What do you own
in
a custom AI solution?
Only what
the contract lists.

A custom AI solution is software built for one job in your company, on a rented AI model. What you own when it’s done is whatever the contract lists.

Tony DrummondFounder & Client Success

The proposal says “custom AI solution” on the cover and runs eleven pages. Somewhere in there should be the answer you’ll want in two years: if we part ways, what do we keep?

Often it isn’t there. The question comes up the week you want to leave, which is the worst week to ask it.

A custom AI solution is a build on top of a rented model

Most custom builds come in five layers, whoever makes them.

The model stays rented either way. Everything above it is yours only if the contract names it.

Your data sits on top, and it’s yours before anyone starts. Below it are the workflow and the prompts written for your job. A prompt is the written instruction the model follows. Then comes the code that links the system to your other software, and the cloud account it runs in.

At the bottom is the AI model, the engine that reads and writes. A handful of companies train these and rent out access. OpenAI, Anthropic and Google are the names you’ve heard. Almost no company your size builds its own.

Buy off-the-shelf when the job is common

In one year, large companies went from close to an even split to buying three uses in four.

• The codedelivered into a repository in your company’s name.
• The prompts and workflow settingsin files your team can open and edit.
• Your data and everything the system producedexportable in a standard format whenever you ask.

• Any model trained or tuned for youand the files that hold what it learned, called weights.
• A list of the outside and open-source parts inside itwith the license each one comes under.
• The right to move the whole thing to another providerand help doing it.

What a build costs follows how many systems it connects and how messy the inputs are. Ask for a fixed quote on one small first piece, and put this list in that first contract.

The AI consultant guide covers what to keep from any outside firm. The agency guide covers the files behind a content workflow. This list is for a build, one layer at a time.

Ask before the first payment. After it, you’re asking a favor.

Switching builders is a handover only if you hold the code, prompts, data and cloud account

If the old builder keeps even one of the four, you start again.

Expect to switch something. Models get replaced by newer ones. The firm that built your system may be gone in three years, or may no longer be the right one.

One question decides whether the next builder picks up your system or starts from nothing.

Questions owners ask about custom AI solutions

• What are custom AI solutions?

  Software made for one job inside your company. It runs on an AI model rented from a company like OpenAI, Anthropic or Google. Who owns the rest gets settled in the contract.

• Should we build custom AI or buy off-the-shelf?

  Buy for the jobs every company has, as long as the product fits how your team already works. Save the build for the one job no product covers.

• How much does a custom AI build cost?

  No fair figure exists until the job is named. Then it turns on what the build touches, the state of what it reads and who runs it later. Start small, at a fixed quote, and decide on the rest after.

• Who owns a custom AI system?

  Whoever the agreement names. The AI model inside it is rented. The code, prompts, workflow settings and cloud account are yours only if they’re written in.

• Can I copyright what my AI system produces?

  Only the parts where a person contributed enough of the expression, the US Copyright Office said in January 2025. Writing the prompt isn’t enough on its own.

• What happens if we want to switch AI providers?

  Own the code, prompts, data and cloud account and the next provider takes over where the last one stopped. If the old builder holds any of them, expect a rebuild.