Is It Time for Guilds to Come Back?

I was listening to a McKinsey podcast recently in which a senior partner made two claims within a few minutes of each other. First, that as AI makes prediction cheap, human judgment becomes the scarce and valuable skill. Second, that organisations should automate their low-level, entry-level tasks and keep the people "who have the capacity to oversee the system."

Nobody in the conversation noticed that the second claim quietly destroys the first.

The entry-level grind is where judgment comes from. It always has been. You don't develop the instinct for a bad requirement, a fragile design or a client about to walk by taking a course on it. You develop it by doing a thousand small versions of the work, getting some wrong, and having someone more experienced nearby when you do. Automate the thousand small versions and you haven't made the workforce more strategic. You've switched off the machine that produces the strategists.

So here is the question that has been circling in my head since: if judgment is about to be the scarce asset, and we are dismantling the way it was made, what institution rebuilds it? I keep arriving at an old answer. I think the guild is coming back. And I think that's mostly a good thing, provided we're honest about which half of the guild we want.

The firm was doing three jobs. We only noticed two.

The twentieth-century corporation earned its place by doing three things more cheaply than the open market could.

The first was coordination. Ronald Coase pointed out nearly ninety years ago that firms exist because organising work through market transactions has costs, and below a certain threshold it's cheaper to organise it inside a company. The second was capital: buildings, machines, infrastructure, distribution, things no individual could afford.

Over the last two decades, platforms unbundled both. Stripe became the finance department. Shopify became retail operations. AWS became the infrastructure team. Marketplaces became the sales force. And now AI agents are absorbing the junior staff layer itself. This is why the one-person business is suddenly everywhere: the firm's first two advantages can be rented by the seat.

But the corporation was quietly doing a third job that nobody priced, because it never appeared on an invoice. It hired cohorts of juniors at a loss, gave them small survivable work, sat them near experienced people, and turned them into professionals with judgment over about a decade. It could afford this because long tenure let it capture the return.

I've written before about how that bargain broke, layoffs taught employees that loyalty flows one way, training budgets shrank, mentoring became a form and then nothing. The point worth repeating is that AI didn't kill the apprenticeship system. It was already dying. AI is the coup de grâce, because it removes the last economic reason to hire the inexperienced at all: juniors used to earn their keep doing the grunt work while they learned, and the grunt work is now automated.

The firm's first two functions have been rebuilt as platforms. Nobody has rebuilt the third.

The solo entrepreneur is spending inherited capital

Which brings me to the narrative being pushed rather hard at the moment: the AI-leveraged solo founder, the one-person company, employment reimagined as a portfolio of ventures.

Look closely at who is actually succeeding at this. Overwhelmingly, they are mid- and late-career people, veterans who spent twenty years inside the old system, absorbed its judgment for free, and are now spending that inheritance with AI as leverage. The model works brilliantly for exactly one generation. It consumes judgment capital that was formed inside the institutions it is replacing, and it replenishes none of it. A twenty-four-year-old following the same playbook gets infinite execution leverage and no mentor: AI hands them a tireless, confident junior associate when what they're missing is the partner at the next desk.

And don't expect the enablement platforms to fix this, because their incentives point the other way. A platform's economics prefer many dependent median users over fewer independent expert ones. A merchant with real judgment can negotiate, switch, or leave; a dependent one keeps paying. Shopify will teach you to use Shopify brilliantly. It will never teach you when Shopify is the wrong answer. Every platform enables execution. None of them enables formation. That gap is structural, not an oversight.

Why the guild, and why now

So: solo work has no formation pathway, the corporation abandoned its one, and platforms won't build one. What's left is the scale in the middle, and here the economics have shifted in a way I don't think has been fully appreciated.

Small firms were historically the free riders of the training system. A ten-person company couldn't afford to carry a junior who'd be net-negative for two years; it hired people the big firms had already trained. If the big firms stop training, the small-company talent supply dies too.

But AI changes the arithmetic at precisely this scale, and in the opposite direction to the corporate story. The reason a junior was unaffordable was the two unproductive years. If AI absorbs the grunt work, a junior plus AI is doing useful work within weeks, and the small firm offers the one thing the corporation never could: whole-business observability. In a ten-person company you cannot hide in a specialised role. You watch the founder handle the angry client. You see the invoice get chased. You're in the room when the pricing decision is made. That's the "learning by watching" that I've argued has to be deliberately rebuilt, except at this scale it happens as a by-product of the office being small.

A master with decades of judgment. A couple of journeymen. Apprentices who are economically viable from month one because the agents do what apprentices' drudgery used to be. That is not a metaphor for a guild. It is a guild, minus the funny hats.

Even the solo entrepreneur story bends toward it. Solo isn't an end state; it's a larval one. The successful founder hits a wall somewhere around employee number two to ten, where they must transmit judgment they've never had to articulate, and discovers personally, that the formation infrastructure doesn't exist. The guild is what the solo economy grows into the moment it succeeds.

The half of the guild we should leave in the ground

Now the honest part because the word "guild" carries baggage it has thoroughly earned.

Historical guilds were not just apprenticeship systems. They were cartels. They fixed prices, restricted entry, punished innovation that threatened members, and passed privilege down family lines. They were abolished across Europe for good reasons, and any modern revival that drifts toward gatekeeping, credential walls, closed shops, artificial scarcity of practitioners deserves to fail the same way.

The distinction that matters: the old guild's legitimacy came from exclusion, controlling who was allowed to practise. The revival's legitimacy has to come from formation, being demonstrably the place where judgment is actually made. One is a moat around the members. The other is a promise to the apprentice. Keep the master-journeyman-apprentice ladder, the observable work, the graduated stakes, the accountability of the master's name attached to the apprentice's development. Leave the protectionism behind, because in a world where AI makes raw output free, exclusion is unenforceable anyway. The only defensible scarcity is genuine judgment, and you can't gatekeep your way to that. You can only grow it.

What the modern guild actually needs

If the small firm is the venue, it still needs a toolkit, because a ten-person company can't improvise an apprenticeship system from scratch. I think it has three parts, and it's worth being precise about them, because two are being built and one is missing.

Rehearsal. Safe, repeatable practice environments; the small survivable rungs before the frightening real thing. These are starting to appear; interview and scenario simulators calibrated by named experts are an early form. One caution I'd carve in stone: AI simulation is fine for rehearsal and dangerous for calibration. A model trained on the internet's average teaches internet-average stakes. Practice against the machine; have your standard set by a human who has actually carried the consequences.

Observation on live work. The apprentice watching the master's actual decisions on actual problems, with graduated rungs from watching, to proposing, to doing with a safety net. This is the heart of it, and it mostly requires deliberateness rather than technology: think out loud, take the awkward call with the junior sitting beside you, let them see the recovery as well as the polish.

Consequence follow-through. The missing third leg, everywhere. Judgment isn't just calibrated against the master; eventually it has to be calibrated against reality. Did the confident call made in March survive contact with September? Almost no organisation closes that loop, decisions are made, outcomes arrive months later, and nothing connects them back to the reasoning. The old guilds had this built in: the master's name went on the work, and the town knew whose roof leaked. Any serious revival needs its equivalent.

The masters are the constraint

Here is the uncomfortable dependency in all of this. A guild is only as good as the masters willing to be watched, and the people holding the deepest reserves of judgment are, disproportionately, the generation now heading for the exit. Every retirement without transmission is a small library fire.

That's why I don't think this is a nostalgic idea. It's a race. The formation machine has been off for long enough that the industry is running on inherited judgment, AI is accelerating the drawdown, and the people who could restock it have maybe a decade of overlap left with the people who need it.

So, is it time for guilds to come back? The formation half, the ladder, the observable master, the name on the work, never deserved to die, and the economics that killed it have just reversed. I'd say it's overdue.

The cartel half can stay buried.