← Steve Bassoli

Dev log

Blending

Eleven days, one designer and one AI coding assistant, a tile game built in small steps and logged as it went. This page is made from the git history and the AI research note.

931commits
131deploys
11days
16test files

Day by day

1 OctA paper game becomes a tile builder

17 commits

2 to 5 OctMultiplayer, bots and triangles

47 commits

6 OctThe designer deletes the game

18 commits

7 OctWater, thick tiles, a performance hunt

105 commits

8 OctThe builder becomes a one-player game

187 commits

9 OctOne rules module, replayed by the server

204 commits

10 OctThe map, the double game, music

216 commits

11 OctGlass

137 commits so far

What the AI research found

  1. A command-line player running the same rules as the page and the server let the AI explore the game from the inside, and made server-side checking of every score possible.
  2. Photographs of the paper game were a fast way to prototype.
  3. A heavy specification framework cost many tokens and made a changing prototype too rigid. It was removed.
  4. Measure before believing: a benchmark said 60 frames a second, the real window said otherwise.
  5. The designer decided every balance question and what to cut. The AI typed and checked.
  6. Every claim was labelled checked or unchecked. The most useful sentence in the AI's replies was where its checking stopped.
  7. A human gate on deploys, plus a rollback log, made fast shipping safe.
  8. Small, atomic requests, checked on the page while the designer watched, beat any plan.

A field report written mostly by the AI, with the designer's words verbatim. A case study, not a controlled experiment.

How it works