Five Agents. Same Question.

The other day I looked at what I had running: five Herdr workspaces, tabs I’d left open for hours, a design tool in another window, and Hermes going in the background. More than I’d ever had going at once. None of it was about the work anymore. It was about proving I could keep up.

The more I lean on these agents, the busier I feel.

At first I tried to hold the whole list in my head. That failed fast, so I moved to to-do apps, then kanban boards, then tracking it inside the LLM tools I was already living in. Structure never came — only a running what’s next that I’d have forgotten by the next session.

None of it stuck. I’d walk away from the machine, and that was the end of it. Next session, I’d start the same triage over again, still not knowing what was worth my time in that moment.

The agent side was its own version of the same problem. I tried tmux to run a few terminals at once. I tried Gas Town and Gas City — tools built for exactly this, several agents running in parallel. I built my own dispatch agent, my version of the Gas Town mayor, routing work between sessions.

None of it made me more productive. It felt like it would — run this task here, start that one over there, oh wait, I forgot about this other one. It didn’t multiply my output. It built a prison of context switching.

So I stopped adding tools to the pile and changed the question. Not how do I work on more at once — how do I know what’s worth working on right now? Then I used Claude to solve the exact problem Claude was creating.

The prompt was something like:

> Help me build a system that answers one question:
what should I be working on right now?

context:
I've tried todo apps, kanban boards, tracking it through chat.
none of it stuck. I want one file, ranked, that tells
me the next right thing to do.

It took a couple of iterations. At one point I reached for the Kanban plugin for Obsidian — the same kind of tool that hadn’t worked for me before. Claude pushed back and suggested a plain ranked list instead.

One file, ranked by priority. One question to answer: what’s the best use of my time right now?

Something like this:

# Command Center

## NOW — this week

- [ ] Message a few people about the automation work
- [ ] Draft the blog post: "The Command Center"

## NEXT — this month

- [ ] Convert a warm yes into a first paid project
- [ ] Ship the next thing on the site

## LATER — someday

- [ ] The idea I keep coming back to but haven't committed to

AI tools didn’t create this problem. They magnified it. Before, I was one person with limited capacity, so I worked one thing at a time — the constraint did the prioritizing for me. Now I can dispatch five agents at once, and it feels like five times the capacity. It isn’t. It’s the same question I always had, only louder and running in five places.

Between family, a day job, and side projects, I don’t have hours to spend guessing what matters. A plain to-do list is useless for that — everything on it looks equally important. Ranking is the whole point. It picks the next thing for me so I’m not re-deciding every time I sit down.

What makes it click is how Claude uses the file. It reads the top item under NOW and points me there. That’s it. The file sticks around between sessions too, so the next morning I’m not piecing it all back together. I open it and keep going.

And it keeps me honest to the parameters I set. Not in a rigid way — the ranking is fluid. When something genuinely more worth my time shows up, it moves to the top, and the rest shifts down. What the file kills is the reflex to chase whatever’s newest because it’s in front of me. A new thing has to earn the top spot.

It never gave me more hours in the day. It made me stop wasting them figuring out where to start.

Yesterday it told me to work on this post. Today it looks like this in practice:

> what should I work on today?

Draft the blog post about the Command Center —
it's at the top of NOW.

I’m still busier than I’ve ever been. Those five workspaces are still open. The difference is I know which one is worth my time.

Knowing what’s next is half of it. Knowing whether it’s worth doing at all is the other half. I built a system for that too — and it already talked me out of an idea I was ready to bet on.