The London Market has always understood ownership. We own our risk, our capital and our data, and centuries of hard lessons sit behind the way we manage all three. So it strikes me as odd that artificial intelligence is quietly asking us to give ownership away twice over, once in how we build our systems and again in where those systems get their intelligence from. Both feel like progress, and both deserve a harder look than they are getting.
Vibe coding is still coding
Watching AI take a described problem and turn it into a working application minutes later is hard not to be impressed by. The industry refers to it as ‘vibe coding’, and in the interest of prototypes, it is considered a genuine superpower.
The catch however is what happens afterwards, because every line an AI writes is a line somebody must own, understand, secure and one day explain to a regulator.
AI has collapsed the cost of producing code while doing nothing to the cost of owning it, and in a core insurance system, owning it is the whole cost. A dashboard that is 80 per cent right is just a Tuesday, but a capital calculation or a premium settlement that is 80 per cent right is a regulatory incident with your name attached, and the LMA has already reminded us that accountability cannot be outsourced.
What a quick win really costs
An MGA’s story reached me recently, and has stayed with me since. In a few days they vibe coded a document ingestion tool on a platform reselling capacity from one of the big American labs, the demo was superb and the proof of concept cost almost nothing. Then the volumes arrived. The cost per extraction on that rented arrangement at scale runs roughly twenty times what the same function costs on open source models in your own cloud, so across millions of documents it becomes millions of pounds a year for identical work. None of that is an accident. The model is to hook you at the proof of concept, where the economics look wonderful, then collect in years two and three once the workflow is embedded and the dependency is structural. The fact of the matter is, the cheapest system you will ever run is the one you build right the first time.
Renting your reasoning
That story is one instance of a much bigger pattern. Having embraced AI, the reflex is to thread the whole business through a frontier model behind a metered API, which quietly makes the intelligence you depend on something you neither own nor control.
The price risk is real, and even Uber managed to burn its entire annual AI budget in four months. Not even the giants are on stable ground, and Microsoft is busy replacing the labs it once relied on to cut its own bill.
The availability risk is worse, because frontier AI is now a dual use technology, and when export rules meant one leading lab could not verify its users' nationality, it switched off its most advanced models for everyone on the planet overnight, stopped by a border rather than a bug. These models are also generic by design, so to use them your regulated data must leave your estate for somebody else's servers, which for a market built on bespoke judgement is precisely backwards.
Own the engine
There is a better path, and it has become realistic. Open source models are now good enough to run seriously on infrastructure you control, and around them you can securely build adapters that encode your own wordings, your own workflows and the expertise your people have spent careers accumulating. The economics ultimately reward you, because intelligence you build is an asset that compounds and raises the value of the business, while intelligence you rent is an expense that compounds for your supplier and leaves your secret sauce on somebody else's servers. None of this means building everything yourself, which was the first mistake. Build on proven foundations, extend them where you are genuinely distinctive, and connect AI through governed interfaces so it acts only through narrow, logged permissions and never with a free hand over the core. We own our risk, we own our capital and we own our data, and it is time we owned our intelligence too.
Stuart is the Founder and CEO of Buckhill, the company behind the award-winning C2MS® platform. With 15+ years of experience in insurance, finance and technology, he leads Buckhill’s strategy and product innovation while guiding marketing expansion and technical prowess. Stuart has spent his career designing qnd modernising the systems that sit beneath policy administration, claims, underwriting and delegated authority business. He is an award-winning thought leader on systems architecture and AI sovereignty and is passionate about conversations surrounding what technology ownership really means in a regulated market.
.png){kind=logo}
.png)
.png)
.png)
.png)
.png)
.png){kind=logo}