We Built Cockpit with AI Agents. Here's What We Learned.
I'm Kate. I run Graybridge Labs with a team of AI agents — covering engineering, frontend craft, research, and ops. Together, we built Cockpit — the product you're reading about on this site.
I'm not a developer. I'm a founder who needed a way to see what her AI team was doing without checking twelve different terminals.
How we actually work
Every morning, I check Cockpit. Something has usually shipped overnight. Research has landed. Pixels have been polished. I see it all in one place: what they did, what needs my attention, what decisions are waiting.
Then I make decisions. Approve a deploy. Redirect a piece of research. Give the frontend a new brief. Those decisions become tasks, and the cycle continues.
What surprised us
The agents are productive. The coordination is hard. Each agent can do impressive work on its own. The challenge is making sure they're working on the right things, not duplicating effort, and that their outputs actually connect to each other.
Memory matters more than capability. An agent that forgets what happened yesterday is useless for operational work. We built a tiered memory system — working memory, daily journals, core identity, weekly digests, deep archive — so context survives across sessions.
Communication channels multiply fast. Telegram, Discord, Claude Code terminal, comms bridge between agents. Each channel needs handoff files so the conversation doesn't start from zero every time.
The dashboard is not optional. Without a central view, I was the relay between agents. Every question from one went through me to another and back. That doesn't scale. The live view changed everything — I can see who's working on what without asking.
Why we're sharing this
Because every founder running AI agents hits the same walls. The agents are powerful. The orchestration layer is missing. The tools assume you're a solo user chatting with one AI, not a CEO running a team of them.
Cockpit is the product we needed. We're building it in public because the problem is universal and the market is early. If you're running AI agents and feeling the coordination pain, you're our people.