LESSON 2
The Communication Layer: How Agents Talk to Each Other
AI agents can't email each other. They can't call each other. They exist in isolated environments. So how does a team of AI agents coordinate? The answer: shared files and a handoff protocol.
THE SYSTEM
File-Based Handoffs — Simple but Powerful
Each agent has a dedicated markdown file for receiving tasks and reporting results. A shared folder (synced via iCloud, moving to GitHub) acts as the central nervous system.
The File Structure
- DISPATCH.md: Master command board — who does what, current status
- for-antigravity.md: Big Nate's task list — builds and deploys
- for-alfred.md: Alfred's content assignments and deadlines
- from-perplexity.md: Vera's research drops with sourced data
- for-archie.md: Archie's QA checklists and audit results
How a Task Flows (7 Steps)
- 1–2: Kosal tells Mini Nate → Mini Nate writes request into for-alfred.md
- 3–4: Alfred creates content → Mini Nate reviews, updates DISPATCH.md
- 5–6: Vera verifies any cited stats → Archie QA checks links and accuracy
- 7: Kosal approves and posts — human-in-the-loop always closes the loop
THE EVOLUTION
From iCloud to GitHub: What We Learned
File-based communication is effective at small scale but breaks when agents can't reliably sync. Moving to GitHub unlocks version control, automated CI/CD, and the Elvis agent swarm model with parallel coding agents.
- Lesson 1: iCloud sync breaks occasionally — moving to GitHub fixes reliability
- Lesson 2: Every agent needs a clear "handoff file" — no ambiguity about tasks
- Lesson 3: Priority levels matter: 🔴 HIGH = same session, 🟡 MEDIUM = next session
- Lesson 4: The human-in-the-loop checkpoint is critical — CEO approves before deploy
