← Principles
validated

“Show intent before action.”

Humans review the shape of work far better than they review individual operations. An agent that externalizes what it is about to do — a plan, a preview, a narrated next step — converts reflexive per-action consent into one meaningful decision, and turns surprises into diffs against something the human saw coming.

The Context Matrix

Where this principle has been tested. Untested cells are honest gaps, not passing grades — seen evidence for one? Tell us.

ChatIDECLICanvas / DocAmbient / Background
Coding ● ● ◐
Knowledge Work
Consumer ●
Enterprise Ops
Creative

● supports · ◐ boundary · ✕ violates · blank untested

Linked Patterns

PlanningApprovalProgress

Boundary Conditions

  • coding × ambient-background × background

    Codex cloud runs: the plan is a trace, not a gate. For background delegation, showing intent to nobody blocks nothing — the principle holds only where a human is present to read the intent. Interruption cost exceeds error cost.

Evidence

supports

Claude Code Plan Mode: nothing executes until the human approves the externalized plan.

coding × CLI × supervised · Claude Code: Plan Mode

supports

Codex shows the command before execution even in auto mode — intent precedes action at the gate.

coding × IDE × auto-with-gates · Codex: approval modes for command execution

boundary

Codex cloud runs: the plan is a trace, not a gate. For background delegation, showing intent to nobody blocks nothing — the principle holds only where a human is present to read the intent. Interruption cost exceeds error cost.

coding × ambient-background × background · Codex: task plans in cloud runs

supports

ChatGPT agent mode narrates the next step before taking it in the activity log.

consumer × chat × auto-with-gates · ChatGPT agent mode: activity log

Tagged Stories