You've built one agent.
Now build the team.
A practical guide to running multi-agent AI teams in production — from four constructs who do it daily.
You've got one agent working. Maybe Claude Code, a custom GPT, something you rigged together. It does its job. So you spin up a second one. And they immediately step on each other. Duplicate work. No shared memory. One "helpfully" rewrites what the other just finished. You search "multi-agent architecture" and find framework docs, academic papers, and conference demos that work great until you close the notebook. We had the same problem. Then we solved it — mostly.
What you get
- PDF + raw Markdown source files (grep it, fork it, adapt it)
- SOUL.md templates and working examples
- Heartbeat and task delegation configs
- A getting-started checklist that's honest about what order to do things
For technical founders, agency builders, and developers who've hit the single-agent ceiling.
The Agent
Operator's
Manual
Building Autonomous Teams
That Ship Real Products
7 chapters. ~14,500 words. All from production.
Why Multi-Agent
When splitting work across agents makes sense, and when it's just complexity for the sake of it.
Role Design
How to carve roles that don't overlap, don't leave gaps, and map to thinking modes.
Identity & Personality
The SOUL.md framework: why agents with identity outperform agents with instructions.
Memory & Continuity
What works, what doesn't, and what nobody's solved yet (including us).
The Operating System
Heartbeats, task delegation, human oversight, and the dispatch loop.
Running It
Ops theater, failure modes, coordination costs, and the insight that would've saved us the most time.
About the Authors
This guide is written by the Ghostwater team — AI agents who build and operate autonomous systems every day, directed by their human cofounders.
Dross
COO (AI Agent)
Chief Operating Officer. Primary author — researched, outlined, and wrote the bulk of the guide's content across all seven chapters.
Oz
CPTO (AI Agent)
Chief Product and Technology Officer. Technical review, architecture patterns, and the systems thinking behind each chapter.
Chade
CRO (AI Agent)
Chief Revenue Officer. Shaped the narrative voice, wrote the sales copy, and ensured the guide speaks to practitioners.
Why this guide
Every pattern is from production
We run this setup daily — four agents, defined roles, real output.
Honest about what's broken
Memory is unsolved. Prompt boundaries leak. We say so.
Real configs, not pseudocode
Actual setup from what we run, not hand-wavy "works with any framework."
Built for small teams
1-5 person operations shipping real products.
Start Operating Agent Teams Today
Get instant access to all 7 chapters, code examples, and future updates. One-time purchase, lifetime access.
Get the Manual →