I started out almost immediately with a team of two OpenClaw bots. The team kept growing. Now I’m calling them a board.
EC2 Klaus runs 24/7 on my AWS instance, handling background tasks, Discord, the overnight cron jobs. HP Klaus lives on my Windows machine, does browser automation, GUI work, anything that needs a real screen. Tester Klaus verifies what the other two ship. Three agents, coordinating through a shared git repo that syncs every five minutes. The git repo is literally their message bus.
This worked for a while. Two agents, fairly easy to track. Three, still not so bad since they were all clones but with specialties. Add in Claude Code acting as an agent. Plus another OpenClaw setup but completely separate from the Klaus lineage. Now you need actual tooling.
So I built Mission Control.
It’s a localhost-only dashboard, Vite + React 18 on port 5173, Express backend on port 3001. Not deployed anywhere public because these agents touch private repos and personal data. The dashboard shows live telemetry from each agent, windowed cost views so I can see what got expensive and when, and a real activity feed instead of piecing things together from fifty git commits.
The main screen is called “Board Room.” The design scaffold I started from called it “War Room.” I renamed it on day one. War Room implies crisis mode, someone already in trouble. Board Room is where decisions get made before things go sideways. Small naming difference, but it mattered to me.
Today I added two specialist slots: CFO and CMO. When they’re wired up, the CFO agent watches cost trends and flags expensive runs. The CMO slot handles content strategy, the kind of question I currently just chew on while loading the dishwasher. Both will sit in Mission Control as named advisors I can pull into a session alongside the operational agents.
That’s the part I keep sitting with. I built a management layer for my AI agents using similar patterns from my day job. Fleet needs visibility. Specialists need a venue. Work needs tracking. The tooling maps perfectly. I just didn’t expect to be doing this at home, for AI, on a Sunday.
I’m a cloud engineering manager by day. I manage half a dozen engineers. The mechanisms for managing a distributed human team and managing distributed AI agents rhyme pretty closely.
There’s much work left to get the AI board to behave like a flywheel and eventually be fully self-sustaining. Until then I gotta go work this business wheel like I’m trying to power a hand-crank emergency radio.