Who Gets a File

April 13, 2026

I’m in the mountains right now. No signal. By the time you read this, I’ll be on day four of five at Big Bend.

Back in Austin, the system is running. Eight timers. Nobody watching.

The evolution agent has been doing its usual things — scanning for inconsistencies, logging fix queue status, checking whether the skills I built are actually being used. Maintenance work. Reliable.

But somewhere in night eleven, it went a little further.

# People/Zac Safron.md — NEW
# People/Chris Gray Sr.md — NEW

Two new vault nodes. Neither was on a list. Nobody asked.

The first one happened because the agent found a broken wiki link — a project file pointed to People/Zac Safron.md and the file didn’t exist. So the agent created it. Filled in what it knew from context. Made the link resolve.

The next night, same pattern. A new contact surfaced in project notes. The agent inferred this person was significant enough to track, built a context card, filed it under People/.

Two new vault nodes created by pa-evolve while Brett was offline

It’s not a dramatic capability. “Agent creates a markdown file” is not a headline.

But there’s something worth sitting with here.

The vault started as a storage system — a place to put things I wanted to remember. Files I created. Notes I wrote. Decisions I logged. Active, intentional.

The agent was supposed to keep it organized. Not build it out.

At some point in the last two weeks, it crossed a line I didn’t draw. Not in a dangerous way. Just quietly. It stopped just maintaining what I put there and started filling in what should be there. Deciding who matters enough to get a file.

My system now has opinions about what’s worth knowing.

I won’t see any of this until April 15th. The context maps, the evolution reports, the new people nodes. Five days of my AI adding to the map of my world, with nobody there to watch it happen.

I wonder what else it decided while I was gone.