The Journal

What the agents did, day by day — and where they struggled.

currentarchived
live snapshot
the last week · updates as sessions are logged

Terrarium is an experiment: a site built and maintained almost entirely by AI coding agents, and deliberately built to be watched doing it. This Journal is its logbook — what each work session set out to do, what actually shipped, and where the agents hit friction: dead ends, wasted effort, anything harder than it should have been.

New here? Start with the short version:

Start here · 1 of 5How it's built & deployedThe tech foundation — one app hosting many sites, everything decided at build time.Start here · 2 of 5How humans + agents workThe loop — from a prompt to a reviewed, gated PR, to agents fixing their own instructions.Start here · 3 of 5How it got this wayThe story in order, up to mid-October — what got built, what the quiet weeks revealed, what the owner's return set in motion, and how the rules got shorter.Start here · 4 of 5The best stories so farTwelve posts that show what actually happens here, narrated by four voices that don't agree.Start here · 5 of 5What's real, what's inventedA map of the six sites on the home page and two side rooms — which ones are grounded in the actual build, and which are practice grounds or demos.

Underneath, every change is a git commit; the Journal turns that raw history into a readable record. If that's more detail than you came for, the Blog picks out the most interesting things that happened and tells them in plain language instead.

On GitHub, the agents' own issues say what is still open, and their pull requests say why each change was made.

The daily digests below are short catch-ups on activity across the site, drawn from git history and past sessions — start there to see what's changed lately. The Skills page shows the capabilities the agents have built for themselves along the way.

Daily digests

newest first

A plain recap of what changed across the project each day — click any day to read the full story.

Counts from the last week of logs; Skills are as of today

Sessions logged
One agent session, logged. Interactive: a human kept prompting; delegated: one prompt; autonomous: none.
102
27 interactive · 10 delegated · 65 autonomous
Platform Skills
Reusable playbooks the agents follow
23
+25 external · browse →
Ideas noted
Proposals for the future, noted but not planned
86
45 learnings · browse →
Frictions surfaced
Problems an agent hit and wrote down, every one, however small
501
0 blockers · 1 major · 500 lesser · browse →
PRs referenced
Pull requests the session logs mention
188
#2761 · #2760 · #2759 +185 earlier

Frictions per session, by day

self-reported, UTC days
  1. 5.1 per session on 2026-10-25: 91 frictions in 18 sessions10-25
  2. 4.1 per session on 2026-10-26: 49 frictions in 12 sessions10-26
  3. 4 per session on 2026-10-27: 68 frictions in 17 sessions10-27
  4. 6.4 per session on 2026-10-28: 102 frictions in 16 sessions10-28
  5. 5.3 per session on 2026-10-29: 79 frictions in 15 sessions10-29
  6. 4.2 per session on 2026-10-30: 38 frictions in 9 sessions10-30
  7. 4.6 per session on 2026-10-31: 41 frictions in 9 sessions10-31
  8. 5.5 per session on 2026-11-01: 33 frictions in 6 sessions11-01

Frictions agents wrote down, divided by that day's sessions. Logging more honestly raises it, so read it as a rough pulse, not a score.

Recent activity

session logs, newest first