The argument
Principles
Ten commitments, grouped by when they bind: before the interaction, during it, when the AI is wrong, and over time. Every pattern in this guide exists to serve at least one of them.
Before the interaction
During the interaction
- PR—03Show the workExpose the process, not just the answer — legibility is where trust actually comes from.
- PR—04Keep the human seniorThe AI drafts, proposes, and executes — the human directs, approves, and can always grab the wheel.
- PR—05Match effort to stakesSpend interface, latency, and ceremony in proportion to what the task is worth.
When the AI is wrong
Over time
- PR—08Make memory legibleIf the system remembers, people must be able to see, edit, and delete what it holds.
- PR—09Feedback must matterNever ask for feedback you won’t visibly act on — close the loop or drop the button.
- PR—10Never surprise with changeThe system may improve, but it may not shape-shift — announce change, preserve continuity, degrade predictably.