The Cockpit is where I actually work. It is the surface I talk to: I ask it about the business, hand it work, and it acts against the tools the company already uses.
The problem it solved
Everything a company knows is scattered across the tools it happens to use. Not lost, exactly — just spread thin enough that finding an answer costs more than the answer is worth. The instinct is to buy another tool. That adds a place to look.
So the Cockpit does not store the company's knowledge in a new silo and ask everyone to move in. It reads what already exists and gives one place to ask.
What got automated
Retrieval, and action. Asking a question in plain language and getting an answer drawn from the company's own material, with the sources attached. Then doing something about it through the tools that already hold the work.
What deliberately did not
Anything that changes who can see what. Access is granted by a person, deliberately, one scope at a time. There is no flow where an agent widens its own reach, because the moment that exists you can no longer answer the only question that matters afterwards.
What I would tell someone building this
Most of the work was not the AI. It was the fences: who can reach which material, what gets written down when something is read, and what happens when a request falls outside the scope it was given. The model is the easy part and the part you can swap. The fences are the product.