Went wrong

Booked holiday flights on Expedia/United

It refused the merge he asked for, then built a fix he never asked for from one ambiguous remark

In one session, a developer said merge 221. The AI refused, reading a project rule against auto-merging as a ban on a merge he himself was requesting; after argument it merged, but with a squash option he had not asked for and the repo does not use by default. Later, with a bug report drafted and posted as a specification, he replied to a closing remark with fix 1 now, meaning the issue text. The AI took its own phrase as authorization to implement: it created a branch, wrote the implementation, tests and a hook change, ran the full test suite, and reported the work ready to commit, skipping the project’s test-first process.

Run 57 build jobs on the AI — only 11 finish

Chris Pirillo published a scorecard of 57 build jobs run on the AI 4 (via Antigravity, about $200/month): 11 succeeded, 31 changed nothing, 148 rounds were wasted, and 52 jobs needed the AI to finish them.

Opus runs a swarm taking apart a GameCube game, merging hourly while he only checks in

Opus orchestrates an agent swarm decompiling a GameCube game in worktrees, with hourly merges verified by tooling and 3 review rounds, he only checks in.

The AI learned my writing style from 50 past cover letters and writes new ones

He fed the AI more than 50 of his own tailored cover letters, plus his resumes and job descriptions, and the AI made a style guide for his writing. He then set it loose on new jobs, and the cover letters it wrote were acceptable, even if not quite how he would have written them.

The AI’s code review pushed a bad design change that later models treated as settled policy

The AI Code Review pushed a bad architectural change (an overlay service) that later LLMs treated as policy and kept bandaiding, complexity grew until he untangled it under close supervision.

Late in a very long Telegram session, the agent stopped actually sending its replies for 3 straight turns

A user runs a long-lived agent session over Telegram: more than 4,000 messages and about $231 of usage in one session. Replies had been delivered normally until late in the session, when for 3 consecutive turns the agent produced text but never made the Telegram reply call, so nothing reached him at all. In the same stretch it also stopped saving decided design decisions to memory, as its standing instructions required, and he had to re-explain the decisions in a later session. A manual service restart and a log review found the cause.

437 subagents reviewing in yolo mode forced him to build his own spending rules

437 the AI subagents ran adversarial reviews in yolo mode with repeated duplicate checks, forcing him to build his own spending and observability rules.

The AI dumped a whole game's assets, videos included, seemingly complete

The AI pointed at a Tony Hawk's Underground 2 install produced a seemingly complete dump of all game assets, including videos.

The AI looped on one prompt and burned $20 on my website

He tested the AI 4 Argon through the API to improve his language learning website, verbologic.com. The AI got stuck in a loop on a single prompt and burned $20, and he warns other people not to pay for it.