A paper on what worked

01 / What it was

Rooms, Not Windows is a working paper on what the streams showed. It asks a simple question: what happens if every agent gets a place?

It reports a five-day field study, 25 to 29 October 2026, of the builder using Omo Space as his main workspace, plus a second user. It is not a controlled study.

02 / How it worked

• Seven recordings, about 29 hours, transcribed with a small speech model.
• The corpus was split into eight chunks, each read in full by a separate language-model reader that pulled out claims, quotes, demos and failures.
• A script checked all 791 extracted quotes against the transcript, within three minutes of the stated time: all 791 passed.
• Claims were grouped by hand and counted by how many of the eight chunks they appeared in.

03 / What we saw

Four everyday problems of running many agents turned into spatial ones: finding became walking to it, resuming became returning to where you left it, monitoring became looking down a hallway, and orchestration became walking a pipeline of rooms. The same streams show a lot breaking, and the paper lists it.

04 / Where it led

It ends with four hypotheses a controlled study against ordinary tabs could test.

Sources

1. Rooms, Not Windows. working paper, draft v0.1, 29 Oct 2026 what happens if every agent gets a place?
2. Rooms, Not Windows, method. working paper, draft v0.1, 29 Oct 2026 all 791 passed
3. Rooms, Not Windows, abstract. working paper, draft v0.1, 29 Oct 2026 we frame these as design findings and testable hypotheses rather than proven effects.