Three kinds of AI tools touch product work. Here is what each one is good at, where each one stops and when to use which.
They recruit participants, run interviews and summarise them. Use one when research itself is the job: a study with a clear question and a researcher who will read the result. They usually stop at the report. Someone still has to connect it to analytics, decide what to change and get it built.
Coding agents
They take a task a person wrote and turn it into code: a feature, a migration, a test suite. Use one when you know what to build and want it built faster. The result depends on how well the task was described. A coding agent does not know that users are confused by a label unless someone tells it.
AI product teammates
A product teammate covers the gap between the two. It listens to users, the team and product data, finds the cause behind a problem, and drafts the fix as a pull request with the evidence attached. Use one for the long tail of small UX problems nobody has time to find. Wirey is an AI product teammate. It lives in Slack or Teams, runs interviews, reads PostHog and opens pull requests in GitHub that wait for your review.
Side by side
| AI interview tool | Coding agent | AI product teammate |
|---|
| Starts from | A research question | A written task | User, team and data signals |
|---|
| Main output | Interview summary | Code | Cause with evidence, plus a pull request |
|---|
| Who decides what ships | Your team | Your team, in review | Your team, in review |
|---|
| Best for | Planned studies | Specified features | Small fixes nobody had time to find |
|---|
When to use which
If you have questions but no time to ask users, start with interviews. If you have a backlog of specified work, use a coding agent. If you keep seeing drops and complaints but nobody has time to work out why and fix it, that is the job of a product teammate. Whatever you pick, keep the same rule for all of them: every change goes through review, checks and branch protection.