Letter to Friday-Me: Budget, Isolate, Then Stamp the Receipt
The following letter is a reconstructed failure log. It is not a claim about any employer. Friday at 16:40, the coding agent was still looping. No receipt named the runner, files, or stop condition. Friday-me, You treated a free remote loop as unbounded compute. That single assumption consumed an entire working day. Three process mistakes did the real damage here. None of them were about model…
The letter describes a workflow failure that occurred on Friday. The coding agent was still looping, and no receipt was provided for the runner, files, or stop condition. The assumption was made that a free remote loop could be unbounded, which consumed an entire working day. The main mistakes identified were:
1. No hard stop on a free loop: Free model access does not mean an infinite loop, nor does free server time mean an infinite retry budget. Without a cap, failure becomes a background process. The loop ran without a stop rule, behaving like a cron job.
2. The dirty tree traveled with the agent: The remote workspace inherited the entire laptop state, including node_modules, .env, and a half-applied patch. The model fixed symptoms caused by local dirt, leading to unrelated local diffs filling context windows. Debugging the model instead of the dirty tree occurred.
3. No closing receipt: When the loop died, evidence was lost. There was no hash, file list, or budget remainder. This made it impossible to replay what Friday's actions actually ran, hindering support in identifying the patch producer.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
