Should you work with us — or with one of these?

The questions worth asking any AI partner

These apply whoever you hire, including us. Most of them are uncomfortable to answer badly, which is exactly why they're useful on a first call.

“What do we own at the end?”

Ask specifically about prompts and the eval suite, not just application code. In an AI build the prompts are a real part of the product, and a handover without them leaves you unable to change behaviour without going back to the vendor. Get the answer in the contract rather than the conversation.

“How will we know if it's working in six months?”

The answer should involve a test suite that someone runs on a schedule. If it involves a dashboard of usage metrics, that measures whether people clicked it, not whether it was right. Those are very different things and only one of them is quality.

“What happens when it's confidently wrong?”

Every AI system is wrong sometimes; the design question is what the system does about it. Look for explicit refusal paths, escalation to a human, and logged reasoning you can replay. A partner who talks only about accuracy percentages hasn't thought about the tail.

“What would make you tell us not to build this?”

Anyone who can't name a scenario where the honest answer is “don't” is selling rather than scoping. The usual real answers are that the data isn't reachable, the volume is too low to justify the build, or the process needs fixing before it's worth automating.

“Who does the work?”

Ask whether the people on the call are the people who will build it. This is the single most common gap between what a pitch promises and what a project delivers, and it is very easy to check up front and very expensive to discover later.