Code Documenting and Testing with AI | AI Talk - Product Makers Romania

Pedro Dobrescu, entrepreneur and product manager at ShortPixel, and "not a QA engineer", told how automated testing finally happened on a WordPress plugin he'd lived with for years. One earlier attempt took six months with a part-time engineer and ended with a single dummy test running. With Claude Code, three weeks got him more than 2,000 tests running. In this session, we went through: Pedro's three steps: let the AI document the code first, then build the test environment, then write the tests, from unit tests up to browser tests A complete feedback loop for the agent: everything in Docker, one command for every test, a run that goes red if nothing ran, and zero retries, because a flaky green is worse than an honest red Pin, then flip: bugs become tests, then permanent regression tests, so Claude never changes the product code The rules he gave Claude Code after it started committing on its own, and why every release still gets a human test pass The CI in GitHub Actions and the AGENTS.md, CLAUDE.md and TESTING.md files behind it A live demo on a small plugin with no tests: one phased prompt, Claude asking for decisions before each phase Questions from the room: skills vs markdown files, docs that eat your tokens, a markdown file per folder, and writing business rules down so the agent doesn't "fix" them 3 in 3: trusting AI within boundaries, what he still does by hand, and the newsletter he reads Tools discussed during the session: Claude Code, Docker, GitHub Actions, PHPUnit, WP-CLI, Composer, Codex Hosts: Pedro Dobrescu -- product manager at ShortPixel, entrepreneur since 2011 Ady Rugină -- creator of the AI Talk series, host of this session, Product Makers Romania Andreea-Ruxandra Dumitru -- AI Talk organising team, co-host of this session, Product Makers Romania Join our community: https://productmakers.ro