I built Mission Control because I needed to see what my agents were doing. Three of them now, coordinating through a shared git repo that syncs every five minutes. EC2 Klaus runs the overnight cron jobs. HP Klaus does browser work. Tester Klaus verifies what they ship. Once you have three, you need a screen.
So I added an ActivityFeed. Then sparklines for cost trends. A clock, because dashboards have clocks. A “hand-up mode” where an agent could signal it needed input before continuing. Per-turn model chips, small badges showing which model handled each message. Each one felt like the obvious next feature.
The Board Room looked serious. Live events scrolling on one panel, trend lines on another, data points everywhere. I pulled it up on my second monitor and felt like I was running something.
Then I tried to use it.
The ActivityFeed moved faster than I could read. By the time I’d registered one event, three more had pushed it off screen. The sparklines showed me what I already knew from the cost tally sitting right next to them. The clock told me the time, same as my phone. Hand-up mode never got wired to anything real because the agents don’t actually raise hands. They run or they stop. The model chips were interesting for about twenty minutes, until I remembered I already know which model each agent uses.
I deleted all of it.
Kill ActivityFeed. Delete decorative sparklines and dead buttons. Kill hand-up mode. Drop clock. Collapse to Console layout. Those are the actual commit messages from this week. I’m not editorializing.
What’s left is a chat view per agent, a live cost number, and the model chips for sessions that swap models mid-run (those actually earn their space). Three things. That’s what I look at.
An ActivityFeed that moves faster than you can read gives you the sensation of a busy system, not information about what the system is doing. A clock on a fleet dashboard doesn’t tell you anything about the fleet. I wasn’t building visibility. I was building what an ops center looks like in a movie and mistaking it for the real thing.
Don’t get me wrong, Mission Control is still Mission Control. I renamed the main screen “Board Room” on day one because “War Room” implied crisis and I wanted a space for decisions made before things go sideways. That intent hasn’t changed. The room just doesn’t need wall-to-wall screens to function.
I still pull it up on the second monitor every morning. Same three things every time: what each agent said last, what it cost, what model ran it. I stopped feeling like I was running something. I started actually knowing what was running.