13.8Interview Talking Point

"Our reporting captures three things for every test: a command log for reproducibility, screenshots for visual debugging, and structured JSON for programmatic analysis. On failure we also keep the YAML accessibility snapshot and the URL, so the agent can re-analyze the failure semantically — and for runner and healer sessions we archive the Playwright trace, which since 1.60 includes HAR network capture. That trace is the audit trail: when someone asks what the AI actually did in the browser, we replay it. The key insight is that AI-driven tests need richer failure artifacts than traditional tests, because the agent may need to reason about the failure to determine if it's a real bug or a test environment issue — and the disk-first CLI gives us most of those artifacts as a side effect of the agent working."