Critique, not commentary.
Chatbots comment. Deskcrit critiques:
- Severity
- ranked, never listed
- Evidence
- measured from your pixels
- Confidence
- honest about uncertainty
- The fix
- actionable as written
Both buttons share the same fill, size, and weight, so the screen offers no default path. At the moment of decision, equal-weight choices measurably slow completion.
primary actions · identical fill · identical height
Pick one primary: keep a single filled button and demote the other to an outlined style — whichever converts better wins the fill.
How it works.
- 01ReadsParses your screen into a context model — type, platform, intent — not just pixels.1 screen · context
- 02ReasonsThinks out loud through heuristic and craft passes. Visible, interruptible — no spinner theater.NN/g 10 · WCAG
- 03ReportsFindings ranked by severity, cited against the pixels, scored for confidence, closed with the smallest concrete fix.severity · evidence · confidence
- 04DefendsChallenge any finding. It engages your actual argument and resolves it on the record.defended · conceded · revised
Designers post screens into Slack and get emoji back.
Honest, senior critique is scarcest right before you ship. I spent twenty years reviewing screens for high-stakes platforms; Deskcrit encodes that judgment into an agent that holds its position under cross-examination. Transparently, it's also a portfolio artifact — the way this agent argues is my work sample.