The internet was built for humans.

Film had Hollywood. Software had Silicon Valley. Hardware ended up in Shenzhen. Cloud quietly concentrated in Seattle.

London built one of the largest service economies in the world. That matters if the next generation of software is built around real industries instead of trying to escape them.

If agents are going to do real work, they will need to live inside systems that decide what is allowed, what gets paid, and what has to be explained later. London sits unusually close to that world.

The reason this feels sudden is that two separate things finally lined up.

Models got good enough to deal with messy requests. They can read something half-formed, guess the intent, and move the task forward.

At the same time, the systems underneath the economy got easier to plug into. Payments, banking, logistics, identity vendors, procurement tools, internal dashboards. More of the world started exposing APIs. Now even internal tools surface through MCP, CLIs, and skills.

When both are true, software can take a request that used to land in somebody's inbox and carry it several steps further on its own. That used to require a person in the middle. Increasingly, it does not.

Real technical waves do not just create companies. They create hubs that other companies, investors, and infrastructure attach to.

If the agent paradigm becomes economically important, the world will produce one or more hubs like that again.

If software is going to act on your behalf, someone still has to answer a set of very old questions. What was it allowed to do? How did it identify itself? Where did the money move? Who signed off on it? Who explains it when something goes wrong?

That is the boring infrastructure that surrounds autonomy: systems, permissions, payments, audit trails, procurement, compliance. The machinery that decides what software is trusted to do.

That is why this will cluster. The companies that build the defining agent businesses will not be the same as the companies building foundation models. They will win by knowing how to operate inside those systems.

The hard part is not making software act. It is making institutions trust it.

If that still sounds abstract, picture the actual companies.

Money. Insurance. Procurement. Trade. Identity. Compliance.

An important London agent company might not look like a chatbot. It might triage insurance claims, gather missing documents, route exceptions, and chase counterparties across messy workflows.

Another might sit inside trade and logistics, resolving customs exceptions, reconciling shipment data, and moving information between freight systems that were never designed to talk cleanly to one another.

Others may look more like procurement infrastructure, vendor due diligence, or compliance remediation than anything the consumer internet taught us to admire.

Building Consent.io has made this feel obvious to me. Cookie banners, GDPR settings, CCPA rules. Customers do not really want to manage any of that directly. They want systems that notice when something is off, adapt to the policy, and keep the compliance layer working without turning it into another manual job.

The UK is unusually well placed to broker this kind of work. London sits inside finance, insurance, law, consulting, trade, and logistics, with dense relationships into Europe, the Middle East, Africa, and Asia. It is a city built around approvals, exceptions, counterparties, and paper trails.

That institutional density is paired with a nearby technical pipeline. Imperial and UCL are in the city. Oxford and Cambridge are close enough to feed founders, researchers, and early employees into the same network. Talent alone does not make a hub, but it matters when it can plug directly into customers, capital, and hard problems.

That matters more than AI hype.

The defining agent companies will not live in chat. They will live in back offices, close to customers, counterparties, regulators, and old operational systems they cannot route around.

The first successful agent companies will probably look less glamorous than people expect.

They will not start as general assistants. They will start by fixing one expensive workflow that currently requires three spreadsheets, six emails, and someone chasing signatures.

One company might handle insurance claims intake and document gathering. Another might reconcile shipment records across freight systems. Another might automate vendor due diligence across procurement teams.

None of these look like consumer AI products. But they sit inside real economic systems that move money, approvals, and risk.

The first agent companies will not replace people. They will remove the most painful step in a workflow large organisations already run every day.

The mistake is assuming every technical wave rewards the same geography.

London did not miss the last cycle because its founders lacked ambition. It missed because the biggest software companies of that era were built far from the systems London knows how to sell into, operate inside, and compound around. Agents may flip that.

If the last generation of startups won by escaping real-world systems, this one may win by embedding inside them.

Silicon Valley is still the best place in the world to start many kinds of software company because it has the deepest concentration of repeat founders, elite engineers, experienced operators, and investors who already know how to scale one.

But this wave may reward something else as well: proximity to regulated markets, procurement-heavy industries, ugly workflows, and cross-border systems where permission matters as much as product taste. That is a game the UK can plausibly win. London sits inside finance, insurance, law, and trade, while Imperial, UCL, Oxford, and Cambridge keep technical talent within its orbit.

The industries everyone calls boring are often the ones still held together by email, PDFs, and people chasing approvals.

Most of the good companies in this category will start with one ugly workflow and one customer who needs it fixed now.

Places like this usually form when a few founders keep solving the kind of problem other people overlook, then stay long enough for the gains to compound locally.

Then the second-order effects kick in. The first good company creates alumni. Early wins create reference customers. Some of the proceeds get reinvested. The next team starts with sharper instincts, warmer introductions, and a reason to build in the same city.

London does not need a slogan. It needs a handful of unusually good companies whose success produces credibility, capital, and founders for the next wave.

If you are building one of these companies in London, stop acting like you need to apologise for your postcode.

For the last twenty years founders learned a practical lesson: much of the venture industry sits in the United States. If you wanted capital, attention, or customers, you often pointed your company in that direction.

That is fine. Take American capital. Build global companies. Spend time in San Francisco if it helps you grow.

But stop pretending the company is not being built here.

If the next generation of software is built around real industries, proximity will matter again. Banks, insurers, regulators, trade networks, compliance teams. A lot of that infrastructure already lives in London.

Taking American capital does not make your company American. Building great companies in London is how London becomes a capital.

Venture capital does not create technology waves. It follows them.

Silicon Valley did not become the centre of software because venture firms declared it so. Venture firms moved there because the companies were already forming.

The same pattern tends to repeat. Founders solve a painful problem for buyers already concentrated in one domain. The first good company creates reference customers, alumni, and operators with money and conviction to back the next one.

American venture capital will continue to play a huge role in the next generation of companies. Many of the best London startups will raise from US investors.

But capital follows gravity. If the defining agent companies start forming close to the large organisations that buy them, regulate them, and create their first hard workflows, investors will go wherever those companies are being built.

Capital moves to where the companies are. Not the other way around.