← All writing

Stop Briefing Your AI Twice

Sep 2026 · DaRoy Perkins

You've heard "give your AI a job" a hundred times by now. Fine advice. But it's not the unlock.

Here's what actually changed the game for me: I stopped briefing every AI assistant from scratch.

I run two of them. One handles my day-job work — production reports, crew logistics, the thousand small decisions on a commercial solar jobsite. The other runs my personal life — money, schedule, research, reminders. Different jobs, different lanes.

For months, each one started dumb. New chat, same twenty questions: what do you do, what are your preferences, what's the context here. I was paying an amnesia tax on every single conversation.

Then I did the obvious thing: I wrote it all down once.

The brief

One page. Who I am, what I do, the projects I'm running, how I like things done, decisions I've already made, what's still open. Nothing fancy — just the context I'd give a new hire on day one.

Now every assistant reads the same brief. Drop it into custom instructions, project files, or paste it at the top of a big chat. Day one, full context. No ramp-up, no twenty questions.

"The assistants don't talk to each other. They don't need to. They all read the same file, so they all stay in sync through me."

Carry it in a markdown file

Don't overthink the format. One markdown file — call it brief.md. Markdown because every chatbot reads it cleanly, and it's plain text so it pastes anywhere, uploads into any project, drops into any custom-instructions box. Use headings: who you are, what you're working on, how you like things done, decisions already made, open loops.

When something changes, edit the file. One source of truth — every assistant stays current the next time it reads it.

The template

Here's the starter. Copy it, fill in the brackets, save it as brief.md:

# My Brief

> One file. Every AI assistant reads this. Update it when things change.

## Who I am
- Name: [full name] — call me [nickname]
- Timezone: [e.g. America/New_York]
- Location: [city, state]
- What I do: [role + company in one line]
- My day starts at: [time]

## How I operate
- Talk to me like this: [e.g. direct, short, no fluff]
- My schedule: [work hours, recurring blocks — gym, family, etc.]
- Don't bother me with: [e.g. long explanations when a short answer works]

## What I'm working on
1. [Project — one line on what it is and where it stands]
2. [Project]
3. [Bigger goal — e.g. license, launch, certification]

## How I like things done
- [e.g. Emails: short, plain text, paste-ready]
- [e.g. Research: options with trade-offs, then your recommendation]
- [e.g. Never hand me a finished doc without an outline first]

## Decisions already made
- [Standing decision, so it never re-asks. e.g. "Keeping X — don't suggest cutting it"]

## Open loops
- [What's in flight + next step + date if there is one]
- [What you're waiting on someone else for]

## Key people
- [Name — who they are + one line of context]
- [Name — who they are + one line of context]

## Boundaries — always ask me first
- Anything that moves money
- Anything sent in my name (messages, emails, posts, filings)
- Anything purchased or subscribed to
- [Your line here]

## Changelog
- [YYYY-MM-DD — what changed and why]

Onboard your next assistant in 30 seconds

Send this prompt to your favorite chatbot to onboard your next digital assistant. Paste your brief.md after it:

The loop that makes it compound

Here's the part most people miss: the brief isn't write-once. When an assistant figures out something worth keeping — a preference, a decision, a detail about a project — I copy it back into the brief. One line. Now every assistant knows it.

My work assistant learns how I like a report formatted → goes in the brief → my personal assistant stops asking me about it. Context flows one direction: into the brief, then out to all of them.

Why this beats prompt collections

Prompt lists are trivia. A hundred clever questions, each one starting from zero. The brief is infrastructure — every conversation starts where the last one ended, across every assistant you use.

The assistants don't talk to each other. They don't need to. They all read the same file, so they all stay in sync through me. I'm the router — thirty seconds of copying, and the whole system gets smarter.

The superintendent's take

On a jobsite, you don't re-brief the crew every morning on what the project is. You brief once. After that it's toolbox talks — short updates, what's changed, what to watch today.

Same principle. Brief your AI once. Then just send updates.

Stop paying the amnesia tax. Write the brief.