Don't Use Generative AI to Replace Discovery with Real Humans

Teresa Torres argues against generative AI tools that try to replace real discovery work, such as synthetic customer interviews, auto-generated opportunity solution trees, or AI-drafted artifacts. She explains that discovery exists to increase the hit rate for creating customer and business value, and that artifacts are by-products of thinking, not the goal. Using ChatGPT examples, she shows that AI-generated personas produce generic, non-actionable, fictional stories that cannot build empathy or differentiation. She offers four guidelines: keep talking to real humans, don't outsource thinking, use AI for analysis but verify its output, and treat AI as one team member. The piece matters because it draws a clear line between efficiency and the competitive advantage that continuous discovery builds.

01Key takeaways

• Discovery's purpose is raising the hit rate for customer and business value, so judge work by the thinking it produces, not the artifacts.
• Synthetic interviews yield generic, non-differentiating insights; keep continuous interviews with real customers as a competitive advantage.
• Avoid letting AI generate the first draft of team artifacts, since it shifts the conversation and invites social loafing.
• Use AI for tedious analysis such as theming and categorizing data, but collaborate iteratively and verify counts, duplicates, and calculations.
• Form individual perspectives before team discussion, then treat AI as one more team member whose output is weighed against others.

02Key sections

Why discovery exists

Discovery aims to raise the odds of creating customer value and business value, structured around a clear outcome, customer opportunities, and solutions that satisfy both. Artifacts like opportunity solution trees are tools for thinking and alignment, not the end product.

Don't replace real humans with synthetic ones

Interviews uncover specific, real past behavior that AI-generated personas cannot reproduce. Torres demonstrates through Netflix examples that synthetic responses are generic, sanitized, and shared by competitors, so they offer no differentiation.

Don't use AI to replace thinking

Generating artifacts with one click skips the examination and team alignment that discovery depends on. AI-produced first drafts let the loudest voice or the tool dominate, encouraging social loafing and eroding diverse team perspectives.

Use AI for analysis, but check its work

Torres shares how she used ChatGPT to theme open-ended survey responses and analyze LinkedIn engagement, iterating collaboratively and verifying counts and math. The approach saved significant time while keeping her actively engaged in the reasoning.

Treat AI as a team member

Following a pattern of individual work before team discussion, AI should be one perspective integrated with others rather than a source of final answers. Thinking must happen first, with AI augmenting rather than replacing team judgment.

03From the post

“I’m disappointed to see the rise of generative AI tools that are designed to replace discovery with real humans. Don’t get me wrong. I’m a big fan of generative AI. I use it daily in both my personal life and at work. But when we use generative”

04Frameworks mentioned

Summary and takeaways written by PM Atlas; quotes are short excerpts. © the original author.