Ask almost any product team why they don’t talk to their users more, and you’ll get the same answer dressed a dozen ways: we don’t have time. Notice what that answer is not. It is not “we’ve run out of things to ask.” Every team is drowning in open questions — why people churn in week two, which half of the onboarding they skip, what the feature they requested actually means. The questions are free and endless. What’s scarce is everything that has to happen around a question before it turns into an answer.
That’s the thing worth being precise about, because it decides what you build. AI User Interviews is built on a single claim: the bottleneck in user research was never the questions. It was the calendar.
What a single interview actually costs
Trace one user interview end to end and the interesting part isn’t the twenty minutes of conversation. It’s the tax around it.
First you recruit — post the screener, chase the responses, filter the people who aren’t a fit, offer the incentive. Then you schedule, which means threading one live human’s availability through another live human’s availability, across time zones, minus the no-shows and the reschedules. Then you sit through it, in real time, one at a time, because a person can only moderate one conversation at once and can’t be in two rooms. Then you write it up — the recording, the notes, the tagging, the quotes pulled by hand — and only after all of that does the interview become something the team can actually use.
The conversation was the cheap part. The scheduling, the moderating, and the synthesis were the expensive part, and they scale the worst way possible: linearly, in a person’s hours. Ten interviews aren’t ten times the insight — they’re ten times the calendar, ten times the sitting-in-the-room, ten times the write-up. So teams do the rational thing. They run five interviews instead of fifty, and they run them late, once the decision is mostly made, as confirmation rather than input. Not because they stopped being curious. Because the tax got too high.
Move the cap, not the questions
Once you see research as calendar-bound rather than question-bound, the design problem inverts. You stop asking “what should we ask?” — teams are already good at that — and start asking “what would it take to make the hundredth interview as cheap as the first?”
Three costs have to fall for that to happen, and they’re exactly the three that AI User Interviews takes on:
- Scheduling → gone. If nobody has to be in the room at the same moment, there’s no calendar to thread. Interviews stop being appointments and become something a study just runs.
- Moderating → parallel. A person can hold one conversation at a time. Remove that limit and a hundred interviews can happen at once instead of one after another. The wall between “a few interviews” and “a lot of them” was always a single moderator’s throat and attention.
- Synthesis → immediate. The write-up was where insight went to die three weeks late. Reading across every transcript as it lands — the recurring theme, the sharp quote, the place people split — turns a pile of recordings into findings the same afternoon, while the decision they inform is still open.
The interviews were never the hard part. The queue in front of them was. Take the queue away and “talk to more users” stops being advice nobody can afford and becomes something you just do.
None of that touches the questions, and that’s the point. You still write the discussion guide. You still decide who you’re talking to and what a good answer looks like. The machine takes the part that was mechanical and expensive — the recruiting queue, the moderator’s finite hours, the manual tagging — and leaves you the part that was always the actual work: deciding what to ask and what the answers mean.
Being honest about what it is, and isn’t
We’d rather show the seam than paper over it — the same discipline that keeps code-results from calling a correlation a proof. So: a moderator that never gets tired and never runs out of hours is a genuine change in what a team can afford to learn. It is not a replacement for a researcher’s judgment. It doesn’t decide which questions matter, it doesn’t know your business, and it won’t tell you what to do with what it hears. Those stay with the person who wrote the guide — the AI runs the study faithfully at a scale a human never could, and hands the thinking back.
There’s a failure mode we designed against, too: cheap interviews invite lazy ones. When talking to a hundred people costs about what talking to one used to, the temptation is to skip the discipline — vague questions, no hypothesis, a fishing trip. Scale is a multiplier, and it multiplies a bad guide just as happily as a good one. The tool makes the interviews cheap; it can’t make them worth running. That part is still on you, and we think it should be.
Why this is the shape of the studio
Every product we build tends to reveal the same move underneath it — find the expensive step everyone treats as fixed, and question whether it has to be. Thyseus decided compute didn’t have to be a data center. ganttchart.ai decided the plan didn’t have to be a spreadsheet, just the ten minutes before a meeting. AI User Interviews decides that talking to your users didn’t have to cost a calendar.
That’s the throughline of a studio that only builds what it runs: the interesting problems are rarely the loud ones out front. They’re the quiet, assumed-fixed costs sitting just behind the thing everybody’s staring at. Here, everyone was staring at the questions. The bottleneck was standing behind them the whole time, holding a calendar.