Category: The Argument

For most of my career, the systems I built were valuable because they were hard to build.

At research labs and then inside large enterprises, my teams and I built machine learning systems that took months of work, rare expertise, real budgets, and long struggles for resources. The difficulty was not incidental to the value. It was the value. If a competitor wanted the same capability, they needed the same scarce people, the same time, and the same money. Scarcity of execution was the moat. We just never had to call it that, because it had always been true.

Looking back now, I feel two things at once. Pride, because the work was genuinely good. And vertigo, because a lot of it could be stood up today in a weekend, by a small team with a subscription. The capability I spent years of my life building is being commoditized.

Here is the part that took me longer to see. The part of that work that mattered most was never the part that was hard to build. It was knowing what to build, what the data actually meant, which requests to push back on, and which impressive system would quietly make things worse. That part has not commoditized. That part is what this piece is about, because the world’s highest-grossing law firm recently bet half a billion dollars on it.

The sentence every boardroom should read

Kirkland & Ellis, which booked $10.6 billion in revenue in 2025, recently announced it would spend roughly $500 million over the next few years building its own AI platform rather than relying only on the tools its competitors can buy. Its chairman, Jon Ballis, compressed the logic into one line. Widely available AI tools, he said, are “raising the floor for everyone.” But, he added, “we don’t get hired for the floor.”

That is one of the most clarifying things a business leader has said about AI strategy in two years, and it cuts against how most companies are spending. The prevailing assumption is that advantage comes from having the most capable AI. Buy the best models, deploy the most agents, automate fastest, and you win. This was a reasonable playbook for almost every previous technology. It is wrong about this one.

We have run this experiment before

It is wrong because capability is commoditizing, and we know what happens when capability commoditizes, because it has happened to every general-purpose technology in modern history. Electricity was an advantage for the firms that could generate it, until it became a utility and the advantage migrated to what firms did with it. Computing was an advantage for firms with mainframe access, until computing became accessible and the advantage migrated to data and process design.

Nicholas Carr made himself famous, and briefly infamous, by calling this pattern in 2003. His Harvard Business Review essay “IT Doesn’t Matter” argued that information technology was following electricity and the railroads from proprietary advantage into shared infrastructure. The argument was bitterly contested at the time. Two decades later it looks prescient. The firms that built durable advantage in the computing era were not the ones with better hardware. They were the ones that developed distinctive organizational capabilities for using what everyone could buy.

Michael Porter gave us the vocabulary for why. Operational effectiveness, doing the same things better, is necessary but not sufficient, because best practices diffuse. Strategy is doing things rivals cannot easily match. Tools that a thousand firms can license are operational effectiveness by definition. They raise everyone’s floor at once, and a tool that raises every floor confers advantage on none of them. That the operating model is the moat, more than the model, is a case I made in detail recently, building on McKinsey’s own evidence. If your AI is the same as the AI across the negotiating table, you have spent money to keep pace, not to pull ahead. Ballis’s floor and ceiling is Carr’s argument and Porter’s distinction, restated by a customer with $500 million on the table.

Even the people selling the technology concede the mechanism. Larry Ellison, whose company is staking its future on enterprise AI, argues that because every major model trains on the same public internet, model outputs are converging and differentiation is eroding. He is right. Shared inputs produce shared reasoning. The implication the vendors are less eager to draw is that buying more of a commoditizing capability is not a strategy. It is a subscription.

Where Ellison’s answer falls short

Ellison’s proposed moat is proprietary data. That is closer, but it imports a mistake. Every enterprise already has data, most of it fragmented across systems, contradictory between departments, disconnected from the outcomes it produced, and ungoverned. Pour that into a powerful reasoning engine and you do not get insight. You get fast, confident reasoning over an incoherent picture, which is worse than slow reasoning, because the confidence hides the incoherence. Having data is not scarce.

What is scarce is data made coherent: owned, reconciled across the organization, connected to the outcomes it produced, and trustworthy enough to reason over, joined to the judgment of the people who know what the numbers mean. I argued earlier in this series why no vendor can sell you this. The platform that stores and retrieves your data is plumbing, and plumbing commoditizes. The coherence is the asset, and it compounds.

What Kirkland is actually buying

Read the Kirkland decision through that lens and it stops looking like a technology splurge and starts looking like strategy. The firm is not paying $500 million for data it already owns or software it could license. Anyone can download its public filings. What cannot be downloaded is how the firm decides: the judgment of 250 of its lawyers, 100 of them partners, encoded in a form every lawyer can draw on for every matter. Kirkland is spending to make its institutional judgment coherent, and to keep it exclusive.

The tell is in the terms. The outside firms building the platform are barred from selling it to any other law firm. If the value were the technology, exclusivity would not matter. Kirkland insists on it because it believes the technology is commoditized and the value is the distinctive judgment the system encodes. A shared tool would dissolve exactly that. A shared tool is also a conduit. Every standard and correction a firm feeds into it can improve the version its rivals rent tomorrow, which is the leakage I traced in Your AI Usage Exhaust Is Someone Else’s Moat. Read that way, Kirkland’s exclusivity clause is that essay’s prescription written into a contract: close the loop, and keep what you encode inside your own walls.

Since the May announcement, the pattern has only hardened. Through June, Kirkland added two more exclusive builds, one for private-equity fund formation and one for litigation, and framed each the way it framed the platform itself: a way to capture the firm’s own judgment and knowledge and keep it exclusive to Kirkland. Three deals in roughly five weeks, and the constant across all of them is the insistence on owning what the tools encode.

None of this means Kirkland is certain to be right. Building rather than buying is a real bet, and it is possible that within a few years a purchasable platform, fed a firm’s own data and tuned to its standards, delivers most of the advantage at a fraction of the cost. Every executive now faces a version of that question. But notice what the question is actually about. It is not whether to have AI. It is what you are trying to own. Kirkland has decided the thing worth owning is not the model and not the data but the coherence that turns both into judgment competitors cannot replicate.

The wrong scoreboard

The firms that misread this will keep score by the wrong numbers. They will count agents deployed and measure speed of adoption, and they will mistake a rising floor for a rising position. I understand the pull of that scoreboard better than most, because I spent years on the other side of it, building the hard things the scoreboard rewarded. The hard things are cheap now.

The firms that read it correctly will ask a harder question: when the models are a commodity and the data is everywhere, what does our organization understand about itself that no competitor can buy?

That was never the intelligence. It is the coherence of the organization putting intelligence to work. Kirkland just put half a billion dollars behind that proposition. The rest of the market is still buying the floor.

This argument runs through my book, Coherence, arriving this Fall. If you want to follow the thinking as it develops, join the list at coherise.com. The one-page decision tool from the book is the first thing I send.

McKinsey published a piece this month that gets the hardest part right. Its argument, in one line: the advantage in AI is not the tools, it is the operating model, and the operating model is the one thing a competitor cannot buy. I agree with almost all of it. I want to push on the part it leaves out, because that part is where most companies are about to get hurt.

Start with what McKinsey gets right, because it is a lot. Efficiency gains from AI, they argue, will become table stakes as the technology spreads. Operating models, unlike software, cannot be purchased or copied overnight, so the durable moat is the organization, not the model. They have the numbers to go with it. Only about a fifth of companies have fundamentally redesigned how they work around AI. Top performers are three times more likely to have done that redesign, and twice as likely to redesign the workflow before choosing the tool. And AI programs run as technology projects fail at more than an eighty percent rate, because they optimize the tool instead of changing how the company works.

If you have read anything I have written, you know why I would agree. This is the commoditization argument. Capability is becoming universal, so capability stops being the edge, and the advantage moves to what the organization can do that a competitor cannot copy. McKinsey and I are looking at the same shift.

Here is where we part.

The scissors cut both ways

McKinsey’s best image is what they call the complexity scissors. As a company grows, revenue grows not on a line but a curve – fast initially and flattening later. But coordination costs, the meetings and committees and management layers, keep climbing. Plot the two lines and they open like a pair of scissors. The gap between them is why big companies slow down, and why fewer than one in ten sustain returns above their cost of capital over a decade.

Their prescription is to use AI to close the scissors. Route decisions through a central orchestration layer. Hand coordination work to agents. Flatten the org. Get faster.

This is the step I want to stop on. AI can close the scissors. It can also open them wider, and nothing in the redesign itself tells you which one you are going to get.

McKinsey is right that agents let you route around the old coordination layer. What they underplay is that agents build a new coordination surface underneath, invisible, machine-speed, and owned by no one.

Every autonomous system you add to speed up a workflow is also a new thing that has to stay consistent with every other autonomous system. A support agent and a billing agent that make different assumptions about the same customer have not reduced coordination cost. They have created a new kind of it, one that does not show up in a meeting because no human is in the loop to notice. McKinsey is right that agents let you route around the old coordination layer. What they underplay is that agents build a new coordination surface underneath, invisible, machine-speed, and owned by no one. Their own report admits the danger in a single line: when agents run inside workflows that were not redesigned for them, errors propagate across the company at machine speed. That is the coordination trap, and it is produced by the very rewiring they recommend.

So the redesign is not the safe move and the caution. The redesign is the risk. Done with the discipline to keep the new systems coherent, it closes the scissors. Done as a race to orchestrate and flatten, it opens them, and it does so faster than the old human version ever could, because now the coordination failures happen at the speed of software.

This is not just my read of the mechanism. IBM’s 2026 study of two thousand CIOs and CTOs found the same trap from the inside. Companies that chase speed let business units move ahead while governance falls behind, gaining local velocity and losing containment. Companies that chase safety slow deployment under review until oversight becomes unmanageable. Both paths, in the study’s own words, accumulate strategic debt. That is the point. The rewiring does not have a safe default. It has two ways to fail and one narrow way to work, and the narrow way runs through coherence.

The rewiring does not have a safe default. It has two ways to fail and one narrow way to work, and the narrow way runs through coherence.

Their own examples make the point

Look closely at the cases McKinsey uses, because they prove the thing the article does not quite say out loud.

The copper miner they profile did not win by deploying more AI. It won by building modular models where roughly sixty percent of the code from the first site was reusable across the next six, so each deployment got faster and cleaner than the last. That is a coherence story wearing a productivity headline. The reuse is only possible because someone designed the systems to fit together before scaling them. The car maker they profile shrank a planning team by more than eighty percent, but the win was not the headcount. It was that the coordination layers between data and decision compressed, because the workflow was redesigned as one coherent thing instead of a chain of handoffs.

In both cases the value came from making the systems cohere, not from the number of systems shipped. McKinsey files this under operating-model redesign. I would file it more precisely: the redesign worked because it was coherent, and it would have failed if it were not. The article treats coherence as a happy property of good redesign. I think it is the whole variable, and that a redesign optimized for speed without it produces the opposite result in the same enterprise.

What to actually do differently

If you take McKinsey’s advice and only McKinsey’s advice, you will redesign for speed and measure yourself on how fast you moved. That is the eighty-percent-failure path wearing better clothes, because deployment speed is exactly the vanity metric that hides the debt building underneath.

The addition is small to state and hard to do. Before you rewire a workflow around agents, decide how those agents will stay consistent with the rest of the company as they multiply. Build the ability to see what your autonomous systems are doing in aggregate, contain them so a failure in one stays in one, and set in advance how much each is allowed to decide. Then measure the redesign not by how fast it shipped but by whether the organization got more coherent or less as it grew. A team that retired four brittle systems and shipped nothing new may have improved your position more than the team that shipped fourteen agents into the trap.

McKinsey is right that the operating model is the moat, and right that most companies are getting this wrong by treating AI as a tool to buy rather than a business to redesign. The correction I would add is that the redesign has a failure mode of its own, and it is the one nobody is watching for. The winners will not be the companies that rewire fastest. They will be the ones that rewire coherently, which is a slower thing to say and a harder thing to build, and the only version that closes the scissors instead of opening them.

The winners will not be the companies that rewire fastest. They will be the ones that rewire coherently.

That is the subject of my book, Coherence, and of everything I am writing here between now and launch. If the rewiring is on your desk right now, the one-page tool I use to sort what to automate, what to redesign, and what to leave alone is the first thing I send when you join the list at coherise.com.

In May, Meta laid off roughly 8,000 people, about ten percent of the company, in a restructuring built around AI. The memo called the cuts the price of leading the most consequential technology shift of our lifetimes. Six weeks later, at an internal town hall on July 2, Mark Zuckerberg told employees that AI agent development “hasn’t really accelerated in the way that we expected.” The reorganization had been messier than planned, he said, and the benefits should arrive in three to six months.

Be precise about what he conceded. He did not say the layoffs were a mistake. His chief AI officer quickly clarified that he meant the whole industry’s progress on agents, not Meta’s alone. Take that at face value. It makes the admission bigger, not smaller. A whole industry restructured its workforce around an acceleration that one of its most aggressive adopters now says is running late. Meta may just be the first to say so out loud.

Every story like this feeds the same debate, and two chief executives sit at its poles. In May, Matthew Prince of Cloudflare wrote in the Wall Street Journal about how he decides which employees to replace with AI. He had just cut more than a fifth of his workforce. Days later, in the New York Times, Goldman Sachs chief David Solomon took the optimist’s side. Yes, AI will disrupt the labor market, he wrote, but America absorbed electrification and the digital revolution before it, and new work will emerge as it always has. One CEO sees replacement beginning. The other sees the economy adapting, as it always has. Both may be right. Both are missing the bigger problem.

AI is not just a labor story

The debate treats AI as a labor substitute. But what AI is really collapsing is the cost of execution, the cost of doing things. Build a workflow, analyze a dataset, draft a document, run a process. Each of these once took real expertise and real time. Now each one takes a goal and an instruction.

When execution gets cheap, the constraint moves. What becomes scarce is not the ability to do things. It is the ability to do them coherently. A company now runs on hundreds of autonomous systems. Keeping them pointed at compatible goals is the hard part. So is making sure the people who answer for the results still understand what they have built well enough to govern it.

That is the inversion neither side of the jobs debate has named. AI is not primarily a labor story. It is a coordination story.

And the people living it already feel the difference. In a 2026 survey of 1,200 executives, more than half, 54 percent, said adopting AI was tearing their company apart. In the same survey, 79 percent said AI applications were being built in silos, and more than a third admitted they could not immediately shut down a misbehaving agent. Those are not complaints about weak models. They are the sound of an organization losing its coordination, not its labor.

Prince’s own reasoning shows why. He built the Cloudflare cuts on a framework from Peter Drucker: every company has builders, sellers, and measurers. AI can now measure cheaply and continuously, work that used to take a lot of people. So the measuring layer can shrink, and the savings can move to the roles that create value. The logic is clean. That is what makes it worth examining, because it rests on one quiet assumption. It assumes the people labeled measurers were only measuring.

In most companies they were not. The person who tracks a process is often the one who notices when it breaks. She knows why it was built that way. She catches the exception the dashboard misses. Cut her as redundant measurement, and you may find you also cut a layer of coordination you never had a name for. Did Cloudflare make that mistake? I cannot say from outside. The point is that the framework cannot even see the question. Neither can the jobs debate it belongs to.

Anticipation, not implementation

The labor story is mostly theater anyway, and the evidence says so. In a survey of more than a thousand executives, 21 percent had made big cuts in anticipation of AI. Only 2 percent tied cuts to AI they had actually deployed. That is a tenfold gap between the future they were cutting for and the present they were living in. New York State added a box to its mass-layoff filings asking whether automation drove the cuts. In the first full year, almost no employer checked it.

The research points the same way. AI’s real effect on jobs shows up in slower hiring, not firing. And that is the quiet danger. According to Accenture’s last Pulse of Change survey, 59% believe young professionals are having a harder time finding jobs due to automation and AI. Juniors who never get hired are the seniors an organization will lack in ten years. The pipeline gets cut without anyone announcing it.

The layoffs are running ahead of the plans that would justify them. In that same 2026 survey, 69 percent of companies were planning AI-related layoffs, yet 39 percent had no formal strategy to earn revenue from the AI they were adopting. Cutting first and figuring out the value later is not a strategy. It is a bet on an acceleration that has not shown up.

Read Meta’s year through that lens. The company cut in anticipation of an acceleration. The acceleration is the very thing its CEO now says has not arrived on schedule.

The oldest instinct in the room

I have seen the underlying mistake before, long before agents existed. Early in the data science era, I sat down with a business unit head to discuss where machine learning could help. Her opening move was to hand me an enormous dataset and ask me to figure out something useful from it. I was the data geek. That was my job.

Capability first, purpose later, is analysis unmoored from the business, and it goes nowhere.

It took several more conversations before the team came around to a different starting point: begin with a business outcome, then work backward to the analysis. Capability first, purpose later, is analysis unmoored from the business, and it goes nowhere.

That instinct never went away. What changed is the friction that used to hold it back. Back then, a capability-first project cost a quarter of work and a real budget. When it produced nothing, it died quietly in a slide deck. Today the same instinct ships an autonomous system into production in an afternoon.

And those systems do not sit still. An agent built for escalations spins up sub-agents to handle edge cases, each with its own logic and even less context than the first. A finance agent reaches for new data sources and builds a picture of the company no human has checked. Every step is reasonable on its own. Nobody approved the whole. The organization did not design this. It just failed to prevent it.

This is how the burden I wrote about last week, complexity debt, actually accumulates: not through failure, but through hundreds of local successes that nobody is coordinating. The old friction was never just cost. It was an accidental governance system. AI removed it without replacing it.

Why the reorganization wasn’t clean

There is also a well-documented reason Zuckerberg’s three-to-six-month promise should be read skeptically, and it has nothing to do with model quality.

Economists Erik Brynjolfsson, Daniel Rock, and Chad Syverson found that big general-purpose technologies follow a productivity J-curve. Early on, measured productivity actually dips. The technology demands a lot of hidden investment first: new processes, new roles, new organizational wiring. That work is real, but it does not show up on the books. The gains come later, once the work is done. It happened with electricity. It happened with computing.

A plan that waits for the models to improve, instead of doing the coherence work, is a plan to sit at the bottom of the J.

For agentic AI, that hidden investment is coherence. It means shared context across systems, wiring the organization can still read, and a clear human owner for every autonomous system. This is organizational work. No model release does it for you. A plan that waits for the models to improve, instead of doing the work, is a plan to sit at the bottom of the J.

This makes the shape of Meta’s cuts worth a pause. Reporting at the time said the layoffs fell hardest on integrity, cybersecurity, and content design, while AI infrastructure and monetization teams were spared. I do not know Meta from the inside, and I will not pretend to. But the pattern raises the question every executive should ask before this trade: how much of what looks like overhead is actually the coordination layer? Cut the people who do the quiet coordinating work, then multiply the systems that need coordinating, and the J-curve does not get shorter. It gets deeper.

What actually closes the gap

History suggests where this leads. Industrialization created coordination problems, and operations management grew up to solve them. Software created questions no engineer or salesperson owned, and product management, data science, and UX design each emerged to answer one. The pattern repeats every time. The new function is dismissed as redundant, then treated as essential, then made table stakes once the companies that built it start winning. The doubters never look wrong until the results come in.

Even the optimists can see the first outline of it. Solomon, arguing that new jobs will emerge, names one. Companies are already hiring people to manage agentic AI, he writes, across implementation, workflows, compliance, and validation, and all of it takes human judgment. He is right. The role is real and growing. Some in the industry are calling it the agent manager, the person who runs a fleet of agents inside one function. Tellingly, the good ones tend to come from the business being automated, not from tech. Forward deployed engineers, brought in to install and tune the systems, are the other half of the early response.

Both are real. Both are also operators, and this is where Solomon stops one step short. An agent manager’s view ends at the edge of the fleet. The support agent manager watches the support agents. The procurement agent manager watches the procurement agents. Each starts and ends the day inside their own dashboard. Nobody is accountable for whether those fleets are working from the same assumptions about the same customer.

An enterprise can staff every function with an excellent agent manager and still have no one whose job is the coherence among them.

That cross-cutting question is invisible from inside any single fleet, and it is the one that goes unowned. An enterprise can staff every function with an excellent agent manager and still have no one whose job is the coherence among them. This is already biting. In a 2026 survey of 621 enterprise leaders, 42 percent said the lack of a clear internal owner had directly delayed an agentic project in the past year. The gap is not theoretical. It is on the calendar, slowing things down right now.

The obvious response is to add a governance layer: a committee that reviews what the machines produce, a stack of approvals, more people whose job is to sign off. That instinct is backward.

Be careful here, because the obvious response is the wrong one. The obvious response is to add a governance layer: a committee that reviews what the machines produce, a stack of approvals, more people whose job is to sign off. That instinct is backward. It is the old apparatus of permission applied to a problem it was never built for, and it cannot keep pace with systems that deploy in an afternoon.

The goal is to build coherence into how the organization is wired, not to post humans at every intersection to hold it together by hand.

The book argues the real work is structural, and it comes first. Before any new role, an organization has to build the infrastructure to see its own systems, to contain them so trouble in one place stays in one place, and to set in advance how much any system is allowed to decide. Most of that is design work, done once and maintained, not review repeated forever. Increasingly the systems do the watching themselves. The goal is to build coherence into how the organization is wired, not to post humans at every intersection to hold it together by hand. An organization that stays coherent through sheer vigilance is coherent only for now, at a cost that does not scale, one lapse of attention from a mess.

Only on top of that structural work does a human role make sense, and it is a smaller thing than a new bureaucracy. It resolves into a handful of capabilities, each one rhyming with a profession we already know. Someone has to decide how the organization’s own systems should fit together, which ones may depend on which, and where the lines between them run. That is product management, turned to face inward. Someone has to establish which systems can be trusted, for what, on your actual work, based on real, measured data. That is what an analytics team does, applied to AI. Someone has to design the moments where a person hands a decision to a system and takes it back. That is akin to UX, pointed at the inside of the company, designing the interfaces between human and machine. And someone has to own all of this close to where the systems are built, while still answering to a view of the whole. That is the embedded finance officer, in a new setting.

In each of these the machines do more of the work every year. What does not pass to them is the judgment about what the organization should permit, and who answers when it goes wrong.

That is the real shift. The job is not to build AI, and it is not to approve it. It is to build the structure that lets an organization gain from what AI does for it, and hold together while it does.

The question that matters

The jobs debate will continue, and it should. Displacement is a genuine human and economic concern. But the debate is incomplete. It asks what AI replaces. The more urgent question is how to manage the complexity AI creates, and whether organizations will recognize that need before the debt comes due.

Zuckerberg has put a clock on Meta’s answer: three to six months. I read that clock differently. It is not set by the next model release. It is set by how fast a company can do the unglamorous work of staying coherent while it automates. That work is what my book, Coherence, is about, and I will keep working through it here in the open. The one-page tool I use to start is the first thing I send when you join the list at coherise.com.

Early in my career, at Mitsubishi Electric Research Labs, I watched a team solve a hard problem in a way that has stuck with me ever since.

The task was automatic highlight reels for baseball games. Real effort had gone into the sophisticated version: systems that could read the play, follow the ball, understand the game. Then someone on the audio side noticed something. Every moment worth keeping had one thing in common. The crowd roared. So they tried the simplest possible approach. Replay the few seconds around each spike in crowd noise. It produced a near-perfect highlight reel, built from a signal anyone could have used.

The sophisticated system was not the advantage. Noticing what mattered was.

I think about that a lot right now, because the enterprise is making the same mistake at scale. We have decided the advantage in AI is capability. The smartest model. The most copilots. The biggest deployment. It is not. And capability is about to stop being scarce at all.

For most of the past decade, the winning move was speed. Adopt faster, automate more, ship before the competition. That worked because building things was hard, and whoever removed that friction first pulled ahead. Agentic AI is ending that era, because it is making execution cheap. When anyone can build, automate, and deploy in an afternoon, speed stops being an edge. Everyone has it.

So what becomes scarce? Coherence. Whether your growing crowd of autonomous systems, and the people accountable for them, still pull in the same direction. When everyone can go fast, coherence is what wins.

The debt that never shows up on a dashboard

Here is what happens when execution gets cheap and no one is watching for this. Everyone makes more. Sales stands up an agent for lead qualification. Finance automates forecasting. HR wires up hiring. Operations builds a copilot for logistics. Each one works. Each one passes its own tests. On paper, the company gets more automated every month.

And it gets quietly harder to run. Decisions start flowing through systems no single person fully understands. Two agents act on assumptions that contradict each other. Every team sharpens its own corner while the whole thing loses its shape. I call this complexity debt, and its defining feature is that you cannot see it directly. You feel it later, as the strange sense that nobody can quite explain why the organization behaves the way it does.

I learned how this happens the embarrassing way, long before agents existed. At United Technologies, I built a data visualization I was proud of. It was dense, information-rich, technically elegant, the kind of thing that impresses other engineers. But users hated it. They found it unreadable, and they were right. I had optimized for the wrong thing. My system was correct at the level I cared about and useless at the level that mattered.

That gap is the whole problem, and it is about to repeat across the enterprise, one agent at a time. A system can be flawless at its task and still make the organization worse. Task-level correctness and organizational coherence are different things. Improving one does nothing for the other, and most companies are measuring only the first.

What coherence is

So what am I asking you to protect? Coherence is not a vibe or a culture slogan. It is structural, and you lose it in specific, recognizable ways.

A coherent organization shares context. Its systems and its people reason from the same facts, so a decision in one place does not silently undercut a decision somewhere else. It keeps its wiring legible, so pulling out one system does not trigger a chain reaction nobody saw coming. And it keeps a human line of accountability for every autonomous system: someone who owns it, knows what it is really doing, and can correct it when it drifts.

None of that means slowing down, and none of it means centralizing. A decentralized company can be perfectly coherent. A centralized one can be a mess. Coherence is not the absence of autonomy. It is what makes autonomy safe to scale.

You cannot supervise your way out of this

The instinct, once a leader feels this, is to watch everything. That instinct fails on contact. You cannot supervise a hundred agents by paying attention harder.

The leaders who handle this well build coherence into the structure instead. They set constraints so whole classes of incoherence cannot arise in the first place. They contain systems so the failures that do occur stay local rather than spreading. They spend their scarce human judgment on the few decisions that truly need it, and let the rest run inside guardrails. It is engineering, not vigilance. The difference is between a company that stays coherent because someone is always watching and one that stays coherent because it was built to.

The advantage no vendor can sell you

This is why I keep coming back to that baseball reel. Intelligence is commoditizing. Everyone will have capable models, mostly the same ones, at mostly the same price. The capability will not be your advantage, any more than the sophisticated summarizer was.

What will not commoditize is the ability to deploy all that intelligence coherently: the judgment to decline the automation that buys a local win at the cost of the whole, the architecture that lets you move aggressively without piling up debt, the discipline to keep the organization legible to itself as it fills with autonomous systems. That is hard, it is specific to you, and there is no vendor who can sell it to you.

So here is the question I would sit with. You can almost certainly tell me, right now, how accurate your models are and maybe, how many agents you have in production. But can you tell me whether your organization is still coherent? Do you have anything that would show you coherence breaking before it breaks something?

Most leaders do not. It is the most expensive blind spot in enterprise AI, and almost no one is looking at it.

That question is what my book, Coherence, is about, and it is what I will be working through here in the open over the coming months. The one-page tool I use to start answering it is the first thing I send when you join the list at coherise.com.

For thirty years, the fastest company usually won. I think that is about to stop being true.

Agentic AI is making execution cheap. When anyone can build, automate, and deploy in an afternoon, speed stops being an edge, because everyone has it. What becomes scarce is coherence: getting the growing crowd of autonomous systems, and the people responsible for them, to pull in the same direction. When everyone can go fast, coherence is what wins.

I did not arrive at that from a strategy deck. I arrived at it from a question I have been chasing since college.

For as long as I can remember, one thing has held my attention above all else: what makes something intelligent. That question pulled me out of discrete math and computer science and into computational neuroscience, where I spent years on a single puzzle. How does the brain turn billions of small, unreliable, independent units into one coherent mind? No neuron is in charge. No neuron can see the whole. And yet, somehow, thought happens. I did not know it at the time, but that puzzle would turn out to be the through-line of my whole career, and eventually, of this book.

I carried the question into industry back when there was no name yet for the job I was doing. I lived through the Big Data years, and then the Data Science years, each one arriving as the answer to everything and each one settling into something quieter and more useful. Then deep learning arrived, and it felt different in kind. The first time I saw what it could really do, I understood that the ground had moved under all of us. I joined Karthik to start Concentric AI on the strength of that conviction, and bet a company on it.

What I had watched build slowly for years then began moving at a pace I had never witnessed. Steady theoretical progress, faster hardware, and ever more compute compounded into generative AI. And now we are somewhere newer still, in the agentic era, where AI no longer just answers. It acts.

Which brings me back to the thing that unsettles me. In narrow domains like math and code, the progress is genuinely breathtaking. And yet inside most enterprises, that progress stubbornly refuses to turn into results. In conversation after conversation, what I hear from leaders is relentless pressure and a quiet fear of falling behind, all of it wrapped around a phrase nobody can quite define: “doing AI.”

Around that fear, the noise is deafening. On one side, breathless promises: networks of thousands of agents that will soon run entire departments on their own. On the other, a shrug dressed up as wisdom: the models will keep getting better, so whatever problem you see today will simply solve itself. Both cannot be true. And what is missing from all of it is not another confident prediction. It is a way to think clearly while everyone around you is loud and certain.

The serious thinking has mostly been elsewhere, and for understandable reasons. The researchers who build these systems are busy making the models, the theory, and the algorithms better. The engineers and tinkerers around them are focused on empirical evaluation, on benchmarks, and on getting individual systems to work. AI safety and responsible AI have concentrated on society and the long horizon, on where powerful models might eventually take us. Governance has approached AI through risk, security, and compliance. Economists and organizational theorists are only beginning to engage, and agentic AI is so new that there is little evidence yet to build rigorous work on. None of that is a failure. It just means the question I kept running into, how a company holds together as it fills with autonomous systems, has fallen into the gap between all of them, still largely unclaimed.

There is only one way I know to cut through noise like that. You stop arguing at the surface and go back to first principles. You find the single thing that actually changed, and you follow it, patiently, wherever it leads, whether or not the destination is fashionable.

The single thing that changed is the cost of execution. It has collapsed toward zero. So I started there and followed it, step by step: to why firms exist at all, to what becomes scarce once doing things is nearly free, and to where advantage has to move next. The word I kept landing on was coherence. The same word from the brain, now describing the enterprise. I began writing to pin it down. Before long, I had a book.

So I am glad to share that my book arrives this Fall. It is called Coherence: The Competitive Advantage AI Can’t Buy. Its argument, reasoned from the ground up, is simple to state and uncomfortable to live by: capability is now for sale to everyone, so it can no longer be your edge. What cannot be bought, and what now decides who wins, is the coherence of the organization putting AI to work.

Between now and launch, I will be writing here regularly, thinking out loud through the ideas in it: why speed stopped winning, the hidden cost of making everything autonomous, and what a leader can actually do about it on a Monday morning. If any of this matches what you are seeing from where you sit, I would be glad to have you along.

You can read more, and sign up for launch-day access, at coherise.com. Everyone who joins also gets the one-page decision tool from the book, for sorting what to automate, what to augment, and what to keep in human hands.

This book cost me more weekends, and more self-doubt, than I bargained for. What kept me going was a small surprise that never quite wore off: the question I once asked about billions of neurons turned out to be the same one facing every leader now trying to hold a company together. I am glad it is nearly here.