The five mistakes we see in every first AI project

After 40+ first AI projects across SMBs, the failure modes are remarkably consistent. (1) Pick a use case nobody asked for. (2) Skip the eval harness. (3) Underestimate the integration work. (4) Hand it off to a junior. (5) Declare victory before you have a month of production data.

Avoid all five and you are already ahead of 80% of the market.