Introduction

An ankle injury gave Teresa Torres 90 days on the couch - and the chance to build her first AI product. Here are 6 lessons she learned.

What’s the problem it solves?

The article explains how to turn AI experiments into real products that solve customer needs. It shows the mistakes to avoid, how to test early, and why building with AI is a never-ending process.

Quick Summary

Teresa Torres used her recovery time to explore AI deeply. Instead of chasing hype, she focused on where AI could solve real problems in product discovery - like speeding up interview analysis. She learned that picking the right problems matters more than adding “AI” for its own sake.

Her testing process showed that large language models can be powerful thought partners, but not replacements for human judgment. She built several prototypes, moving from simple chat tools to structured workflows. Along the way, she discovered that consistency is tough, and that evals (AI tests) are critical to know if your product is actually good.

Torres also realized building an AI product is not a one-time project. It requires constant iteration, weekly experiments, and careful handling of sensitive user data. She stresses the ethical side: users must know what’s being stored, for how long, and why. In her case, partnering with an already SOC 2-compliant tool solved the data responsibility challenge.

Key Takeaways

• Don’t add AI features just to follow hype - tie them to real customer needs.
• Prototypes should be tested with safe, engaged users first.
• Chatbots aren’t always the best UX - AI can work better in workflows or behind-the-scenes.
• Consistency across different inputs is hard - evals are the only way to measure quality.
• AI products require continuous investment, not one-off launches.
• Data handling is an ethical responsibility - transparency and limits build trust.

What to do

• Start with customer problems, not AI tech.
• Use domain expertise to guide your AI’s context.
• Test with a safe beta group before real customers.
• Break complex tasks into smaller AI steps (workflows).
• Set up evals to measure your product’s quality.
• Expect to keep iterating every week - the work never ends.
• Be upfront about data collection, delete what you don’t need, and consider partners for compliance.

‍