#63 Teresa Torres - Taste won't save your product

August 10, 2026

Ben

38:52

mins

Listen Now

38:52

mins

Five years after Continuous Discovery Habits, Teresa Torres has done something few methodology authors do: she stopped just teaching the method and started building the tools. In this conversation with Ben, she explains what happened when a product discovery coach picked up AI engineering, why most teams get AI-assisted discovery wrong, and where the line runs between thinking with AI and letting AI think for you.

‍Teresa Torres doesn't buy the "all you need is taste" narrative. She calls it Silicon Valley elitism, and she has a better explanation for what actually separates good product work from bad.

About this episode

The conversation starts with the taste debate. The current Silicon Valley story goes: roles collapse into builder roles, and the ones who win are the ones with taste. Teresa dismantles that framing. A founder's taste doesn't matter, the customer's taste does, and matching it is a learnable skill built from expertise, hard-won customer knowledge, and craft. Not something you absorb by being exposed to tasteful things.

From there it gets practical. Ben describes the AI theater he sees in large organizations: transcripts pasted into an agent, turned into a PRD, dropped into SharePoint, summarized by someone else's agent into three sentences for a meeting. Teresa's answer is blunt: the documentation was never the job. Referencing a recent David Brooks piece in The Atlantic, she draws the line that runs through the whole episode. Some people use AI to think better. Some people use AI to think for them.

The core of the episode is interview synthesis. Teresa is building AI-generated interview snapshots in partnership with Vistaly: upload a transcript, get key moments and opportunities extracted, and after three interviews a first draft of your opportunity solution tree. She explains why this is much more than a prompt. Synthesis needs a point of view, a research question, hallucination checks, framing checks, and correction loops. Ben tests this against his own experiment, a Notion wiki-LLM built to synthesize 27 customer interviews in a single week, and shares exactly where it broke: the LLM took his deliberate provocations at face value and worked wrong data into the synthesis.

The last third covers the build-vs-buy question after the "SaaS is dead" hype, why Teresa built her own task manager but would never build her own blog software, and a surprising symmetry: Ben learned to onboard employees by first learning to give context to AI agents, and Teresa teaches LLMs qualitative synthesis the same way she has taught humans for years.

Key takeaways

The taste narrative is a great-man myth. What matters is not the founder's taste but whether the founder can match the customer's taste, and that comes from expertise and customer knowledge, not exposure.

AI synthesis quality tracks your own synthesis skill. If you're bad at synthesis and dump transcripts into an LLM, you get shallow, surface-level output that confirms what you already knew.

Production AI is a different beast from personal tinkering. Reliable output across hundreds of users means evals, correction loops, and failure modes that keep coming. At some point you hire a tool instead of becoming the expert yourself.

LLMs take confident statements in transcripts at face value. Interviewing techniques like deliberate exaggeration corrupt the data unless the synthesis layer knows about them.

Tacit knowledge doesn't transfer through checklists. Deliberate practice with expert feedback is the only path, and AI can compress the practice loop with simulated interviews and instant feedback.

"I don't think a founder's taste matters. I think their customer's taste is what matters."

"Some people let the documentation become the job. And some people continue to think for themselves."

About the guest

Teresa Torres is a product discovery coach, speaker, and author of Continuous Discovery Habits, which has sold over 135,000 copies since 2021 and is currently the subject of a year-long five-year anniversary book club. Through Product Talk she has taught discovery skills to thousands of product people at companies from early-stage startups to global enterprises. She is now building AI-powered discovery tools, including AI-generated interview snapshots in partnership with Vistaly.