How to choose an AI implementation partner: 12 questions that separate builders from resellers
The market is now flooded with agencies claiming AI implementation. Twelve questions will tell you, in an hour, whether the person across the table can actually build.
Between 2024 and 2026 the number of firms describing themselves as AI implementation partners went up by an order of magnitude. Perhaps one in ten of them can actually ship production software. The rest are reselling other people's tools, running strategy decks, or subcontracting the real work to a distant team you will never meet. What follows is the diligence we would apply if we were the buyer.
Twelve questions to ask before you sign
1. Who on your team writes the code?
Named humans. Not "our engineering practice" or "our partners." If the builders don't have names, they are not the ones being sold to you.
2. Show me three production systems you shipped this year.
Not slides. Not case studies. Live URLs, logins, or screen shares. A real firm can walk through what they shipped, why it was built that way, and what went wrong. If they can't, walk out.
3. What's your rate of "we decided not to build this"?
Firms that never say no are order-takers. The good ones talk clients out of a build at least a quarter of the time.
4. Who owns the code when the engagement ends?
You do. Full stop. Any answer involving "licensed access to our platform" is a subscription contract, not a build.
5. What's your stack, and why?
Not a list. A conviction. Firms with an opinion are firms that ship. Firms that say "we're flexible with any stack" usually don't have one at all.
6. What does your first deliverable look like?
Ask for the shortest cycle they've ever run — day one to first working thing. If the answer is measured in weeks, delivery is not the culture. See our note on why software delivery got fast .
7. How do you handle data we can't share with third-party models?
If they can't articulate a coherent private-model story, they have not built anything serious for a regulated buyer.
8. Who monitors the systems after they ship?
Delivery is easy. Uptime is the work. Ask for their alerting, logging, and on-call posture. Vague answers are a red flag.
9. What is your team size, and what is the ratio of builders to salespeople?
In good firms, most of the roster ships. In bad firms, the sales team is larger than the engineering team.
10. Have you turned down a client this year, and why?
The answer reveals whether they have a real philosophy about fit — or whether they will take any file, whether or not they can serve it.
11. What's your average client tenure?
Firms whose clients renew for three years are doing work that lasts. Firms whose clients disappear after six months are shipping demos.
12. Would you take this call again for free in six months?
The answer to this one is not the answer. It is how they answer. A real partner is happy to say yes and mean it. A reseller will get quiet.
Firms that never say no are order-takers. The good ones talk clients out of a build at least a quarter of the time.
Where we sit
VX-N is a small firm on purpose. The people who diagnose the file are the people who write the code. If the twelve questions above sound like table stakes to you, we should probably talk. If they sound like a foreign language, read the note on what an AI consultant actually does first, and then come back.
If any of this maps to a decision you're weighing, a call is the fastest way to know whether we're the right people to build it. If we're not the fit, we'll say so.