The AI consulting market has exploded — and with it, the number of profiles selling vision without ever having shipped a single working line of code to production. Here’s how to sort them out before signing.
1. “Show me a project you’ve shipped, not a demo”
A demo proves a concept works in a controlled environment. A shipped project proves it holds up against real users, real data, real edge cases. Ask for concrete examples — architecture, technical decisions, measured results.
2. “How do you evaluate whether an AI agent actually works?”
If the answer revolves around “we test and adjust,” be cautious. A serious consultant talks about evaluation datasets, precision metrics, regression tests. Without that, you’re flying blind once in production.
3. “What can AI NOT do in my case?”
A good consultant knows how to say no. If every idea gets validated without nuance, that’s a red flag — either they don’t understand your constraints, or they’re telling you what you want to hear rather than the truth.
4. “Who actually codes the project?”
Many AI consultants lead without ever touching code — which can work fine if a technical team is already in place. But if you’re looking for someone who understands the real constraints of LLMs in production, a profile who codes themselves has a direct advantage: they validate feasibility before selling you a roadmap.
5. “What’s your plan if the first prototype doesn’t work?”
The answer should be concrete: short iterations, regular validation checkpoints, success criteria defined upfront. A vague answer signals a lack of method.
What to remember
The right instinct isn’t to look for the most impressive talker — it’s to look for someone who asks these questions themselves before you have to.
Stéphanie Caumont
AI Product Owner · Learn more