The bot can run the desk. I still decide.
I spent a day inside the CEO OS and caught myself taking the top of the queue without asking why it was there. The bot can run the desk. Deciding is still my half of the job.
I posted about the CEO OS yesterday. Show me the next work, hide the rest. That was the whole idea and I still want it.
Then I spent a day inside it, which is a different thing than writing about it.
It worked. I sat down, the next piece was ready, and I picked it up without asking why it was first. Did that a few times before I caught it. Nothing broke. The queue was fine. I just stopped doing my half of the job, because the answer arrived already formed and taking it felt like progress.
Wayne Sutton published something the same day I did, and one line of it stuck. “Bots running operations. Not a bot thinking for you.” His piece is here. I read it on the exact day I’d been letting the queue think for me, so it landed harder than it might have otherwise.
Three lives, one me
I’m still running three things that don’t care about each other. Personal, Switchboard Live, Diger Studios.
Personal is the one I starve first when work gets loud. I know that about myself now, so it gets its own lane instead of competing for scraps. Switchboard Live runs at a pace I don’t set. Diger Studios is where I ship.
The agents do the work. Cursor cloud agents in the repos, an assistant sitting on each of the three lives. That part is handled in a way it wasn’t in March, when everything ran through a Slack hub and every loop started with me typing the first message.
What’s left over is me deciding. Smaller than it sounds. Also heavier than it sounds.
A desk I talk to
Grok Bot on my desktop is a desk I talk to, and I keep coming back to that word because it’s the whole difference. A pile of automation needs feeding. A desk has someone at it.
So I sit down and something is ready. Then I’m supposed to push on it. Why this one first. What got left out. What would I have picked if I’d built the list myself this morning.
Most days pushing on it changes something. Not always the item. Sometimes just the order, sometimes the version of the task I thought I was doing.
And when I don’t push on it, I should be honest about what that is. Rubber-stamping the top of the queue isn’t running an operation. It’s quitting quietly while looking busy. That’s the failure I’m watching for in myself, more than a bad answer from the bot.
The part I can’t hand off
Thinking stays mine. So does working out which problem I’m actually solving, which is usually the harder half and the first thing I drop when the day gets full.
Then orchestration, which I underrated. Three lives and a set of agents that will do whatever I point them at, fast. Pointing badly is now exactly as fast as pointing well, and nothing in the queue protects me from that.
The bot can prepare, sort, draft, chase, and keep the noise off me. It can’t tell me the thing is worth doing.
Until the next post
One rule I’m adding for myself. Ask why it’s first before I pick it up.
Small, but it’s the difference between an OS and an autopilot I didn’t mean to install. I’ll run it that way for a while and pay attention to what I hand back.
Still building in public. Let’s see how this goes.
This is part of a series on how I’m using agents to run the work. Follow me on X.