How to start discovery inside feature factories

Whether you love or hate AI, you cannot neglect it.

9 out of 10 CEOs are demanding that teams ship AI features.

I bet you get annoyed with the push to ship AI for the sake of it, and yet you may benefit from it if you reframe your thinking.

Instead of “Which AI features should we ship?” ask “Which problems can AI solve now that we couldn’t in the past?”

Teresa Torres asked precisely that question, and in a few months, shipped her first AI product, the Interview Coach.

When Teresa had an accident while playing Ice Hockey, she had to stop and stay quiet at home for months. Instead of complaining or doing nothing, she spent those months going from heavy AI user to building her first AI product. Not because she's a developer. Not because she had a team. Because she put her energy and critical thinking into it and started building.

This isn't another "AI will change everything" podcast. This is what one of the world's smartest product people learned building with AI for the first time.

Alloy brought this episode to you. Prototypes that look exactly like your product.

Check it out now

Here are a few valuable links for you:

* Read Teresa’s content for free

* Building My First AI Product: 6 Lessons from My 90-Day Deep Dive

* Connect with Teresa Torres

You can watch this one on YouTube as well:

10 Key Takeaways From Building an AI Product

1. Start With Assumptions, Not Interviews

"Even if you can't test them, even if you never get to the point where you run an assumption test, just doing the exercise of asking what needs to be true in order for this idea to work will help you see the flaws in your ideas."

Most teams want to jump straight to customer interviews. But when business stakeholders own customer relationships (especially in B2B), assumptions become your entry point.

List what must be true for your idea to work. You'll immediately spot the fatal flaws.

2. Discovery Becomes Everything When Delivery Becomes Cheap

"We're going to go through a period where companies think they should build every idea that they have. And then we're very quickly going to realize this leads to terrible products."

If AI makes building features trivial, companies will build everything. Products will become an incoherent mess trying to serve everyone.

The companies that survive will be those obsessed with deciding what not to build. Discovery isn't dying - it's becoming more important than ever.

3. AI as Thought Partner, Not Replacement

"I'm much more a fan of the, it's gonna augment our ability to do our tasks... when I work with it as a thought partner, it's pretty fantastic."

Stop trying to outsource work to AI. Instead, pair with it like you would with a senior colleague. Teresa uses Claude to challenge her writing, suggest improvements, and push her toward better storytelling.

The magic happens in the collaboration, not the delegation.

4. Building AI Products Is Easier Than Ever (But Still Not Easy)

"For people that have high agency, it is easier than ever to build. It's still not easy, but it's easier than ever to build."

Teresa went from never building an AI product to deploying one in production within months. But don't mistake "easier" for "simple." You still need to understand your problem deeply, design thoughtful workflows, and iterate relentlessly (over several months).

5. Single Prompts Don't Scale, Workflows Do

"Most AI products are not single prompt. They're workflows or they're agents where the LLM is doing a variety of smaller tasks and then aggregating them to get a bigger response."

Teresa's interview coach began with a single prompt that evaluated four dimensions. When the AI got confused between criteria, she split it into four separate prompts with specific contexts.

Breaking complex tasks into smaller, focused steps dramatically improves AI reliability.

6. You Need Evals to Build Quality AI Products

"When you push to production, you start to see errors that you just can't identify before you're in production... evals are a very systematic way of looking at what errors you're getting."

Teresa spent more time building evaluation systems than building the initial product. Evals help you measure error rates, test improvements, and maintain quality as you iterate. Without them, you cannot guarantee the quality of the output.

7. Critical Thinking Becomes Your Superpower

"Critical thinking is always the most important skill. It used to be like when we had gatekeepers on most of our content, we could outsource the critical thinking to the editors and the publishers. But we don't have those gatekeepers anymore."

With AI generating an infinite amount of content and eliminating traditional quality filters, your ability to separate signal from noise becomes invaluable.

Don't just consume - evaluate sources, question claims, and go deep on concepts that matter.

8. Go to the Source, Avoid Surface-Level Content

"When you learn about a new idea, if you hear about an idea from somebody where they didn't go deep on that idea, there's some danger in misinterpreting it and really not understanding the core value."

The internet floods us with shallow takes on important concepts. When learning about frameworks like Jobs to Be Done or Opportunity Solution Tree, find the original sources and people who've implemented them extensively.

Surface-level interpretations often miss the point entirely.

9. Use AI as an Additional Team Member

"Claude found opportunities that I did not, which really surprised me, but it missed a lot of opportunities that I found... I absolutely would add AI to my team."

Don't outsource synthesis to AI, but definitely include it as a team member.

Have everyone identify opportunities individually, then compare notes with Claude's analysis. You'll catch opportunities you each missed alone.

10. Pain Teaches Faster Than Success

"I think what's going to happen is AI is going to accelerate those really painful business results. And the silver lining is we're going to get to realizing discovery is even more important faster."

Companies will build too much with AI, confuse customers, and face painful consequences. The survivors will learn that speed without direction is chaos. This pain will create a renaissance in product discovery practices.

While others debate whether AI threatens product management, Teresa built an AI product that helps product managers improve their skills. She didn't wait for permission or perfect knowledge - she identified a problem and solved it.

The future belongs to product people who see AI as their thought partner, not their replacement. Those who use it to amplify their judgment, not replace it. Those who understand that when building becomes easy, deciding what to build becomes everything.

Teresa's broken ankle led to a breakthrough product. What will your constraint create?

Teresa Torres teaches product discovery to thousands through her courses and blog at ProductTalk.org. Her new AI interview coach is available through her continuous interviewing course, and she's launching a podcast called "Just Now Possible" focused on AI product-building stories.

Are You Ready to Advance Your Career?

After 17 years on the road, I crafted the 100X PM Mastermind. It’s the best cohort you can get to help you move from product manager to product leader.

10 live lessons, including unscripted talks, and 30 minutes of free advisory after each lesson. You bring the challenge, we solve it together.

Join our upcoming 100X PM Mastermind.

Here’s what people say about it:

And this is our program to transform you into a product leader.

Have a lovely day, David.

This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit dpereira.substack.com/subscribe