Stuff I'm building around testing with AI. Plus a guesthouse in the mountains where I try AI with my own money.
AI proposes, a human decides.
You give it a requirement and AI drafts acceptance criteria, test scenarios and ready-to-run Playwright tests. I built it because I'd seen plenty of generated tests that pass and check nothing. So instead of pass rate, it tracks how much you can trust each test.
+ alsoThere are human gates between the steps. Nothing moves on until QA says yes.
One set of QA rules for people and agents.
A set of QA rules where every rule has its own ID. Agents in Claude Code and Copilot read them and cite them, the way a colleague would in a code review.
+ alsoWhen an agent says which rule a test breaks, nobody has to ask why.
Even a PM can order a load test.
A framework on top of k6. A product manager who's never seen a script can order a load test with it. They answer five questions, or leave it all to the AI configurator.
+ alsoAt the end they get a report they can read without a translator.
An AI sidekick for manual testers.
A Chrome extension that sits next to a manual tester. It suggests what else to try and helps write up a finding.
+ alsoIt has one rule it can't skip: no claim without evidence. Every finding needs a screenshot, a log or an exact step.
A Master Test Plan that comes out the same every time.
A Master Test Plan generator. Code builds the structure, scope and rules from the input data, so the same data gives you the same plan.
+ alsoAI is only allowed to write the prose, like the intro and the summary.
A mountain guesthouse as my lab.
We run a family guesthouse in the Krkonoše mountains. I collect pricing and occupancy data and see what AI can get out of it. Bit by bit I'm automating the routine around bookings and messages to guests.
