Frame
Define what success changes.
Name the person, decision, artefact, or behaviour that changes when the task succeeds. A broad activity is not yet an outcome.
Control question: What will be different when this is done?The Operator Method
A repeatable path from request to trusted result.
The method makes the hidden decisions visible before AI-generated work becomes expensive, public, or difficult to reverse.
Ground
Supply the source truth.
Bring forward the facts, examples, files, permissions, constraints, and missing inputs that should govern the work.
Control question: Which evidence may the system trust?Generate
Produce a visible first result.
Ask for an outline, plan, sample, calculation, patch, or dry run that is cheaper to correct than a polished but misdirected deliverable.
Control question: What is the smallest useful result to inspect?Inspect
Check the work, not the confidence.
Review the real file, code, calculation, page, source, or customer journey against explicit acceptance tests.
Control question: What independent evidence proves this works?Improve
Correct with specific feedback.
State what failed, where it failed, what must remain unchanged, and how the next result will be evaluated.
Control question: Which correction removes the highest-cost weakness?Preserve
Make approved work reusable.
Save the correction in a template, instruction file, checklist, skill, standard operating procedure, or regression test.
Control question: Where should this learning live for the next run?The BRIEF framework
Prompts work better when the operating controls are visible.
BRIEF turns a vague request into a usable work order without filling the prompt with theatrical role descriptions.
Background
The context that changes the work.
Result
The outcome and audience.
Inputs
The sources, facts, and examples.
Execution
The process, limits, and permissions.
Final checks
The acceptance tests and output format.
Route the work
Choose the instrument before refining the instruction.
ChatGPT
Dialogue, explanation, exploration, brainstorming, rehearsal, and iterative thinking.
ChatGPT Work
Research, analysis, and polished documents, spreadsheets, presentations, and sites.
Codex
Repository inspection, file changes, commands, tests, diffs, and verified implementation.
KAI-style system
A bounded audience, approved knowledge, consistent guidance, guardrails, and human handoff.