Reducing cognitive load

Keeping my brain focused on what matters has always been important to me. Agents can do more of the work, but I still have to keep track of the threads, decisions and unfinished tasks. A pull request here, a message there, something waiting on a reply. It all takes up space.

I used to think “orchestrator threads” were the answer: one place to manage the other agents. I wasn’t far off, but I’ve changed my mind about what that conversation should be. I originally hated the idea of a thread with permanent context. I wanted a clean slate each day. But my work doesn’t reset each morning, and neither do conversations with people.

An ongoing conversation means I shouldn’t have to reconstruct where we left off. What did we decide? What’s still waiting? What needs my attention? If an agent can keep track of that and follow through, I can stop carrying the whole list in my head.

I want my main AI surface to be proactive and flexible enough to fit different ways of working. I want to trust it with responsibility and know it will finish the task. I want it to ask questions when it doesn’t know, including asking other people. I want it to say no and push back. I want to be able to treat it like an employee.

That’s where I think we’re heading. The thing I care about is being able to hand something over and let it leave my head, knowing it will come back when it actually needs me.