Buy vs build vs own: the third option nobody talks about.
The debate is stuck between two false options. Buying leaves you a tenant. Building from scratch bankrupts you. Owning an OS is neither.
Every CTO in financial services has had the buy-vs-build conversation about AI. It is the wrong conversation. Buying leaves the firm renting intelligence forever. Building — meaning training your own frontier model — bankrupts you and misses the point. There is a third option, and it is the one serious firms are choosing.
The three options, honestly
- Buy: subscribe to a vendor's product. Fast, cheap, non-proprietary, non-compounding, non-portable.
- Build (naïve): train your own model. Expensive, slow, mostly a distraction — the frontier will always be ahead.
- Own the OS: assemble a private platform on which the firm's own data, workflows, agents and skills live, using the best model for each job — private or frontier — inside a router the firm controls.
Own the OS. Rent the models when it makes sense. Never confuse the two.
The economics
Owning the OS is a build cost that lands between a serious buy and a serious build — typically £75k–£300k for the first shippable slice, then a monthly Operator retainer. The compounding return sits on top of that: every workflow, agent and skill the firm ships onto its OS is a proprietary asset that does not need to be rebuilt.
Where to start
Every KJ Capital engagement begins with a two-week £15k AI Diagnostic. We map your existing data estate, model your firm's OS across the six layers above, cost the build, name the first three skills to ship, and hand you a plan your board can act on. It is deliberately priced to be a rounding error against the value it unlocks.
FAQ
What if we already bought a vendor product?
Keep it. Slot it in behind your router. Own the layers around it — Firm Memory, Skills, Workflows — so the day you swap that vendor, nothing else has to move.
Isn't owning the OS just building?
No. Owning is assembly plus ownership. You use open-weight models trained by others, frontier APIs where they earn it, and open-source infra. What you own is the shape, the data, the skills and the guardrails — not the model weights.
Start with the £15k AI Diagnostic
Two weeks, one costed OS blueprint, three named skills to ship first.