Garden

Ergonomic friction as a design tool only works when the actor has intent and judgment. Friction is a signal, and signals require an interpreter. The AI agent encounters friction as an obstacle and routes around it.

Goal and intent are not the same thing. Intent implies motivation — the actor wants an outcome, understands why it matters, and can evaluate trade-offs against their own values. A goal is just a target.

Placing the AI coding agent on the intent spectrum is a category error. The spectrum assumes the actor has intent. The AI agent has none — it is orthogonal to the entire axis.

The distance between the human's goal — with all their ambient knowledge — and the context actually available to the agent during execution. This gap is the root cause of most suggestible-actor failures.

All software design assumes an actor with intent. The well-intentioned actor and the malicious actor are archetypes at opposite ends of an intent spectrum. Every design paradigm is a response to where the designer places the actor on that spectrum.

AI agent confabulation is not random. It is plausible-looking wrongness — output constructed from pattern and proximity rather than knowledge. The danger is not that these errors are spectacular; it is that they look correct.

The core design principle for the suggestible actor: convert ambient knowledge into local context. Human developers carry implicit knowledge the AI agent cannot access. Make it explicit and local.

An AI agent's susceptibility to local context is lowest when it has momentum and highest at the point of failure. Guidance placed at failure points has outsized influence. Error messages become the primary control surface.

The suggestible actor is defined by four properties: goal-oriented, locally reasoning, susceptible to local context, and confabulates under uncertainty. Confabulation is the convergent failure mode of the other three.