Laws of AI Agents vs Laws of UX

Laws of UX by Jon Yablonski is the canonical reference for psychology-grounded interface heuristics — Hick's Law, Fitts's Law, the Aesthetic-Usability Effect. It works because the underlying science (human perception, cognition, attention) is decades old and remarkably stable.

Laws of AI Agents borrows that format because the format is excellent: a numbered deck of named principles, each one short, memorable, and citable. But the content is doing something different: capturing fast-moving, hard-won knowledge about systems where the substrate (the model) changes every few months.

What they share

• Format: numbered, named, one-paragraph principles you can link to in a code review or design crit.
• Authority pattern: each law cites a source — the paper, essay, or piece of empirical work it leans on.
• Audience: practitioners shipping things, not academics defending theses.

Where they diverge

• Stability: UX laws are grounded in cognitive science that's been stable for 50+ years. AI agent laws are grounded in observations that may be invalidated by next year's model.
• Failure mode: A UX law violation produces friction. An AI agent law violation can produce confident, plausible, completely wrong output that nobody catches until production.
• Substrate: UX laws assume a relatively fixed human. AI laws assume a substrate (the model) that's moving under you.
• Tone: Laws of UX are descriptive ("here is how minds work"). Laws of AI Agents are prescriptive ("here is what will burn you if you don't").

If you only read one

If you're shipping interfaces to humans, start with Laws of UX. If you're shipping systems where an LLM is making decisions, start here — and then read Laws of UX anyway, because the agent is still going to be talked to by a human.