Greptile launches TREX: an AI code reviewer that actually runs your code

TREX: An AI code reviewer that runs your code

Greptile's TREX moves beyond reading diffs—it runs the code, takes screenshots, and catches runtime bugs. It works as a sub-agent of the main reviewer, sharing context and running parallel investigations to avoid re-scanning the codebase. The post explains that an earlier standalone agent produced irrelevant tests, and a single combined agent got overloaded. The fix was an orchestrator that spawns dedicated TREX agents per issue, each handling setup like auth and feature flags on its own.

Why it matters: Greptile added an execution layer to AI code review — not just reading diffs, but running code, taking screenshots, catching runtime bugs. The post has real architectural lessons and failure modes, hitting H and K. But the audience is narrow and R is absent, so it lands right ...