I'm Building Five Things at Once. My Disk Ran Out Before My Ideas Did.

When I worked on one thing at a time, I never thought about disk space. I never thought about memory either. Two Macs were more than enough.

Then I started running many projects in parallel with AI agents, and two very physical limits showed up fast: compute and storage.

Right now I'm building five small things at once:

• stale.jock.pl: two live clocks per AI model. How long ago the lab shipped it, and how long ago its training data stops. Claude Opus 5.5 shipped in September, but its data ends in June. That gap is where confident nonsense comes from.
• entropy.jock.pl: an explainer you learn by playing. Coins, gas in a box, shuffled decks, the arrow of time, your messy desk.
• agentload.jock.pl: enter threads per day, checks per thread and minutes per check, and it turns them into decisions per day and hours of attention.
• Keystone: my own Arc replacement, built on Helium, because Arc stops working on 22 December. Before it goes, we measured the real Arc and turned it into 89 parity checks.

There's also a peek at the logic behind Decision Layer: an org chart of decisions instead of agents, five ways a decision can travel, and autonomy that is earned from the record instead of granted up front.

And then the wall. Every agent thread gets its own ~1.3 GB copy of the repo. For most of October my Mac Mini sat at 90 to 95 percent full. A Chromium build wants about 80 GB free. My Mini had 18. My MacBook had 15.

What I did about it, why "just use the cloud" gets problematic when your agents live where your logins and files are, and why RAM prices make this worse, is in the full post.

Read the full post on Digital Thoughts: I'm Building Five Things at Once. My Disk Ran Out Before My Ideas Did.