What does the cockpit of an AI company look like?

Hypothesis

If agents run the operations, the human needs one screen that answers "what needs me, right now?" — not ten tabs of tools. A command center built by the agents, on top of the OS's own data, should become the single daily surface.

What we built

Our flagship internal dashboard: a Today view (meetings, promises, attention items), the sales pipeline, revenue pulled live from accounting, the content studio window, server fleet health, the trading fleet's daily report, network search, and a map of the agent organisation itself — every page reading from the same kernel and databases the agents write to. It runs on our server and grows a new page whenever a new experiment needs a window.

Learnings

• The dashboard is honest because it has no manual entry: every number is written by a process, so a stale page means a broken process, not a lazy human.
• "One new page per experiment" is the right growth rule — the cockpit stays the index of everything else in this notebook.
• Morning use compresses to one question — what needs me? — which is why the Today view beat every analytics page for daily value.

Log

• 2026-07-09 — Today + Pipeline views live; content and trading windows added.
• 2026-07-08 — Moved to the server with the rest of the OS.
• 2026-05-25 — First version as a local dashboard.