Rented by the hour.

Two firms. Same answer. One logo.

Picture a boardroom. Two firms present for the same mandate — a strategy, a platform build, a transformation program; the scene plays out identically across consulting, technology delivery, and engineering services. The proposals are close to identical: same diagnosis, similar architecture, comparable rigor. One firm carries a storied logo and a rate card to match. The other doesn't. The buyer picks the logo, and everyone in the room, including the buyer, quietly knows the thinking was equal.

So what, exactly, was purchased?

I spent years inside the rooms where that premium was manufactured, and I've spent the last decade competing against it from outside. The honest answer is that the premium was never one thing. It was four things, bundled so tightly that buyers — and often the firms themselves — stopped distinguishing them. AI has now pulled the bundle apart, and it treats the four components very differently. Understanding which is which is, I'd argue, the most commercially important question in professional services right now.

What the premium was actually buying.

Note what this breakdown is and isn't. It's not a theory of what these firms are — it's a theory of what clients were paying extra for. That distinction matters, because AI doesn't attack firms. It attacks scarcities. And of these four, it reaches exactly three.

Three moats just got re-priced.

Re-priced — not erased. The distinction is the difference between an argument that survives scrutiny and one that doesn't, so let me be precise: AI hasn't made expertise worthless. It has crushed the scarcity value of expertise's codified forms.

Knowledge capital first. The largest field experiment on this to date — run by Harvard researchers with BCG's own consultants, 758 of them — found that for tasks within AI's competence, consultants using it completed roughly 12% more work, 25% faster, at markedly higher quality. Read that from the buyer's side: a meaningful slice of what the doctrine engine used to make scarce is now on the table for anyone. The engineering half of the industry is living the same moment: reference architectures, integration patterns, and accelerator libraries — the codified assets that justified premium rate cards — are precisely what frontier models now reproduce on demand. And watch the incumbents' own behavior, which is the tell. McKinsey built an internal AI trained on its knowledge base; by the firm's own account it now fields half a million prompts a month, used by three-quarters of the firm. That is not a company dabbling; that is a company defending. But read the defense carefully, because it concedes the thesis: an internal AI makes the incumbent faster — it does not make the incumbent scarce. The premium was never "we have playbooks." It was "you cannot get this thinking anywhere else." Once doctrine-grade reasoning is available to every buyer and every boutique, owning your own machine layer protects your margins, not your moat.

Capacity second. The pyramid is visibly cracking. Graduate intake at the Big Four fell by double digits — nearly thirty percent at one firm — with the UK, where the hiring numbers are public, offering the cleanest window into the pyramid's economics. Starting salaries have been frozen for years. One of the majors is dismantling its traditional analyst-to-manager title ladder entirely, and the industry's own commentators now describe the firm of the future as a "diamond" — thin at the base — rather than a pyramid. And the arithmetic hits technology delivery even harder than consulting: agentic coding compresses into days what was billed in person-weeks, and the great offshore pyramids — built on armies of fresh graduates doing exactly the work AI does first — are slowing their intake for the same reason the Big Four are. A senior McKinsey technology leader put the question publicly: "Do we need armies of business analysts creating PowerPoints? No." When the pyramids' owners start asking it aloud, the moat is already gone.

Craft third — with a scalpel, not a hammer. One datapoint stands for the whole category: a contract review that billed roughly $2,480 across six hours of associate time, reproduced by a frontier model in eleven seconds for a few cents. In engineering the same collapse wears a different face: the polished proof-of-concept that once took a solutions team six weeks is now a weekend's work, and production-grade scaffolding — tests, documentation, deployment pipelines — generates alongside the code. Polish — the board-ready deck, the flawless demo — is no longer scarce, and a signal that anyone can send is no longer a signal. But craft was never only polish. Framing the problem, reading the politics of a room, sequencing a recommendation so a board can actually swallow it — and in engineering, the system-design call, the tradeoff nobody asked about, knowing what not to build — that is craft as judgment, and it has not been leveled. The same Harvard/BCG experiment proves both halves at once: on a task deliberately designed outside AI's competence, consultants using it were significantly more wrong than those without it. The frontier is jagged, and knowing where it runs is itself expertise.

The fourth moat holds — and splits the market.

Now the part the triumphalists skip. There is no dataset showing the trust premium collapsing. The legal market is the cleanest mirror we have, because its numbers are public: midsize firms are growing demand faster than the elite tier — a real leak of routinized work downstream. And yet, over the same period, elite firms posted double-digit revenue and profit growth, with top litigators crossing $3,000 an hour. Both things are true at once. That isn't collapse. It's segmentation: the routine middle leaks downward, while the highest-stakes work rewards reputational insurance more than ever. When the bet is existential, the buyer still pays for cover.

And one concession the thesis owes its critics: trust is not fully independent of the other three moats. Clients trusted the elite firms partly because of demonstrated analytical horsepower and immaculate craft — so as those level, the foundations under the brand erode too, slowly, from below. Trust is the most durable moat. It is not a permanent one.

Trust can't be leveled. It can only be flanked.

Here is the asymmetry this essay exists to name. The first three moats get leveled: AI raises everyone's floor, and the scarcity premium dies. Trust doesn't work that way — no model grants you an alumni network or a forty-year brand. But trust has a different vulnerability: it can be made unnecessary. The challenger doesn't out-trust the incumbent. The challenger shrinks the territory where trust is required.

Flank one: proof over promise. Ask why "nobody got fired for hiring the logo" existed at all. Because engagements were big, slow, opaque bets — and the brand was insurance against their failure. AI attacks this not by giving anyone a brand, but by shrinking the bet the brand was insuring. When a challenger walks in with a working demonstration built at their own cost, a diagnosis produced in days, and a first phase measured in weeks — the purchase changes category. It stops being a leap of faith and becomes an evidence-based continuation. Small, fast, visible bets don't need insurance. The incumbent sells a promise backed by a logo; the challenger sells proof backed by a prototype.

Flank two: borrow the door. Networks can't be generated, but they can be partnered. A challenger pairs a locally trusted front — a partner with standing, a director with a reputation, an entity of record in the buyer's market — with leveled capability behind it. The door is borrowed; what walks through it is entirely new. This move is as old as trade. What's new is that the capability side of the pairing is no longer the hard part.

Notice the trajectory this produces: flanking works from below, upward. It captures the commoditizing tier first, builds evidence, then climbs — the classic disruption path. Trust isn't defeated in this story. It's bypassed for a growing share of the market and concentrated at the top of a shrinking pyramid, which is exactly the reshaping the incumbents' own hiring numbers describe.

This argument has ancestors.

Intellectual honesty requires naming them. Clayton Christensen and colleagues argued in 2013 that consulting would be disrupted as its knowledge assets unbundled from bespoke human delivery — the thesis before the tool existed. Richard and Daniel Susskind argued a decade ago that technology dismantles the professions' production monopoly while leaving the trust and accountability function standing. And more recently, Grace Shao has made the sharpest adjacent observation: that much of what looked like a moat in professional services was really a labor-cost structure the market mistook for defensibility.

What this essay adds to that lineage is two things. First, the taxonomy: naming the four components of the premium — knowledge capital, capacity, craft, trust — as distinct scarcities that AI treats differently, rather than one bundle that either survives or doesn't. Second, the asymmetry: three of those moats get leveled; the fourth can only be flanked. That distinction — leveled versus flanked — is, I believe, the most useful lens available for reading what happens to this industry over the next five years, because it tells you where the fight happens: not brand against brand, but bet size against bet size.