AI Output Evaluation

Expert evaluation for Contra Labs, assessing AI-generated design output as a working designer rather than a general reviewer.

Most evaluation of AI output stops at whether something looks finished. That is the easy half, and the half a model is best at faking.

The harder question is whether it works. Whether the hierarchy leads the eye where the argument needs it to go. Whether the type is doing a job or filling a space. Whether a layout that scans cleanly is actually making a point, or quietly making none.

Twenty years of building this material by hand is the difference between "this feels off" and naming the decision that broke it. A note saying something looks cluttered is not actionable. A note identifying the hierarchy choice that caused it is.

This engagement is covered by an NDA. No client, product, output, interface, criteria or methodology appears here. Everything above is my own background and approach, brought to this work rather than learned from it.

If you are evaluating AI output and want a reviewer who has actually made the thing being imitated, that is what I do. Fixed scope, fixed deadline, delivered on schedule.