What to ask an AI partner before you sign

Before signing with an AI partner, the questions that protect you most aren't about technology but about incentives and ownership: what the success metric is, who will own the code and the configuration, how and when the system is handed off to your team, and where human control sits. A good partner answers all four without hedging. If they dodge the one about the end of the project or the one about ownership, the technical answers matter less: the risk isn't in the model, it's in the relationship.

"What is the success metric and when is it ready?"

Ask that the project be tied to a single business metric —time saved, volume handled, errors avoided— set before starting. This question filters a lot: a partner who can't define when the system is finished is describing, without saying so, an endless project. The metric is also the only way to demonstrate AI ROI beyond the sense that "it's going well." If it can't be defined, the use case probably isn't mature yet.

"Who will own the code, the prompts and the configuration?"

The answer has to be unambiguous: you. Not only in the contract, but in practice —with documentation your team understands and open standards instead of black boxes. Real ownership is what separates having an AI in production system from renting access to one. If the code, the prompts or the integrations stay on the provider's side, every future change will go through their invoice.

"What does the handoff to my team look like?"

A partner aligned with your interests treats its own exit as a goal. Ask how the system is documented, how your team is trained and how it ensures you can operate without them. Against the model of large consulting firms —big teams billing for months, with no incentive to finish— what you want is the opposite: someone who enters, builds on what you already have and leaves when your team runs the system on its own. The handoff isn't the last phase; it's designed from the first.

"Where do human control and traceability sit?"

Finally, ask how the system is governed once running: at which points a person steps in —the human-in-the-loop pattern— and how what the system did and why is recorded. That traceability is what lets you audit and account, especially in sensitive processes. A partner who designs governance from the start is thinking about you operating it; one who leaves it "for later" is thinking about continuing to operate it themselves.

The question behind all of them

The four questions measure the same thing from different angles: does this partner's model win when you become autonomous, or when you keep depending on them? That's the line that separates an applied-AI partner from an endless project. Make it explicit before signing and the rest of the negotiation sorts itself out.

How we approach it at Codara

At Codara we answer these four questions the same way every time: an agreed metric before starting, full ownership on your side, handoff from day one and governance with human control and traceability. If you want to ask us these questions about your specific case, let's talk about your AI project.