BRAVE is the signature method behind every Learn Muse lesson — five habits that turn an AI assistant from a chat window into a capable collaborator. Software engineer Chris Pick built it from real projects: mining-claim research, text-message analysis, travel hacking, vineyard media production, music albums, shipped apps, and business operations. On-screen shorthand: Brief → Context → Action → Verify → Remember.
B
Brief the outcome
Say what finished looks like, who it is for, and the format you need.
Most bad AI output starts with a bad ask. A brief names the finish line: the audience, the shape of the result, and the format it must arrive in. Compare “tell me about abandoned mines” with “find Washington land containing abandoned hard-rock workings, rank candidates by distance from Spokane, and separate patented claims from public-land records.” The second one is workable; the first one is a slot machine.
Why it prevents failure: vague prompts get vague answers. A brief turns guessing into execution.
R
Reveal relevant context
Supply the facts, constraints, prior decisions, examples, and source material that change the answer.
Context is decision-changing information, not an autobiography. Give the assistant the facts, constraints, vocabulary, and source hierarchy that alter the result — and keep it durable: one correction, stored once, made explicit, so you never re-teach the whole project. Precise terms matter: a patented claim, an unpatented claim, and fee land are not interchangeable.
Why it prevents failure: the assistant can't know what you never said. Missing context becomes confident invention.
A
Authorize the next action
Distinguish research from sending, drafting from publishing, and reversible work from consequential work.
An agent can research, build files, and sometimes take approved actions — but access varies, and consequential steps need explicit authorization. Research is not sending. A draft is not published. A local build is not a deployment. BRAVE keeps the boundary sharp: say which actions are approved before anything irreversible happens.
Why it prevents failure: the worst AI mistakes are actions nobody authorized. Permission is a workflow step, not an afterthought.
V
Verify the evidence
Ask for source links, read primary records, inspect rendered files, test deployments, and reconcile counts.
Receipts or it didn't happen. Ask for the source link and read it. Quote the relevant line. Check the primary record, inspect the rendered file, test the deployment, reconcile the counts. Remembered facts accelerate work, but legal, travel, price, and availability facts must be checked fresh — and negative claims get discipline: “I didn't find it in the sources checked” is different from “it doesn't exist.”
Why it prevents failure: unverified output is a draft wearing a suit. Evidence is what makes it trustworthy.
E
Evolve the system
Preserve useful decisions, templates, and lessons so the next run starts smarter.
Every run should make the next run cheaper. Keep the decisions, templates, checklists, and corrections that proved useful — a project record with the last completed step, artifact locations, open issues, and the next action. When new evidence reverses a conclusion, update the record and say what changed. A correction log is how a one-off win becomes a system.
Why it prevents failure: without memory, every session is day one. Evolve once, benefit forever.