Engineering at Meta logoEngineering at Meta
|AI

The Death of Traditional Testing: Agentic Development Broke a 50-Year-Old Field, JiTTesting Can Revive It

2026-02-11
4 min read
0

Endigest AI Core Summary

This post introduces Just-in-Time Tests (JiTTests), an AI-driven testing approach designed to keep pace with agentic software development.

  • Traditional testing relies on static, manually authored test suites that require constant maintenance and struggle with false positives as code change velocity increases.
  • JiTTests are generated on-the-fly by LLMs the moment a pull request is submitted, targeting regressions introduced by a specific code change.
  • The process infers the intent of a change, creates mutants (deliberately faulty code versions) to simulate failures, then generates and runs tests against those mutants.
  • Rule-based and LLM-based assessors work together to filter out false positives and surface only true regression signals.
  • Tests are not stored in the codebase, eliminating maintenance overhead and requiring human review only when an actual bug is caught.
Tags:
#DevInfra
#ML Applications