PerfLint puts AI optimization on measured ground

I have shared before the example how I used LLM and Tracy to profile a godot game.
And recently I experimented a lot with automating profiling our Unity game at work with the help of agents.

So I’m always skeptical when an AI agent “optimizes” a Unity project by reading files and guessing or even hallucinating (happened a few times in our team).
PerfLint takes a better route: deterministic rules measure assets, import settings, memory and build size, then expose findings through Unity CLI/MCP so an agent can explain or draft a fix.

The author reports Addressables duplication dropping from 337 assets to 5, and bundles shrinking from 1.29 GB to 805 MB. More interestingly, the tool caught an agent’s false texture diagnosis before it changed anything.

I haven’t tried it yet, take the numbers with a grain of salt, but “measure first, let AI explain” is the direction I want.

Have you tried it before or after this latest update?

https://discussions.unity.com/t/released-perflint-free-local-audit-for-build-size-performance-and-unity-6-migration/1734256

#performance #ai