Build a lean multi-agent back office for one business
A founder split solo-company work across eight Copilot agents that shared memory, checked one another, and handled publishing and admin.
- What they did
- The founder created eight separate agent files in Markdown inside a GitHub Copilot repository. The agents could call MCP tools tied to X, dev.to, Sentry, and scheduling, and they all updated the same JSONL knowledge graph so they could reuse context. A COO agent led a morning check-in and assigned tasks, while a trigger table routed specific items between roles, such as having marketing pass product statements to the lawyer. An Improver agent drafted suggested changes as diffs, and the founder reviewed those edits before they landed.
- How it went
- Two months later, the setup had generated €6.09 in revenue. It also came with about €42 per month in infrastructure costs, 84+ tweets, and 5 articles, while keeping marketing work below an hour a week and tax filings on time. Concurrent writes broke the memory file until the founder switched the server to a forked copy.
- Tips
- A smaller starting point worked better: begin with COO, Marketing, and Accountant before adding the rest. The review step between agents should be part of the design from day one, and the build itself took roughly 40 hours.
Try it yourself with Claude Code
Help me set up a small team of AI agents for [business description]. I want separate role files for jobs like operations, finance, legal, marketing, and engineering, plus one shared JSONL memory file they all can read and update. Show me how each role should hand off work to the others, and include a final human review step before anything is sent, paid, deleted, or submitted. Finish by giving me one simple interaction to test first.
Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.
Get 3 agent uses like this every morning
With the day's top AI news. Free, and one click to unsubscribe.
Related agent uses
Build a hiring rubric from competency levels
A commenter used Copilot Pro to turn company competency definitions into interview questions, scoring, and role fit guidance.
Use Claude Cowork to grade papers and plan lessons
A teacher used Claude Cowork to speed up lesson planning, marking, slide creation, and grade analysis across several classes.
Triage email by voice while you walk
Use voice commands to sort email, draft replies, and capture follow-ups while you are away from your desk.
Run your business for a week with narrow AI agents
A solo owner let three Claude agents handle inbox, content, and support for seven days.
Schedule project file summaries and email them
Set up Codex to read a project folder on a schedule and email a summary through Gmail.
Route each task to the cheapest capable model
A user changed OpenClaw so it could choose lower-cost models for CRM work and email drafting, with manual approval before any outside action.