Martins Sipenko had never written a Playwright test before this. He works on backend and platform engineering, and for years end-to-end testing was someone else's job. Then he needed a real suite for his own product, with **no one** around to write it, so he handed the work to coding agents.

He ran Codex and Claude Code against the codebase, wired up the Playwright MCP, and drove the rest from the CLI, coming out with a working end-to-end suite covering the product's core flows.

This talk is an unglamorous report on how that actually went: the parts the agents handled better than expected, the tests that passed for the wrong reasons, the flakiness he had to chase down, and the uncomfortable moment of staring at a green build full of tests he had not really read.

Martins walks through the workflow that survived contact with reality, the rules he ended up enforcing on the agents, and where he thinks a human still has to stay in the loop. It is an honest account from someone who started at zero, not a vendor demo.