Own your AI, do not rent it

Owning your AI means holding the code, the prompts, the configuration and the data yourself, documented well enough that another developer could take the system over tomorrow. Renting means paying monthly for a black box that runs inside your business on terms someone else can change. The distinction decides who controls the system, what it costs over time, and whether it can ever be audited.

Much of the AI market sells the second model in the language of the first. The subscription is framed as a partnership, and the fact that you may never inspect what is running is framed as simplicity.

What does AI vendor lock-in actually cost?

Lock-in prices itself in three ways. Price rises you have to accept, because switching means rebuilding on another platform from zero. Behaviour changes you have to absorb, because the supplier can swap the underlying model overnight and your workflow behaves differently by morning, with no notice and no rollback. And a roadmap you have to wait for, because the feature your business needs sits wherever it sits in someone else's backlog.

Each month of history the platform accumulates raises the exit cost, so your negotiating position weakens the longer you stay. That dynamic is the product working as designed.

Why does ownership matter for governance and audit?

An auditor, a regulator or a board can only sign off on a system that can be inspected, and you cannot grant inspection rights over something you are never allowed to see. Prompts deserve particular attention here, because a prompt that decides how customer messages are answered is operating policy. A firm should be able to read its own policy, version it and change it, in the same way it can read its own procedures manual.

What should a handover actually include?

The source repository, the prompt library, the configuration, an export path for the data, and documentation written for whoever comes next, walked through in person rather than emailed as a zip file. One of our builds shipped an AI feature MVP for a B2B SaaS startup in 4 weeks and avoided 2 engineering sprints of internal build time, and the startup retained ownership of the code and configuration throughout. That feature survives any future change of supplier, including us.

This is the advisory habit applied to software. An adviser leaves the client with an asset and the ability to run it, and we hold our builds to the same standard. The handover pack is where that standard shows.