Talent Will Gather Where Risk Is Priced and Rules Are Written

Alfred Marshall noticed this in the industrial towns of Victorian England. Put enough people practising the same difficult trade in one place and the trade’s secrets stop being secrets. Apprentices absorb it. Good ideas get “taken up by others and combined with suggestions of their own.” The place becomes self-reinforcing, because the craft is easier to learn there than anywhere else.

Every great talent center in the last century was built around one scarce craft. Detroit had the engine. Hollywood had the moving picture. Wall Street had capital. Silicon Valley had the tool: the craft of turning an idea into a working piece of technology. First chips, then software, then the internet, then AI itself. For seventy years, if you wanted to be someone in technology, you went where that craft was in the air.

Here is the turn. The craft the Valley was organised around is the one being absorbed first. When a model can write the software, design the interface and wire up the system, building a tool stops being scarce. It becomes something you order from the intelligence layer, the way you order electricity. The mystery goes out of the air.

So what is still scarce? I made the case in The Signature Is the Product: when the work is manufactured, what people pay for is someone willing to stand behind it. That is risk. How likely it is to fail, what failure costs, who pays, how much capital has to be held against it, and what the contract says when it goes wrong.

That craft has a home, and it isn’t Palo Alto. It lives where three things are practised side by side, all day, in public:

• Capital

  Markets that price risk every second and put money behind the answer.

• Insurance

  The people whose whole job is knowing what it costs to be wrong. London has had that market since coffee houses.

• Law

  The contracts that say who owes what when it breaks. Much of the world’s commerce is written under New York or English law.

Only a handful of cities have all three in walking distance. That is why the next talent center is a financial capital, and why I would put it more precisely. It is not trading that pulls people in. It is the meeting of capital, insurance and law, because that is where the decision about what can be stood behind actually gets made.

Be clear about who is writing the rules here. It is not the labs, and at first it is not the lawmakers either. It is the financial services sector: the insurers and the banks. Every commercial business that uses AI needs an insurance policy and a line of credit, and the conditions on those two things are the real AI rules that business will live under. Insurers decide what AI use is coverable. Banks decide what AI risk they will lend against. That is the core of the signature economy: insurance companies decide AI policy, long before any statute does.

And the financial sector does not stop at its own forms. It lobbies. When insurers and banks want liability assigned a certain way, or a public backstop for catastrophic AI losses, they take that to Washington, Brussels and the other capitals. So the second magnet is the capitals where they lobby, and it pulls for the same reason as the first. Whoever can speak machine, risk and government in the same sentence is going to be among the scarcest people alive.

Why not the labs?

The obvious objection is that the talent will simply follow the frontier, to the labs where the models are made. I think the opposite happens. The frontier of AI research is itself being handed to AI. The most important work in a leading lab is increasingly the supervision of systems that are researching how to improve their own code. That is the purest case of the craft being absorbed: the tool that builds the tool that builds the tool. The lab of the future is a secure building full of compute, overseen by a small number of people. It is enormously important, and it will not be where the crowd is. You cannot gather a generation around a craft that the machine has taken over. You gather it around the decisions the machine cannot make for itself: how much risk to carry, and what society will permit.

Not with the lab: with whoever sets the terms

Palantir already works this way. Most of its people sit not in a central lab but out in the field, where the software meets the real decisions. The product is made centrally. The value is made where it gets used and judged.

Apply that to the stand-behind economy and follow it one step further. The customer using AI is the ordinary commercial business: the restaurant, the manufacturer, the clinic, the law firm. But that business does not get to decide on its own how it may use AI. Its insurer and its bank decide, through the terms of its coverage and its credit. So the talent that matters most does not sit with the model maker, or even with the business. It sits with the insurers and banks writing the terms every business must meet, and with the people carrying those terms into government. Those are concentrated in exactly two kinds of city: the financial capitals and the political ones.

1. The prestige job flips from builder to underwriter.

   For a generation, the ambitious graduate wanted to start a company or join one early. By around 2030, the coveted first job is on the risk side: underwriting AI, structuring warranties, auditing machine work, writing the contracts. The status will follow the scarcity, and the scarcity is judgment about failure.

2. The AI labs’ centre of gravity drifts to New York, London and Washington.

   The models will still be trained wherever the power is, and supervised by a small crew. But the part of an AI company that closes the big deals, signs the warranties, partners with insurers and negotiates with governments will need to live beside its counterparties. Watch for a major lab moving a senior risk, legal, policy or enterprise function out of California.

3. Insurers and banks become the biggest employers of AI talent.

   The best technical people will not sit in the lab that makes the model. They will sit inside insurers and banks, writing the conditions every commercial customer’s AI must meet, testing whether a business’s AI practice is coverable, and deciding what can be lent against. The financial sector becomes the de facto AI standards body, and it hires like one.

4. The financial sector’s AI lobby becomes the one that matters.

   The loudest AI voices in Washington and Brussels today are the labs. I expect the most effective ones by the end of the decade to be insurers and banks, arguing over who carries liability and who backstops the catastrophe. The person who can explain how a model actually fails, in the language of risk, to a committee will be worth more than the person who can build one.

5. A new kind of firm is born in the financial centers: the signature house.

   Part insurer, part law firm, part engineering shop, part bank. It sells guaranteed outcomes: this process will be done, to this standard, and here is who pays if it is not. It will look strange to the Valley because its product is not a tool. Its product is a promise with capital behind it. I follow this one all the way through in The Signature House.

6. Silicon Valley becomes a foundry town.

   It stays rich, powerful and essential, the way a great industrial town is essential. But it becomes a place of a few enormous, capital-heavy companies rather than a place where the ambitious go to find each other. Its status goes before its money does. Most people will notice only when the brightest young people they know stop moving there.

7. Asia’s seat is decided by law, not technology.

   There will be an Asian center of the stand-behind economy, and the contest for it will not be about engineers. It will go to whichever city the world’s contracts trust most, which is a question about courts and rules, not chips. That is an open contest, and it is being decided now.

8. Every region grows a small signature tier.

   The same logic works at local scale. In a city like Charleston, the local bank, the local insurance agent, the local law firm and whoever delivers AI work to local businesses become a small version of the triangle. Local trust becomes an asset again, because someone near you has to be answerable.

I co-founded a software company, E2open. The tool was the game, and the tool culture was the culture I knew. Charleston AI is built the other way. It is a service provider, and it makes and keeps promises in an unusual order: we do the work, the customer receives it, inspects it, puts it to use, and only then pays. The customer pays after. That one choice changes the whole risk dynamic, because the risk of the work not being right sits with us, not with the customer, until the customer is satisfied. It is a small, local version of exactly what this piece is about. The tool is becoming the cheap part. What holds value is someone who stands behind the result before asking to be paid for it.

This piece is the third in a set. When Intelligence Comes Off an Assembly Line asked what happens when thinking becomes a manufactured good. The Signature Is the Product argued that the value moves to accountability, and that insurers will write the rules. This one asks where the people go: to where risk is priced and rules are written.

Already true

Building software is being absorbed by AI faster than any other craft. Insurers and lawyers have started rewriting the terms under which AI work is covered. The largest AI companies already keep major offices in New York, London and Washington, and AI policy is now argued in every major capital. Leading labs already use their own models to accelerate their research, and insurers have begun writing AI into, and out of, their policies.

What has to happen

The scarce skill has to move from building to standing behind and getting permission. If the hard, well-paid, status-bearing questions in AI become “who is liable, how is it priced, and what is allowed,” the people will follow them, because they always follow the scarce craft.

Where I am probably wrong

Silicon Valley has absorbed every previous shift, from chips to software to the internet to AI, by reinventing itself, and venture capital is itself a risk-pricing business. It could simply learn the new craft at home. My answer is that venture prices the risk of building something, while the new risk is the risk of running it at scale, and that is the craft of insurers, lenders and lawyers. The other way I could be wrong: talent may not need a room at all anymore. If the unwritten part of the craft can travel over a screen, Marshall’s air stops mattering, and there is no new center, only a network.