I’ve been having a lot of fun playing around with the Playwright MCP server in Claude Code, as a way to conduct what I’m calling “self-driving product discovery and QA”.
In simple terms, throw an AI agent into your product, have it navigate and click around around, and then document what it see’s and learns along the way.
I’ve experimented with a bunch of interesting use cases here:
Basic QA/testing (evaluate these test cases)
Self-documentation (read the docs, use the feature, then improve the docs)
Accessibility reviews (go through this flow, evaluate against standards X and Y)
As with everything in this space, it’s a tad clunky and slow today, but I think this is a fascinating window into the future of product development and management.
Just today, whilst testing this out on a few feature we merged into our staging environment today, this very workflow uncovered a UX quirk in our product that I frankly had never considered.
I watched the agent try to do something in our product, and get stuck. After a few seconds, it figured it out, but it reminded me that, as b2b product leaders, there is no better way to assess usability and intuitiveness, than throwing fresh pairs of eyes at your software.
It just so happens, an AI agent is a very cheap and effective way to do just that, at scale, all the time, for every feature and workflow in your entire product.
