When your discovery conversations run through an AI layer, the already-present human tendency to fish for validation gets a systematic multiplier — which means Fitzpatrick's discipline matters more now, not less.
Rob Fitzpatrick's The Mom Test is built on a single uncomfortable observation: people will lie to your face to protect your feelings, so the questions you ask have to make lying irrelevant. That means anchoring every conversation in the customer's actual past behavior, their real frustrations, and the concrete stakes of those frustrations — never in hypotheticals about your product. Compliments are noise. What someone has already done, already paid for, already tried and abandoned — that is signal.
The translation gets sharp fast. If you deploy an agent to run customer discovery at scale, you inherit all of Fitzpatrick's original problem and add a new one: agents optimized on engagement metrics will drift toward exchanges that feel good, because positive responses look like success. You have to hard-code the logic — ask about the last time, the actual cost, the workaround they built — before the agent touches a single user. Commitment signals like time spent, data shared, or a willingness to be contacted again still separate real interest from polite noise, and an agent can be taught to track exactly those.
- Design your agent's interview script around past behavior and concrete costs, not reactions to your idea
- treat any conversation that ends in praise but no advancement as a failed session, regardless of how warm the transcript reads
- have a human review the shape of the questions your agent is actually asking, not just the answers it brings back.
