Teresa Torres

Product discovery coach; author of Continuous Discovery Habits; advocates for weekly customer interviews as a non-negotiable product practice.

Last updated: 2026-04-13

Overview

Teresa Torres is a product coach and researcher best known for formalizing continuous discovery — the practice of maintaining a weekly cadence of customer interviews throughout the product development cycle, not just during dedicated research phases. She writes at producttalk.org and authored Continuous Discovery Habits (2021).

Her empirical work includes interviewing real product teams to surface what actually works in practice, as opposed to what frameworks prescribe.

Key Ideas

• Continuous discovery: product teams should do at least one customer interview per week, every week — not as a project, as a habit
• Opportunity Solution Trees: structured maps that connect customer opportunities (unmet needs, pain points) to solution ideas to experiments, keeping the customer problem as the root of all decisions
• Stay close to customers for iteration: teams that succeed with AI products don’t just ship and watch data — they maintain direct customer contact to interpret why things work or don’t
• Domain expertise as a discovery shortcut: in her AI product research, teams with deep domain knowledge (former teachers, SREs, government officials) made better product decisions faster — domain expertise is a substitute for some discovery work, not all of it

On AI Product Teams (producttalk.org research)

Torres interviewed 9 AI product teams in production. Key findings that relate to her discovery philosophy:

• Small cross-functional teams (2–3 people) ship faster — aligns with Zhuo, but Torres adds that boundary-spanning experience (not just generalism) is the key trait
• Domain expertise drives architecture decisions — a former teacher designs better eval rubrics; a former SRE encodes better debugging patterns
• Starting narrow is a discovery practice: beginning with one specific use case and watching real customer behavior before expanding is effectively continuous discovery at the product level
• Getting AI to say “I don’t know” is an unsolved challenge — LLM positivity bias produces confident wrong answers; teams combat this through architectural uncertainty management

Relationship to Zhuo / Prototype-and-Prune

Torres partially validates Zhuo: small teams, rapid iteration, and narrow focus are confirmed by real AI teams. But she implicitly answers the customer gap in Zhuo’s framework — the teams that succeed stay close to customers throughout, not just via post-launch pruning. Domain expertise is load-bearing where customer access is limited.

See prototype-and-prune for the open question Torres helps address.

Relationship to Rabois / Anti-Customer-Feedback

Keith Rabois argues customer feedback is actively harmful for consumer and SMB products because consumers make subconscious decisions they can’t accurately describe. Torres argues for continuous customer contact throughout the product cycle.

The tension is more apparent than real when you look at what each is actually prescribing:

• Rabois objects to survey/interview-style feedback on product direction for consumer decisions that are subconscious — asking “would you use this?” or “why did you buy that?”
• Torres prescribes discovery interviews that explore customer problems, contexts, and behaviors — not asking customers what to build

However, Rabois’s underlying point (customers can’t articulate what they want) is a real challenge for Torres’s interview method too. Her response would be that continuous discovery is about understanding problems and contexts, not gathering feature requests. The weak form of her method is exactly what Rabois is criticizing.

See keith-rabois for the counter-position.

Connections

• prototype-and-prune — her research validates small teams + narrow start; her philosophy challenges the “ship and prune” substitution for customer discovery
• product-development — empirical counterweight to Zhuo’s prescriptive framework
• agent-evaluation — her AI team research documents the eval progression pattern (spreadsheets → assertions → LLM-as-judge)
• keith-rabois — counter-position on customer feedback for consumer products

Sources

• Learning from 8 AI Product Teams — added 2026-04-12