“A king who concentrates all craft within his court will find his kingdom brittle; a king who enables the craftsmen to practise freely will find his kingdom resilient.”
— Kautilya, Arthashastra (~300 BCE)
Kautilya was not making a moral argument. He ran the Maurya Empire’s statecraft the way a structural engineer runs load calculations — what holds, what breaks, and under what conditions. His point about craftsmen was an observation about brittleness: concentrate all the capability inside the palace walls and you get efficiency right up until the moment something goes wrong and there is nobody outside who knows how to fix it. Distribute the capability and the whole system gets harder to damage. I keep coming back to this because it maps so precisely onto what is about to happen in the enterprise economy. Three decades of concentrating operational capability inside large institutions — the IT infrastructure, the compliance machinery, the processing scale, all of it behind walls that only the biggest organisations could afford to build — and AI is about to blow those walls open.
In brief: Everyone argues about how to cushion the blow when AI eliminates knowledge-work jobs. This article asks a different question: what happens when the same AI that eliminated those roles also eliminates the operational barriers that kept those professionals trapped inside large institutions? The answer is market fragmentation — and a kind of redistribution that no policy paper is going to deliver.
This is Part 3 of a multi-part argument. It builds on the diagnosis in “Where the Light Is” and the architectural prescription in “Copilot Is Not the Architecture.” The frameworks from those articles — the distinction between the integration layer and the workaround layer, the composite integration stack, the promotion of agentic capability to the integration layer — are the foundation beneath this piece. This article follows the architectural argument to its economic conclusion.
The Moment After
Take a large specialty insurer or a mid-tier bank that has done the hard architectural work described in the first two articles. It has moved AI from the workaround layer into the integration stack. Agentic systems that used to sit between applications, operating screens as substitutes for human operators, now run as integration infrastructure. Systems that never talked to each other exchange data through a semantic mediation layer. The long tail of integration — the thousands of small, expensive, manual bridging operations that used to eat up an army of knowledge workers — is dissolving.
The board is satisfied. The architecture has proven out. And twelve hundred people have received generous severance packages.
Most commentary about AI and the enterprise stops right here — jobs lost, hand-wringing, next topic. But those twelve hundred are underwriters, claims people, compliance analysts, and operations coordinators with decades of domain expertise and professional networks built over entire careers. This is not a factory floor where the loom replaced the weaver. These people know things — specific things about specific markets, risks, clients, and regulatory regimes. The operational context got pulled out from under them, but the knowledge did not go anywhere.
Some get redirected into the work that stays human. Judgment under genuine ambiguity, relationship management, ethical oversight, regulatory interpretation where the regulation itself is unclear. But the maths does not bend. The enterprise does not need twelve hundred people making judgment calls. It needs two hundred, maybe three hundred if you count the roles that the transformation creates. The other nine hundred are surplus — not because they lack talent, but because the work that justified their positions turned out to be pattern recognition across disconnected databases, dressed up as expertise. That work is gone and it is not coming back.
The organisation did the right thing. It solved the integration problem rather than perpetuating the workaround, and the displacement is what happens when you do the correct thing. Pretending otherwise does not help the nine hundred people who need to figure out what comes next.
The Moat
Here is where the standard analysis goes wrong, and I think this is the most important section in this entire article so I am going to take my time with it.
The conventional framing treats the enterprise as a fixed thing that now employs fewer people. The optimist says new jobs will appear. The pessimist says the state needs to provide a safety net. But both of them assume the same thing — that enterprises are the natural units of economic organisation, and that individuals participate in the economy by being employed by one. Neither asks whether the displaced professional might not need either the enterprise or the state.
The enterprise is not a fixed thing — it is a bundle of capabilities, and AI is unbundling it. I have watched this from the enterprise software industry perspective for over thirty years. We can take any vertical as an example: so let me use insurance and \ sketch the pattern in other industries more briefly. A handful of large carriers and a few dozen established MGAs have controlled the specialty insurance market for as long as I can remember. Not because their underwriting judgment was better than anybody else’s — the best underwriters I have met were individuals, not departments — but because they were the only ones who could sustain the operational apparatus. The IT systems, the back-office processing, the compliance reporting, the claims infrastructure, and the data centres. Expensive to build, expensive to run, and no individual could afford any of it no matter how good their underwriting judgment was.
What kept these markets concentrated was not talent. It was operational overhead — the capital infrastructure, the scale economics that justify expensive processing when you have a thousand claims a day but not when you have ten, and the sheer gravitational pull of being the place where all the best people already work. The expertise always lived in the individual; the operational wrapper belonged to the institution. Clients paid for the wrapper because there was no other way to get the expertise delivered.
Commercial law, same picture. The biggest firms dominated because they had the research teams, the document production pipeline, and the admin support — not because every partner was sharper than a talented solo practitioner. A good lawyer working alone simply could not deliver at scale. The delivery infrastructure did not exist outside the firm.
AI collapses all of this at once, and the speed of it is what I think most commentary underestimates. Capital infrastructure is becoming cloud plus AI — the compliance platform, the processing engine, and the back office are available as services now, at a fraction of what it cost to build them internally. A two-person underwriting agency can access operational capability that ten years ago required a hundred people and a building. Scale economics dissolve because AI handles volume — ten claims or ten thousand, the marginal cost approaches zero, and the small operator is no longer at a disadvantage.
But the part I want to spend time on is what happens to talent concentration, because I think this is where the argument gets interesting and most people miss it entirely. The domain expert no longer needs the enterprise’s organisational wrapper. She brings the judgment, the relationships, and the market knowledge accumulated over twenty years. AI provides the operational muscle — the admin, the compliance docs, the data work, and the reporting. I have seen this combination work in practice and it is more responsive and more specialised than the enterprise department she left. The expert is not diminished by leaving the institution — and I think many of them know it already.
The expertise was always the individual’s. The operational wrapper was the enterprise’s. AI replaces the wrapper.
The Ventures
What do the displaced professionals actually build? Insurance is the example I know best and I am going to spend the most time on it.
Experienced underwriters form Managing General Agents and specialist underwriting agencies. They go after the market segments their former employers considered too small or too weird — specialty lines, niche geographies, emerging risk categories that don’t fit the standard models. AI handles policy administration, claims triage, and regulatory reporting. The underwriter brings what was always hers: the judgment to assess risks that the algorithms cannot price. Capacity providers — reinsurers, ILS funds — back the MGA’s judgment with capital. The person who was a cost centre inside a big insurer becomes a principal in her own market.
I have watched the operational overhead of running an MGA for twenty years — the IT infrastructure, the back-office processing, and the compliance machinery. That was always the barrier, not the underwriting skill and not the relationships. The plumbing. And the plumbing is what AI replaces.
Commercial law follows the same pattern — senior practitioners launching boutique firms for mid-market clients the big firms considered low-margin, with AI handling the procedural work, the research, and the document generation. Financial advisory, same thing — portfolio analysts forming independent shops where AI handles data synthesis and compliance docs while the advisor competes on insight rather than on assets-under-management scale, which was always the enterprise’s moat and not the analyst’s. I will not develop audit, consulting, or healthcare administration at the same length — the pattern is the same and I don’t think repeating it adds anything. The directional point is that where an industry had ten big players, it could end up with four hundred specialised ones, and the economics of that shift are already visible if you know where to look.
The Hard Questions
The credibility of this argument depends on not waving away the objections.
“This is just the gig economy with better branding.” I take this one seriously because the ugly version of this story is real and I have seen versions of it play out. The freed-up professionals end up as precarious contractors with no benefits, racing each other to the bottom on price. That happens.
But there is a structural difference between selling your time by the hour and building something that compounds, and it is worth being precise about where the line falls. The gig worker drives an Uber — undifferentiated service, no recurring relationship, competing on availability. The MGA owner underwrites specialty risk in a niche her former employer would not touch because the volume was too low. The boutique lawyer advises mid-market companies on cross-border regulatory problems that the big firms priced themselves out of. These people have clients, not tasks. They have recurring revenue, not gig fees. AI handles the operational machinery that used to require fifty people but it does not handle the twenty years of underwriting judgment or the instinct for which risks are mispriced. That stays scarce. Not every venture works out — obviously. But the ones that do are businesses, not gigs, and the distinction matters because it determines whether the displaced professional is building equity or burning time.
“Incumbents will entrench, not fragment.” Desktop publishing fragmented print media. Cloud computing fragmented enterprise software from a handful of ERP vendors to thousands of SaaS companies. When operational overhead drops by an order of magnitude, specialists who could not previously sustain a business suddenly can — and I don’t think incumbents entrench when the cost floor falls out. They lose their edges to people who know more about less.
“Capital requirements are still real.” Yes, an MGA needs capacity backing and a law firm needs professional indemnity insurance — not every barrier is operational and I am not going to pretend otherwise. But the operational cost was always the bigger barrier by a wide margin, and strip that out and what remains is fundable.
“Most people will not become entrepreneurs.” Fair. But even if ten percent of displaced professionals take the path, the market impact is large, because those ten percent are the most experienced and most connected operators in their fields. The argument is not that everyone becomes a founder. It is that for the first time the option exists at scale because the operational barrier — the IT, the compliance infrastructure, and the admin — is largely gone.
The Real Redistribution
The standard redistribution debate — how much to tax, whom to tax, what to fund with the proceeds — is about rearranging the output of a fixed economic structure. It accepts that the economy produces concentrated wealth and then argues about how to spread it around. Progressive taxation, universal basic income, transition funds, and retraining programmes are all patches applied to an economic architecture that concentrates by default. I have sat through enough policy discussions to know the shape of this argument and I think it misses the point entirely.
The architecture itself is changing. AI does not just redistribute the output — it redistributes the means of production. I use that phrase deliberately because it carries weight and I know it, but I am making a capitalist argument, not a socialist one. The operational capability that was locked inside enterprises — the processing power, the scale economics, the compliance infrastructure, and the admin machinery — is now available to individuals as technology, not as charity and not as policy.
The result is redistribution through participation rather than redistribution through extraction — more players entering markets that were closed to them by operational barriers. Instead of ten carriers controlling a specialty insurance market, four hundred MGAs and specialist underwriters competing across hundreds of niches, each running their own operations, each building equity in their own enterprises. The wealth is generated differently because the cost of participation has collapsed.
What Comes After
Return to the twelve hundred people. The pessimist sees a crisis. The optimist promises retraining. I think both are looking at the wrong thing.
Some of those twelve hundred carry domain expertise that their former employer spent decades building into them — underwriting judgment, claims knowledge, compliance instinct, and market relationships. That expertise is now unbundled from the operational wrapper that was the only way to deploy it at scale. The same AI that made the enterprise need fewer of them now makes it possible for them to compete without the enterprise. Not all of them will, but enough of them will to change the shape of the market — fragmenting concentrated industries into ecosystems of specialised, AI-enabled ventures.
And that creates a new problem. Four hundred participants in a specialty insurance market need to transact with each other, with carriers, with regulators, and with customers. The transaction infrastructure that lets a fragmented market work as fluidly as a concentrated one becomes the critical thing. What that infrastructure looks like, how it works, and why cooperative governance is probably the only model that can hold it together — that is a separate conversation entirely (see “Rethinking the Transaction”). But the demand for it follows directly from the market structure this article describes, and ignoring it means the fragmentation produces chaos instead of competition.
For the diagnostic foundation — why AI is overwhelmingly aimed at the workaround layer — see “Rethinking AI for Automation: Where the Light Is.” For the architectural prescription — the composite integration stack and the promotion of agentic capability to the integration layer — see “Rethinking AI for Automation: Copilot Is Not the Architecture.” For a deeper analysis of where competitive moats actually live when transactions flow through an intelligent integration layer, see “Rethinking the Data Moat.”