indiehackerssoftware-testing
Use a different model to generate tests for AI-written code
At a glance
A builder assigned implementation to Claude and test generation to Gemini during an application audit. This workflow is listed under software-testing, cross-model-review, application-audits.
- Agents / tools
- Claude, Gemini, FastAPI, Next.js, PostgreSQL
Building a tool for these workflows? See weekly rates.
What they did
The author used Claude to write code and Gemini to produce tests independently for a prompt-generation platform. The broader audit also changed authentication, request limits, migrations and router structure.
How it went
The author reports edge cases they would have missed and an increase from 57 to 137 passing tests after the audit.
Worth knowing
The test count followed several implementation and security changes; it does not isolate the effect of using a second model.
Save ideas in this browser. Upvotes count across the board.