realistic about adoption. If a tool requires 10 hours to learn, is the payoff worth it? Sometimes yes (Celonis for process mining). Often no (some hyper-specialized tools with narrow use cases).

The Map Is Not the Territory

Here’s the uncomfortable truth: having this map doesn’t make me a better professional.

It makes me a faster professional. A more efficient professional. Someone who can deliver partner-quality work with senior associate effort.

But the judgment: knowing which problem framing will resonate with this particular CFO, sensing the political currents in a steering committee, reading between the lines of what a client isn’t saying. That’s still human.

The map just makes sure that when I have that judgment, I can execute on it without drowning in the tactical work.

Your Assignment (If You’re Ready)

This week, don’t try to build your full map. Just do this:

1. Pick your three most-used professional skills (the ones you do weekly)

2. Identify one AI tool for each that you’ll test for the next 30 days

3. Set a decision date: Will you keep it, swap it, or abandon it?

Keep it simple. Iterate. Version.

The goal isn’t the perfect map. The goal is your map, one that actually reflects how you work, where you add value, and where AI can genuinely help rather than just add complexity.

Because in the AI afterlife, the professionals who thrive aren’t the ones with access to the best AI. They’re the ones who know exactly which AI to use, when to use it, and when to trust their own judgment instead.

What’s your experience building your AI toolkit? Hit reply and tell me where you’re getting stuck. I read every response.

And if you found this useful, forward it to one professional who’s drowning in AI tools but not sure which ones actually matter.

Subscribe now

Share

I’ve spent a lot of time in my articles exploring how AI is reshaping consulting—the skills it enhances, the tasks it automates, and yes, the capabilities it threatens to replace. But today I want to flip the script.

But there is a paradox that often gets overlooked: the more AI democratizes technical skills, the more valuable your human skills become.

When every consultant on the planet has access to the same AI powered research tools, the same automated analysis, the same instant deliverable generation, what’s left to differentiate you? The answer isn’t more technology. It’s everything technology can’t replicate.

World Class Organizational Change Management

I work with an incredible senior consultant named David Fraley. David handles organizational change management for our SAP implementations, arguably one of the most human intensive aspects of enterprise transformation work.

Here’s what I’ve observed about David: whether the year is 1940 or 2040, the skills he brings to the table are critical.

That statement might seem hyperbolic until you understand what those skills actually are. David doesn’t just manage change projects. He connects with people navigating uncertainty. He reads the room when a manufacturing supervisor is terrified that the new system will expose gaps in her knowledge. He senses when an IT director is nodding along in meetings but quietly sabotaging the project behind closed doors. He understands that the executive sponsor’s impatience isn’t arrogance—it’s pressure from a board that doesn’t understand implementation timelines.

Every single one of our clients has connected and responded to David, to the betterment of the project. Not because David delivers incredible PROSCI deliverables (though he does). But because he creates something AI cannot manufacture: genuine human trust.

No large language model, not even the most sophisticated agentic system, can sit across from a worried operations manager and make them feel heard. No algorithm can navigate the political minefield of a family owned company where the founder’s grandson is threatened by incoming technology. No AI can sense that the real reason training sessions keep getting “rescheduled” is that the department head never wanted this project to succeed in the first place.

David does all of this, and it’s what separates implementations that succeed from the 70% that fail.

The Evergreen Skillset

So what exactly are these skills that transcend technological eras? Recent research from MIT Sloan identifies several human capabilities that complement AI’s shortcomings—capabilities that become more valuable, not less, as AI advances:

Empathy and Emotional Intelligence

AI can detect emotions through sentiment analysis. It can even simulate empathetic responses. But it cannot experience what another person is going through. It cannot share in someone’s anxiety about job security during a transformation. In professional services, this matters enormously. When you’re asking people to change how they’ve done their jobs for twenty years, they need to feel understood, not analyzed.

Judgment and Ethical Reasoning

AI processes data and produces outputs. It cannot weigh human values or cultural norms. It cannot make a call about whether pushing back a go live date is worth the political capital it will cost. It cannot navigate the gray areas where the “right” answer depends entirely on context only a human can fully grasp.

Critical Thinking in Open-Ended Systems

AI excels at pattern recognition within defined parameters. But consulting rarely operates within defined parameters. We deal with ambiguity, incomplete information, competing stakeholder interests, and situations where the “data” tells you one thing and your gut tells you another. That synthesis, data plus intuition plus experience plus judgment, remains uniquely human.

Cultural Fluency and Contextual Awareness

Here’s something AI fundamentally cannot do: understand the real culture of an organization. Not the values on the lobby wall, but the actual unwritten rules. Who really makes decisions. What happened three years ago that nobody talks about but everyone remembers. Why the Singapore office operates completely differently from the Boston headquarters despite identical org charts.

This cultural intelligence, the ability to read nuances, navigate unstated hierarchies, and adapt your approach based on context, is what separates transformative consultants from competent ones.

Presence and Human Connection

There’s emerging research on what makes roles truly irreplaceable by AI, and “presence” keeps appearing as a key factor. The ability to build connections through physical interaction. To foster innovation through real-time collaboration. To provide reassurance that can only come from another human being. As one researcher put it: “Machines can offer data. Humans offer reassurance.”

The Paradox of AI Democratization

Here’s what most people miss about the current moment in consulting: AI isn’t just changing what we do. It’s changing what matters.

When sophisticated analysis was expensive and rare, it was a source of competitive advantage. Now that AI can generate comparable analysis in minutes, analysis itself becomes table stakes. The premium shifts to interpretation, application, and the human relationships that determine whether insights actually get implemented.

This is the cruel irony for firms that have spent decades building their brands around analytical horsepower. That capability is being commoditized in real time. What’s not being commoditized, what can’t be commoditized, is the ability to help people through change. To inspire belief in a new direction. To earn trust that translates into action.

What This Means For You

If you’re reading this newsletter, you’re probably already thinking about how AI affects your career or your practice. Here’s my challenge to you: don’t just think about which AI tools to adopt. Think about which human skills to deepen.

Because in a world where everyone has access to the same AI capabilities, your differentiation isn’t going to be who has the better prompt engineering. It’s going to be who can do what AI cannot:

• Walk into a tense stakeholder meeting and actually change the emotional temperature in the room

• Listen between the lines of what a client is saying to understand what they actually need

• Adapt your communication style, vocabulary, and approach based on who you’re talking to and what’s really going on beneath the surface

• Build the kind of trust that makes people willing to take risks and embrace change

• Navigate organizational politics with enough finesse to actually get things done

These aren’t nice to haves. In the AI era, they’re your competitive moat.

The Bottom Line

I started this piece talking about David Fraley because he embodies something I want every consultant to understand: the future doesn’t belong to those who adopt AI fastest. It belongs to those who combine AI capabilities with irreplaceable human skills.

David will be valuable whether he’s working with punch cards or quantum computers. That’s not because he ignores technology, he embraces it and has put in his 10,000 hours on the skills that really matter leveraging technology. It’s because he never loses sight of what makes transformation actually work: human beings deciding to change.

The tools evolve. The fundamental truth doesn’t.

Your technical skills can be augmented. Your analytical capabilities can be automated. But your capacity for genuine human connection, nuanced judgment, and earned trust?

Those are evergreen. Cultivate them.

What human skills have you found to be most valuable in your consulting work? I’d love to hear your perspective in the comments.

For me, that tool has been NotebookLM.

Not because it’s some magical replacement for consultants or because it promises to “reinvent everything,” but because it fits incredibly well into the reality of modern consulting — especially in the kind of complex, regulated, enterprise environments I’ve dealt with in my career.

What NotebookLM Actually Is

At a very simple level, NotebookLM lets you upload your own materials — PDFs, documents, links, transcripts — and then ask questions against them. The key difference is that it doesn’t answer from the general internet or vague training data. It answers from your sources, and it shows you where the answer came from.

That may not sound exciting at first, but once you use it for real work, it clicks.

There’s no complicated setup, no prompt gymnastics, no need to understand how models work. If you can drag a document into a browser and type a question, you can use NotebookLM. And it’s free, which still feels a little unbelievable given how useful it is.

A Real Example From the Work We Do

Not long ago, we had to go deep on revenue recognition, specifically IFRS 15 and ASC 606. Anyone who has spent time with these standards knows what that usually looks like: massive PDFs, dense language, edge cases everywhere, and guidance that’s technically precise but very hard to translate into how systems and processes actually work.

In the past, this kind of work would typically concentrate knowledge in a small number of senior people. Everyone else would either wait, skim summaries, or ask the same questions repeatedly and hope they were interpreting things correctly.

Instead, we took all of the source material — the thousand-page documents, the detailed guidance, the examples — and dropped them into NotebookLM.

What changed almost immediately was how we interacted with the information. We weren’t just reading or summarizing anymore. We were asking questions that reflected our real use cases. How does this apply to a specific delivery model? What matters for this type of contract? How would I explain this to a client without using accounting language?

NotebookLM didn’t just give us shorter versions of the text. It helped us build a focused, searchable knowledge base that reflected how we actually think and work.

Why This Has Mattered in our Practice

This is where it stopped feeling like an interesting tool and started feeling genuinely useful for our practice.

By making deep, complex knowledge more accessible, we’ve been able to raise delivery quality while also reducing friction. Junior consultants don’t have to guess or feel uncomfortable asking “basic” questions. They can explore the material in plain language, at their own pace, and still stay anchored to the real standards.

For senior consultants, it means less time repeatedly explaining the same concepts and more time spent on judgment, design decisions, and client conversations. For the practice overall, it’s meant more consistency across teams, faster onboarding, and better outcomes delivered at lower cost.

The important thing is that this isn’t about replacing expertise. It’s about distributing it more effectively. NotebookLM becomes a shared layer of understanding that everyone can tap into, regardless of tenure.

More Than Just Text on a Page

One of the underrated aspects of NotebookLM is how flexible it is in the way people can learn from it. Sometimes a written summary is enough. Other times, having a conversational explanation or an audio-style walkthrough makes the difference. Being able to move between those modes, all grounded in the same source material, has been incredibly helpful for different learning styles across the team.

It’s also changed how we think about enablement. Instead of static documents that age quickly, we now have living notebooks that evolve as the practice evolves.

The Bigger Shift I’m Seeing

What NotebookLM really represents to me is a shift in how knowledge flows through a consulting organization. Instead of expertise being locked in documents or people’s heads, it becomes something you can interact with naturally. You ask questions the way you’d ask a colleague, and you get answers that are grounded, explainable, and relevant to your context.

And again, the simplicity matters. There’s no long adoption curve here. No transformation program. No heavy investment. You can start small, experiment, and immediately see value.

That it’s free almost feels secondary, but it’s worth saying out loud: the barrier to entry is essentially zero.

If you’re in consulting and you haven’t spent time with NotebookLM yet, I’d strongly encourage you to try it. Not because it’s flashy or revolutionary, but because it quietly makes teams more confident, more consistent, and more effective at the work they already do.

Has anyone else had some experience with Notebook LM they would like to share?

Subscribe now

Share

Share

Subscribe now

Every Friday afternoon at my first client engagement, leadership would gather in a conference room. No consultants invited.

They were deciding who to fire that week.

For twelve straight weeks, they had been cutting one or more consultants from our team. Partners. Managers. Associates. Nobody was safe. The client believed they were paying for seasoned experts and getting freshly minted graduates instead. They weren’t wrong. I was one of them.

I had arrived just days earlier, and I realized something that changed the trajectory of my career: if I didn’t overdeliver—immediately and consistently—I could be gone by next Friday.

So I made a decision that still shapes how I work today. If I was going to fail, it would never be because I hadn’t put in the work.

The Effort Equation

Angela Duckworth’s research on grit crystallizes something I learned the hard way in those early consulting years:

Talent × Effort = Skill

Skill × Effort = Achievement

Effort shows up twice. It’s the multiplier that bridges the gap between raw talent and meaningful achievement. I wasn’t the most polished or experienced consultant in that room. But I could out-prepare, out-research, and outwork just about anyone.

That first engagement, I worked fourteen to sixteen hour days, seven days a week. Arrived at 7 a.m., stayed until 9 p.m., then went home and prepared for another three to four hours for the next day. Every client question got treated like a final exam. Every meeting felt like a performance review.

There’s an old saying in consulting: you only have to be one day ahead of the client. I tried to stay three. Sometimes I barely managed one or two, but the intent and preparation mattered.

Fear may have been the initial driver. Discipline was what kept me going.

What’s Dying vs. What Endures

Here’s the thing about discipline in the AI era: the nature of the work has transformed, but the underlying principle hasn’t moved an inch.

What’s dying: The endless manual slog of preparation that once consumed consultants’ evenings. Hours spent combing through binders, building slide decks by hand, tracking data across disconnected systems. That grinding mechanical labor? AI handles it now.

What endures: The mental habits that let you perform consistently when it matters most. The ability to prepare thoroughly when others cut corners. To stay focused when distractions multiply. To maintain quality standards even under pressure.

In the old days, my fourteen-hour days were a form of manual navigation through dense technical complexity. I used raw grit to painstakingly map out SAP manuals and client requirements, staying one step ahead through sheer volume of effort.

Today, discipline has evolved from manual labor to systematic orchestration. You still need grit, but you apply it to calibrating the tools. Instead of spending ten hours trekking through data, you spend those hours perfecting the prompts and logic that allow AI to illuminate ten thousand pages in ten seconds.

Your value is no longer found in how much complexity you can endure, but in how quickly you can dissolve it to reveal the path forward.

Four Principles That Still Apply

Effort as the Great Equalizer. Just as I learned facing those Friday firing meetings, discipline and preparation can overcome experience gaps. Endurance outlasts enthusiasm. Clients will always value consultants who show up ready, informed, and dependable. AI handles the mechanical prep now—but the consultant’s role shifts from performing manual labor to gathering insights, validating assumptions, and making strategic decisions.

Preparation Over Guesswork. I tried to stay three days ahead of the client instead of the traditional one. Consultants have always been judged on their readiness. Preparation signals respect for the client, the project, and the stakes. Today, AI can simulate client interactions and anticipate challenges before you ever walk into the room. Preparation becomes proactive planning rather than reactive scrambling.

Knowledge Compression. Great consultants digest massive amounts of information and present it back with clarity. I absorbed SAP’s complexity by treating it like video game levels—each area of configuration like a new challenge, each deadline like a boss fight. That skill of simplifying complexity remains timeless. AI accelerates it. The role shifts from absorbing everything manually to asking the right questions, validating outputs, and applying insights at a higher strategic level.

Trust as Currency. No matter how advanced technology becomes, consulting runs on trust. My survival at that first client came down to them believing in my reliability and preparation, not just my technical knowledge. AI can supply knowledge at speed, but consultants must channel it into human presence and judgment. Trust remains paramount—AI just helps you earn it more effectively.

The Paradox

Discipline creates freedom.

When preparation becomes automatic, you’re free to focus on higher-level thinking. When your work habits are solid, you can take bigger risks. When clients trust your thoroughness, they give you access to more interesting challenges.

This isn’t about perfectionism or burnout. It’s about building a foundation strong enough to support the career you actually want.

In a world where knowledge is abundant and AI can generate answers instantly, your competitive advantage isn’t what you know. It’s how reliably you can apply what you know to create value for others.

That reliability is entirely within your control.

What’s your take? Did you have a “Friday firing meeting” moment early in your career that shaped your work ethic? I’d love to hear your story in the comments.

s

Subscribe now

are hungry for practical workflows, not philosophical hand wringing about whether AI will replace us.

But here’s the uncomfortable truth nobody wants to discuss: the economics that made consulting profitable for 50 years are breaking. Not slowly or eventually, but right now.

If you’re billing time and materials for work that AI can do in minutes, you’re not just leaving money on the table. You’re training your clients to see you as overhead.

This piece is about what comes next.

The Pricing Model That Built the Industry (And Why It’s Dying)

Traditional consulting economics are simple:

Revenue = Hours × Rate × Utilization

Every partner knows this formula. Every staffing model optimizes for it. Every promotion decision reflects it.

The model worked because of an implicit assumption: more complexity requires more hours, and more hours from more senior people.

AI breaks that assumption in three ways.

Compression. Work that took a team of three analysts two weeks now takes one consultant two days. The Synthesis Engine prompt from last week can turn 40 hours of interview coding into 4 hours of structured insight. The hours are gone and they’re not coming back.

Commoditization of the Middle. The middle of consulting, including research, synthesis, deck building, and option generation, is exactly where AI is strongest. These tasks are low context, pattern rich, and infinitely repeatable. If your value proposition lives in the middle, you’re competing with software.

Client Sophistication. Your clients are using the same tools. They know what ChatGPT can do and what it costs. And they’re starting to ask a reasonable question: why am I paying $400 per hour for something I could prompt myself?

The Squeeze Is Already Here

The RFP You Used to Win. A mid market company needs a current state assessment and recommendations for their supply chain. Two years ago, this was a $300K engagement with a team of four over eight weeks. Today, a smart internal team with AI can get 70% of the way there in two weeks. They might need you for the last 30%, the judgment, the politics, the implementation credibility, but they’re not paying $300K for it.

The Staff Aug Trap. You’ve been placing contractors at $150 per hour for compliance work. The client just piloted an AI tool that handles 60% of the documentation. They don’t need five contractors anymore, they need two, plus someone who can manage the AI workflow. Your revenue just dropped 60% and your margin on the remaining work is worse because now you’re competing with every other firm that also has AI augmented delivery.

The Partner Who Can’t Explain the Value. A partner wins a deal based on relationships. The team delivers using AI augmented workflows and finishes three weeks early. The client is thrilled but now they’re asking why they budgeted for eight weeks when it only took five. The partner doesn’t have a good answer because the pricing was based on effort, not outcome.

Three Models That Actually Work

Here’s what I’m seeing and driving in my firm UST and others that have figured this out, including in regulated industries where moving fast and breaking things isn’t an option.

Model 1: Outcome Based Pricing

You price the engagement based on the value delivered, not the hours consumed. Define the outcome with precision, price to a percentage of the value created or the cost avoided, then use AI to compress your delivery cost while keeping the price anchored to outcomes.

Instead of billing $250K for a digital transformation roadmap, you price $400K for a board approved investment plan that secures budget allocation within 90 days. The deliverable is the decision, not the deck.

The catch is that you need confidence in your ability to deliver and you absorb risk. You also need clients willing to define outcomes clearly, which many aren’t. This works best for high stakes strategic work where the value is obvious and measurable.

Model 2: Subscription and Retainer

This is ongoing access to senior expertise at a fixed monthly rate, with AI handling the surge work. The client pays $20K to $50K per month for access to a senior advisor. AI handles research, synthesis, first drafts, and ongoing monitoring. You show up for the decisions, the politics, and the moments that matter.

A life sciences company retains you as a fractional transformation leader for $35K per month. You attend their leadership meetings, advise on major decisions, and deploy AI augmented analysis as needed. No big SOWs and no utilization math, just ongoing value.

The catch is you can only have three to five of these relationships before you’re overextended. It doesn’t scale like a traditional practice. This works best for mid market companies who need senior expertise but can’t justify Big 4 project fees.

Model 3: Productized Offerings

You package a repeatable methodology into a fixed scope, fixed price offering that AI makes efficient to deliver. Identify something you do repeatedly that follows a pattern, standardize the inputs, prompts, workflow, and outputs, then price it as a product.

The Prompt Pack from last week isn’t just a set of prompts. It’s the backbone of a productized engagement model: Intake Pack to Framer to Synthesis to Options to Red Team to Decision Memo. Same flow every time with predictable cost and predictable outcome.

The catch is you have to resist the urge to customize everything because scope creep kills the economics. This works best for practices with repeatable methodologies like compliance, implementation readiness, and operational assessments.

The Pricing Conversation You Need to Have

Here’s the question that separates firms that thrive from firms that slowly bleed: what are you actually selling?

If your answer is smart people by the hour, you’re in trouble. If your answer is that you compress time to decision on high stakes problems, you have a value proposition that AI enhances rather than threatens.

The firms I see winning have internalized a shift from selling hours to selling outcomes, from staffing for utilization to staffing for expertise density, from growing by adding bodies to growing by adding leverage, from competing on methodology to competing on judgment and speed, and from pricing based on effort to pricing based on value at risk.

What This Means for Your Career

If you’re junior, learn the AI workflows now. Become the person who can do in one day what used to take a team a week.

If you’re mid career, move toward client facing work, facilitation, and judgment calls. The middle tasks that built your skillset are being automated and your next decade depends on moving up the stack.

If you’re a partner or practice leader, rethink your pricing model before your clients do it for you. Pilot outcome based pricing on one engagement this quarter and see what breaks.

If you’re independent or considering it, the fractional model has never been more viable. Senior expertise plus AI leverage is a compelling package if you can sell it.

The Uncomfortable Math

The traditional model prices a four person team over eight weeks at a blended rate of $275 per hour. That’s roughly $400K in revenue.

An AI augmented engagement delivering the same outcome might need two consultants for five weeks. Same hourly rate, but now you’re looking at $110K in revenue. Your margin percentage improves because your cost base dropped, but your absolute profit is significantly worse.

You delivered the same value in less time with fewer people. That’s the trap.

The only way out is to break the link between hours and price. Charge for the outcome, the speed, and the certainty. Or watch your revenue compress while your competitors figure it out first.

What’s Next

Next week, I’ll share a story from my first consulting engagement, where every Friday afternoon leadership gathered in a conference room to decide which consultants to fire that week. For twelve straight weeks, they cut someone from our team. That experience taught me something about discipline and preparation that still shapes how I work today, and it explains why effort counts twice in Duckworth’s grit equation, especially now that AI has transformed what “preparation” actually means.

Reply and tell me: which pricing model feels most relevant to your situation right now?

— Chris

t

Share

Share The Consulting Afterlife

If you’ve experimented with AI in consulting, you’ve probably had the same mixed experience everyone has:

• Sometimes it’s shockingly good.

• Sometimes it’s confidently wrong.

• And most of the time the difference isn’t the model—it’s the inputs, the prompt, and whether you have a repeatable workflow.

The firms that win won’t be the ones with the best “AI enthusiasts.” They’ll be the ones who turn AI into an operating system—a consistent way of working that produces speed and reliability.

That starts with a Prompt Pack.

Not 200 prompts. Not a library nobody uses. A small set of repeatable prompts that map to the phases of an engagement, and plug directly into your delivery chain.

Below are the 7 prompts I’ve found most useful. They’re designed to support the AI Leverage Map from last week:

• automate what’s low context + low consequence

• accelerate what’s low context + high consequence (with verification)

• augment what’s high context + low consequence

• assure what’s high context + high consequence

Use these as templates. Save them. Standardize them. Train your team on them. The leverage comes from consistency.

Before you start: the Intake Pack (don’t skip this)

AI doesn’t run on inspiration. It runs on inputs.

For any engagement, assemble a simple Intake Pack and paste it above every prompt:

• Objective: What outcome are we driving?

• Decisions needed: What must be decided, by when, by whom?

• Constraints: Timeline, budget, regulation, capacity, politics.

• Context: Industry, org structure, current state, known issues.

• Artifacts: Notes, docs, systems info, agendas, prior decks.

• Stakeholders: Names/roles + what they care about.

This is the difference between “magic” and “garbage.”

The 7 Prompt Pack

1) The Engagement Framer (Day 0)

Use this to turn a fuzzy ask into a crisp engagement definition.

Prompt:
“Using the Intake Pack below, create:

1. a one-paragraph problem statement,

2. the top 5 outcomes (measurable),

3. the top 7 assumptions we are making,

4. the top 10 questions we must answer,

5. and a draft scope boundary: what is in / what is explicitly out.
   Also list the 5 failure modes that most commonly derail engagements like this.”

Why it matters: It gets you to clarity fast—and surfaces “hidden landmines” early.

2) The Stakeholder Reality Check (Early discovery)

This is for the politics you can’t put on a slide.

Prompt:
“Based on the stakeholders listed below and the context provided, infer:

• likely incentives and fears for each stakeholder,

• where alignment is most fragile,

• where we should expect passive resistance,

• and what each stakeholder needs to hear to support the decision.
  Output a one-page ‘alignment plan’ with recommended 1:1s, messages, and sequencing.”

Why it matters: AI can’t know the politics, but it can help you think through second-order effects.

3) The Workshop Designer (When you need alignment)

Use this whenever you’re about to faciliate something that matters.

Prompt:
“Design a 90-minute workshop to drive these decisions: [list].
Include: agenda with timings, questions to ask, breakout prompts, decision rules, and a facilitation script.
Add a ‘derailment plan’: what to do if participants are stuck, defensive, or arguing.
End with a decision capture format we can paste into meeting notes.”

Why it matters: Great workshops aren’t meetings—they’re decision machines.

4) The Synthesis Engine (Mid-engagement)

Turn messy notes into clean insights without inventing facts.

Prompt:
“Here are raw notes/artifacts from interviews/workshops.
First: summarize only what is explicitly stated (no assumptions).
Second: extract themes and tensions, each with supporting quotes or references to the notes.
Third: list ‘unknowns’ and the minimum data needed to resolve each.
Finally: propose 3 narrative storylines for executives, each with a different framing.”

Why it matters: This is how you move from noise → signal safely.

5) The Option Generator + Tradeoff Table (When choices matter)

AI is excellent at generating options. Your job is making them real.

Prompt:
“Generate 3–5 viable options for achieving the outcomes given the constraints.
For each option, provide:

• what it prioritizes, what it sacrifices

• risks (execution, organizational, regulatory)

• prerequisites (capabilities, decisions, investments)

• expected timeline and key milestones

• ‘how this fails’ scenarios
  Then output a tradeoff table that an exec can understand in 60 seconds.”

Why it matters: Clients don’t need more ideas. They need structured tradeoffs.

6) The Red Team (Before you present anything important)

If you only add one prompt to your workflow, make it this one.

Prompt:
“Act as a skeptical client executive and try to tear this recommendation apart.
Identify:

• where the reasoning is weak or missing data

• likely objections by stakeholder type (CFO, CIO, QA/Regulatory, Ops leader)

• hidden assumptions

• what would make this fail in implementation
  Then propose modifications to strengthen the plan, and list the top 10 questions we should be ready to answer.”

Why it matters: This catches “confident wrongness” before it hits the client.

7) The Executive Decision Memo (The deliverable that replaces the deck)

If you want to move up the stack, shift from decks to decision memos.

Prompt:
“Write a 1–2 page executive decision memo with:

• Decision requested (clear)

• Context (brief)

• Options considered (short)

• Recommended option + rationale in plain language

• Tradeoffs and risks (explicit)

• What has to be true for success

• Next 30/60/90 day plan

• Reversible vs irreversible elements
  Use a tone that is direct and senior. No consultant fluff.”

Why it matters: Decision memos create speed, alignment, and accountability.

The Rule That Keeps This From Becoming Chaos

Here’s the operating principle:

AI drafts. Humans assure.

So bake in two checkpoints:

1. Source check: What inputs support each major claim?

2. Reality check: What breaks when this hits real execution?

If you can’t answer those two things, you don’t have a recommendation—you have a draft.

How to deploy this in your team (in one week)

If you’re leading a practice or delivery team, do this:

• Pick one live engagement.

• Mandate the Intake Pack.

• Standardize these 7 prompts.

• Require Red Team before every exec readout.

• Replace at least one slide-heavy deck with a decision memo.

• Measure cycle time: discovery → recommendation → decision.

You’ll feel the leverage immediately.

And you’ll also learn where AI creates risk in your environment—which is exactly where your human differentiation gets stronger.

What’s Next

Next Tuesday, I’ll share the piece most people avoid because it’s uncomfortable:

The New Economics of Consulting—how AI changes pricing power, why time-and-materials gets squeezed, and the three packaging models I see working (even in regulated industries).

If you want to reply: which of these prompts would create the biggest immediate leverage for your work right now—Framer, Red Team, or Decision Memo?

— Chris

P.S. If you’re thinking, “This is great, but my client won’t let us use AI,” you can still use the Prompt Pack internally on sanitized inputs, then validate with your standard governance. The point isn’t the tool. The point is the workflow.