YouTube Script: Module 3 — Boot

Title: "Your AI Agent Is Wasting 15 Minutes Every Session — Here's The Fix" Duration: ~8 minutes Style: Screen recording + talking head intercuts


HOOK (0:00 - 0:45)

[SCREEN: Terminal showing an agent session. Red text: "python3: command not found" → "test directory not found" → "which branch am I on?"]

VILIUS (voiceover): "This was my first month working with AI agents. Every session, the same corrections. Use python3.11, not python3. Tests are in the tests/ directory. Don't commit to main. Fifteen minutes of orientation before a single line of useful work."

[SCREEN: Cut to talking head]

VILIUS (on camera): "I tracked it. Over a month of daily sessions, I spent seven hours just orienting my agent. That's almost a full workday. And the fix? It takes ten minutes. One file. I'll show you."


THE PROBLEM (0:45 - 2:00)

[SCREEN: Split screen — left side shows clean project, right side shows agent making wrong assumptions]

VILIUS (voiceover): "Here's what happens. You open a terminal, you type a prompt, and the agent has zero context. It doesn't know what project it's in. It doesn't know your Python version. It doesn't know your test conventions. It guesses. And guesses wrong."

[SCREEN: List of common agent mistakes scrolling — wrong Python, wrong test runner, wrong git branch, invented directories, deployment scares]

VILIUS (voiceover): "The agent isn't stupid. You just haven't told it anything. And that's what most people get wrong — they treat agents like search engines. Type a query, get an answer, move on. But agents are collaborators. And collaborators need onboarding."


THE SOLUTION: AGENTS.md (2:00 - 4:30)

[SCREEN: VS Code open, creating a new file called AGENTS.md]

VILIUS (on camera): "The fix is a single file called AGENTS.md. It sits in your project root. Most AI coding tools read it automatically when you start a session. It's the source of truth for your project — the agent's onboarding document."

[SCREEN: AGENTS.md being filled in, line by line]

VILIUS (voiceover): "Five things go in this file. One — what the project is. One sentence. Two — the exact stack. Not 'Python' — 'Python 3.11 at /opt/homebrew/bin/python3.11'. Give the agent the exact path. Three — conventions. Exact commands for testing, linting, committing. Four — key files. So the agent knows where things are without asking. Five — autonomy rules. When should the agent ask permission, and when should it just proceed?"

[SCREEN: Highlight each section as it's described]


THE TEMPLATE (4:30 - 5:30)

[SCREEN: Full AGENTS.md template displayed]

VILIUS (voiceover): "Here's the template. Fill in five fields — project name, language path, three conventions, three key files, one autonomy rule. Save it. Ten minutes. That's the entire boot pattern."

[SCREEN: Template collapses to a single card showing "5 fields, 10 minutes"]

VILIUS (on camera): "If you want to automate this — and you should, because environments change — I've included a Python script in the course materials that discovers your environment and generates AGENTS.md automatically. Run it after you update Python, switch Node versions, or restructure your directories."


BEFORE AND AFTER (5:30 - 6:45)

[SCREEN: Split screen — left is "Before Boot" showing agent corrections, right is "After Boot" showing agent getting everything right]

VILIUS (voiceover): "Here's what this looks like in practice. Before boot — every session, I corrected the Python version four times, the test directory three times, the git workflow twice. Fifteen minutes of orientation."

[SCREEN: Timer counting up — 15 minutes wasted per session]

VILIUS (voiceover): "After boot — the agent reads AGENTS.md, uses the right Python, runs the right tests, commits to a feature branch. Zero corrections. Thirty seconds of orientation."

[SCREEN: Timer showing 30 seconds → 14.5 minutes of productive work]

VILIUS (on camera): "Over a month — that's seven hours reclaimed. Seven hours of actual work that would have been spent correcting an agent that didn't have context."


THE SELF-TEST (6:45 - 7:30)

[SCREEN: Fresh terminal session starting]

VILIUS (voiceover): "Here's how you know it worked. Start a fresh session. Ask your agent: what Python version should I use for this project? If it answers from memory — from AGENTS.md — you did it right. If it guesses, your AGENTS.md isn't specific enough, or your tool isn't reading it."

[SCREEN: Agent responds correctly — "Python 3.11 at /opt/homebrew/bin/python3.11"]

VILIUS (on camera): "One more thing. If you work in a regulated industry — finance, healthcare, government — add a compliance boundaries section. 'Never access patient-data directory.' 'Never commit credentials.' State the rules upfront. Your agent won't enforce them by itself — that comes in Pattern 9 — but the boundaries make violations harder and give you a clear reference when reviewing."


OUTRO (7:30 - 8:00)

[SCREEN: Works With Agents logo + "10 Patterns" overview]

VILIUS (voiceover): "AGENTS.md is the foundation. It gives your agent static knowledge — what the project is, how to work in it. In the next module, we add dynamic knowledge — skills. Reusable procedures your agent loads on demand. When it's done a task before, it doesn't figure it out again from scratch. It loads the skill."

[SCREEN: Call to action — "Download the Boot template at workswithagents.com/learn"]

VILIUS (on camera): "Try this today. Ten minutes. AGENTS.md in your project root. Start a fresh session and ask your agent what Python version to use. If you're still correcting it after that — send me a message. But you won't be."


PRODUCTION NOTES

  • B-roll needed: Terminal sessions showing before/after agent behaviour, VS Code editing AGENTS.md
  • Graphics: The "5 fields, 10 minutes" card, timer comparison graphic, compliance boundaries callout
  • Music: Minimal, ambient — don't compete with the explanation
  • Captions: Yes — the terminal demonstrations need them for the commands

← Back to all patterns

Spotted something?

Suggest an improvement, report an error, or just say hi.