"Building is cheap now, but don't build everything": Teresa Torres on AI product management
Most product leaders are experimenting with AI. Teresa Torres spent three months building with it, full-time, from scratch, with no engineering background, and ended up shipping a production-grade AI product.
The product is the Interview Coach, an AI-powered feedback tool that replaced the live cohort course she had been running since 2017. In our latest webinar, one of the leading voices in product discovery and author of Continuous Discovery Habits shared the full story: what worked, what failed, and what she would do differently.
Her journey reveals that building with AI isn't just about adding a feature; it's about a fundamental shift in how product teams architect, synthesize, and define quality. Here are her five sharpest takes from the session.
can support your team.
Book a demo
1. "Building is very cheap now. That doesn't mean we should build every idea we have."
Product managers might feel tempted to treat low-cost building as a green light. If spinning up a prototype takes an afternoon instead of a sprint, why not just try everything? But Teresa's argument is that speed of execution has never been the hard part of product development. Deciding what to build has always been the constraint, and that hasn't changed.
If anything, cheap building makes discovery more important, not less. The faster you can ship the wrong thing, the more damage you can do. Product leaders who equate "we can build it faster" with "we don't need to validate first" are setting their teams up for a different kind of waste: Not wasted engineering time, but wasted direction.
2. "I see people write on LinkedIn, 'How are we supposed to train our customers to expect mediocre responses from LLMs?' That's not your job. Your job is to get good responses from the LLM."
There's a version of AI product development where the strategy is essentially “Lower the bar and call it a feature.” Teresa Torres is not interested in that version.
Her point is about accountability. If your AI feature produces inconsistent or low-quality output, the answer isn't to explain that away to users, but to fix the underlying problem.
Teresa suggests different ways of keeping high quality standards:
Smarter – not longer – prompts: Instead of writing “mega-prompts,” orchestrate a series of small tasks that build up to the big task, to keep the LLM’s context window from filling up too quickly.
Evals: “We can put evals in place as guardrails to tell our models, ‘You cannot show this to a customer if it doesn't match this criteria.'”
Feedback loops: “As people move into agents, we need this reflective loop of ‘do something, evaluate it, do something, evaluate it.’ Can the agent manage those cycles itself so that you're improving the overall quality?”
None of this is easy. But the teams doing it well are the ones building products that actually earn user trust over time.
3. "The most common antipattern right now is lazy AI interview synthesis."
This one landed hard, and it's worth unpacking.
The workflow Teresa Torres is describing (and warning against) goes something like this: Conduct a customer interview, dump the transcript into an AI tool, read the summary, and move on. It feels productive. It saves time. But it's probably capturing about 5% of what was actually in that conversation.
Teresa's approach is the opposite: Do your own synthesis first, then ask AI to do it, then compare. Every time she does this, she catches things the AI missed. Every time, the AI catches things she missed. The value isn't in replacing the synthesis step; it's in having two different perspectives working on the same material.
For product leaders: If your team is running customer interviews but outsourcing the analysis entirely to AI, you're not doing discovery. You're generating summaries.
4. "If you're building on the edge of what's just now possible, you have to live on that edge."
Teresa challenged how many product leaders are engaging with AI: periodically and cautiously.
Her case is that you can't develop a genuine intuition for what AI can and can't do through occasional experiments or secondhand reports. You have to use it every day, at the edges of what it's capable of, and pay attention to where it struggles. That's how you build what she calls a "builder toolkit”: A real understanding of context windows, task decomposition, evals, and the patterns that separate reliable AI systems from unreliable ones.
Her own experience mirrors that. Starting with no IDE experience and no GitHub knowledge, she went from zero to shipping her AI Interview Coach in roughly three months, using Claude and ChatGPT as tutors along the way.
Teresa isn’t suggesting everyone needs to become an engineer. But product managers who aren't regularly working with these tools will find themselves increasingly out of their depth as AI becomes the default substrate for everything they build.
Want to hear Teresa Torres expand on all of this, including the full story of how she went from zero engineering experience to shipping a live AI product in three months? Watch “Behind the build: Teresa Torres’ first AI product” on demand.
Francisca Berger Cabral
how airfocus can support your team.
Book a demo
Read also
Explore how airfocus can support your team
Explore how airfocus can support your team