Resources
Playbooks, guides, and best practices for AI-native E2E testing.
Definition of Done for AI-Generated Code
The classic definition of done assumed the person who wrote the code and the person who confirmed it worked were the same trustworthy human. When an agent writes the code, writes the test, and reports the result, done collapses into self-report. Here is how to split it back into three checkable states.
Spec Project: Taking a Feature From PRD to Test Evidence
Most spec-driven workflows stop when the agent reports the code is written. Spec Project is an agent skill that carries one feature through specification, implementation, and evidence, keeping the spec, the code, and the test report in agreement as the product changes.
What Does an Agent-Run Test Suite Actually Cost Compared to Plain CI?
Somebody on Hacker News asked the honest version of this question: is an agent-driven suite an order of magnitude more expensive than my existing CI running Playwright tests? The answer depends on one variable most people never measure, and the arithmetic is simple enough to do at your desk.
AGENTS.md: Using It as a Testing Contract, Not Just Style Notes
Most AGENTS.md files are house style: package manager, commit format, folders to avoid. The higher-value use is a testing contract that tells the agent how to run the suite, which flows are critical, and what it must never do to get green. Here is what to put in it, a worked example, and how to check the agent is honouring it.
Do AI-Generated Tests Actually Catch Bugs?
Sometimes, and you do not have to guess which. Coverage percentage says nothing about whether a suite would notice a broken feature. Mutation testing does: break the code deliberately and count how many breakages the tests report. Here is how to run that measurement and what to do with the answer.
Playwright's Test Agents: Planner, Generator and Healer
Playwright now ships three first-party test agents, installed with npx playwright init-agents. The planner explores your app and writes a markdown plan, the generator turns that plan into test files while checking selectors live, and the healer repairs failures until they pass or guardrails stop it. Here is what each one actually does, where it fits, and how the approach compares with tests that resolve intent at run time.