How to Run Customer Discovery with AI Agents

Most solo founders skip customer discovery. Not because they think it's unnecessary, but because the work around it is time-consuming: writing interview guides, recruiting participants, taking structured notes, and then analyzing patterns across 15 different conversations. That's a part-time job before you've built anything.

Running customer discovery with AI agents removes most of that overhead. Not a watered-down version. The actual process.

What Customer Discovery Involves

Customer discovery is the process of talking to real or potential customers to understand the problem you're solving, how they currently deal with it, and what a better solution would look like to them.

Done before you build, it tells you whether the problem is real, how people talk about it, and who has it badly enough to pay. Done after you build, it tells you why adoption is slow or which features matter most.

The process has four distinct parts: writing an interview guide, recruiting the right participants, running the conversations, and analyzing patterns across the responses. AI agents don't replace the conversations. You still get on calls or send surveys. What agents handle is everything around those conversations.

How to Run Customer Discovery with AI Agents

This is a working process you can complete in under a week.

1. Write your interview guide: Start with a Research Specialist from the Specialized department. Give it your product idea, your target customer type, and the specific assumption you're testing. It will produce a structured guide with 8-12 open-ended questions, follow-up probes for each, and a clear objective per section. This takes 10 minutes instead of 3 hours. Ask it to write questions that reveal behavior, not opinions — "What did you do last time this happened?" rather than "Would you use a tool that fixed this?"

2. Create a participant screener: Before you schedule a single call, confirm that participants match your target. Ask the Research Specialist to generate a short screener survey (5-6 questions) that filters for the right profile. You can share this via email or a simple form before booking time.

3. Build a note-taking template: Ask the Research Specialist to produce a structured note-capture template matched to your interview guide. One row per question. Space for exact quotes. A field for surprises or off-script moments. Consistent notes across 12 conversations are far easier to analyze than freeform scribbles.

4. Run the calls yourself: This is the part agents can't do. Twenty minutes per person, following the guide, filling in the template during or immediately after each call. Don't summarize in real time. Capture what people actually said.

5. Analyze the patterns: Once you have 8-12 completed templates, paste them all into a new session with the Research Specialist. Ask it to identify recurring themes, rank the most common pain points, and flag responses that didn't fit the pattern. What takes a researcher half a day to synthesize takes about 15 minutes here.

6. Write a discovery brief: Ask the Research Specialist to produce a one-page brief summarizing what you learned, which assumptions held, which were wrong, and what to build or test next. This document becomes the artifact you reference in every product decision that follows.

A Real Example

You're building a project tracking tool for freelance designers. Your assumption is that they lose time chasing clients for feedback approvals.

You open a session with your Research Specialist and ask it to build an interview guide for testing whether freelance designers struggle with feedback loops. It produces 10 questions covering current workflow, where delays happen, what they've already tried, and what a working solution would look like.

You run 12 calls over two weeks, 20 minutes each. You paste your 12 completed templates back to the Research Specialist and ask: "What are the top three pain points these designers share, and which of my starting assumptions were wrong?"

The analysis comes back. The biggest pain isn't chasing feedback. It's that feedback arrives across five different channels simultaneously — email, Slack, PDF comments, voice notes, and direct messages — and consolidating it before acting takes 30-40 minutes per revision round. Your original assumption was wrong, but you now have a more specific and more painful problem to solve. And you found it before writing a single line of code.

That's what discovery is for.

Common Mistakes

Running discovery after you've already built. Discovery is most useful before you commit to a direction. Running it afterward means you're validating decisions already made, which is harder to act on and easier to rationalize away.

Writing leading questions. If your guide asks "How frustrated do you get when clients go quiet?" you're not learning anything — you're confirming a bias. The Research Specialist defaults to open-ended, behavior-focused questions. Check any questions you add yourself against the same standard.

Treating 3 calls as enough. Three conversations might give you a feeling. They won't give you a pattern. You need at least 8 completed interviews before recurring themes become meaningful. Themes that appear in 3 out of 5 calls may disappear in the next 10.

Skipping the synthesis step. Notes are not insights. The analysis pass, where you look across all responses at once and ask what repeats, is where learning actually happens. Don't skip it because you're tired from the calls.

Optimizing for confirmations. If every call confirms your hypothesis, your screener is probably selecting for people who already want what you're building. Push the Research Specialist to include questions designed to surface the cases where your solution wouldn't help.

Bottom Line

Customer discovery doesn't have to be a research project you defer until you have more time. With a Research Specialist handling the guide, the screener, the note template, and the analysis, you spend your time on the part that requires a human: the actual conversations.

That's roughly 80% of the administrative work removed. The 20% that's left is the part that produces the insight.

If you're pre-product, run 10 calls before you build anything. If you're post-launch and growth is slow, run 10 more. The Specialized department has the agent that handles everything around those calls. You just have to show up for them.

---

Ready to put this into practice? Browse the departments and start with whichever handles your biggest current bottleneck.