Skip to content
All frameworks

Diagnostic · Chapter 6

Five Laws Diagnostic

Five laws for AI-era usability. Five diagnostic questions. Run them against any product you are building or evaluating.

The laws are not replacements for Nielsen's heuristics. They sit on top of them. A product that satisfies all five new laws but violates error prevention is still a broken product. The foundation holds. These extend it into territory the foundation was not built to cover.

01

Predictable Unpredictability

Design for variation, but make the variation legible. Unexplained unpredictability is not acceptable.

Diagnostic question

Can a user who has been using this product for a month explain why outputs vary? If not, the system’s variability is not legible.

02

The Prompt Is the Interface

Prompt design is UX. The interface has a job before the user types anything: show them what a well-formed request looks like.

Diagnostic question

Does the interface help users write better instructions, or does it accept whatever they type and return whatever it returns? If the latter, prompt design has been treated as the user’s problem rather than the product’s.

03

Trust Is the Usability Metric

Trust is not a quality consideration adjacent to usability. It is the metric that determines whether all the other metrics are reachable.

Diagnostic question

At what point does a user stop verifying outputs and start trusting them? If that point never comes, the trust layer has failed.

04

Muscle Memory Is Temporary

Adaptive interfaces can be better than fixed ones. But adaptation has a cost. The designer has to account for it.

Diagnostic question

Does a user who takes a week off from the product need to relearn anything? If yes, ask whether the adaptation that caused the change was worth the cost.

05

Context Replaces Consistency

Context is the new consistency. Context that is wrong, hidden, or locked in is worse than no context at all.

Diagnostic question

What context does the system have about this user? Is it using that context? Does the user know it is being used? Can they correct it if it is wrong?

Answer all five to see your result.

The five laws

  1. 01

    Set Clear Boundaries

    AI outputs naturally vary. Usability breaks when users cannot predict the range of possible answers. Design so users understand what the system can and cannot do.

  2. 02

    Guide User Inputs

    A blank prompt box creates user paralysis. Provide clear suggestions, shortcuts, and context so users do not have to guess what to type.

  3. 03

    Earn Trust First

    If users do not trust your AI, they will double-check every single output. Building trust is essential to making your AI features usable.

  4. 04

    Keep Core Controls Fixed

    While generative UI can adapt to tasks, keep high-frequency controls in fixed positions so users do not lose muscle memory.

  5. 05

    Use Context Responsibly

    AI adapts using user data. Ensure that context usage is accurate, transparent, and easy for the user to view or modify at any time.

From the book

Chapter 6 covers how the old laws bend and what replaces them.