January 7, 2026

AI writes tests; devs write roasts

Show HN: KeelTest – AI-driven VS Code unit test generator with bug discovery

AI test bot claims it finds bugs — devs demand receipts

TLDR: KeelTest claims it auto-writes tests and flags real code bugs, with a flashy demo and free tier. The community is intrigued but skeptical, debating how it distinguishes bad code from bad tests, griping about credits, and warning against boilerplate test spam — demanding proof this actually saves time.

KeelTest popped into the spotlight promising to auto-generate Python unit tests in VS Code and even find bugs before production. The demo shows it catching a real gotcha — a function that fails to log a failed notification — while touting a 90% average pass rate and a "quality score: 100%". But on Hacker News, the crowd came ready with popcorn and questions.

The loudest chorus? Skeptics asking how this thing tells a real code bug from a bad test. As ericyd presses, the “why” of a failure is messy — and the community wants proof, not vibes. Another hot thread: unnecessary tests. joshuaisaact warns that large language models love churning out boilerplate “happy path” tests that feel impressive but don’t catch anything spicy. Meanwhile, rcarmo side-eyes with a Copilot comparison and wonders if other AI agents even track dependencies properly.

Pricing drama also bubbled up: the credits system confused folks, with jokes about turning debugging into arcade tokens. Some cheered the free forever tier (install here), others raised eyebrows at the 75% pass rate next to a 100% “quality” badge. Verdict from the crowd: cool pitch, sharp demo, but show us how the bot knows the difference — and keep the fluff out of our test suites.

Key Points

  • KeelTest is an alpha VS Code extension that generates runnable pytest suites for Python and flags source bugs with fix suggestions.
  • The tool uses deep AST-based static analysis and a multi-step verification pipeline; tests run in a sandbox environment.
  • An example run shows 6/8 tests passing (75%), 2 bugs found, and a total runtime of 1m22s, with tests written to a file (test_notifications.py).
  • Plans include a free tier (7 credits/month) and paid tiers (Starter $9.99, Pro $19.99) with features like priority queue, usage analytics, and early access (Pro).
  • Reported weekly-updated stats cite 847 test suites generated, 127 source bugs discovered, and a 91% average pass rate.

Hottest takes

"How does this tool know that the test it wrote indicates an issue in the source vs an issue in the test?" — ericyd
"it'll write you a ton of boilerplate happy path tests that aren't wrong, per se, they're just pointless" — joshuaisaact
"Weird. Copilot knows what tests are" — rcarmo
Made with <3 by @siedrix and @shesho from CDMX. Powered by Forge&Hive.