UX audits in the agentic era

Classic UX auditing has a stable toolkit: heuristic evaluation, UX laws, severity scoring. But when products start shipping agentic surfaces (copilots, autonomous flows, generative UI), auditors face failure modes that Nielsen never wrote about.

New friction, old laws

Hick's Law says more choices slow decisions. An agent offering '12 suggested next actions' is Hick's Law wearing a new costume. Fitts's Law governs pointing at targets; a confirm button that appears in different places depending on generated content violates it unpredictably. The laws didn't change. The surfaces did.

What we now score for

Our audits weight several agent-specific dimensions: recoverability (what happens after the AI is wrong), expectation setting (does the interface communicate confidence honestly), interruption cost (how expensive is it to stop and correct), and trust calibration (do users over-trust or under-trust based on early interactions). Each maps to concrete, fixable interface patterns.

The audit is more valuable now, not less

Teams shipping AI features fast accumulate friction faster than ever. A structured audit catches the compounding issues (confusing empty states, missing undo paths, silent failures) before they calcify into 'the way our product feels.' Two weeks of scrutiny is still the cheapest velocity investment available.