Ask HN: How do you pilot a service company full of AI agents?

I've been running my small service company (marketing agency) with a ton of AI agents for a few months now. Got NanoClaw running (OpenClaw-like) with custom automations, tools connected, agents doing some work for me. It's amazing when it works.

But there's a missing piece i can't find, at the cross road between interface and AI backend.

A single Slack/Telegram doesn't work when you're on 10 topics at the same times. Terminal is great but has no memory. Nothing works if you actually want to pilot your company with agents.

In an ideal words i would need to approve tasks the AI is suggesting, run benchmarks on new automations and skills, navigate my knowledge, manage who on my team can trigger what, access what, and have everything running on my claude code subscription (ofc haha)

What I'm missing feels like a desktop app, part cockpit, part Obsidian, part approval queue, part agent manager, part team dashboard, part kanban board

Is this already being built somewhere? Am I the only one feeling this? Should everyone builds his own version of this?