My AI builder had 12 jobs last night. The 4 it didn't do are the part I'd pay for

Description and use

The AI builder completed 6 of 12 jobs, but 4 were stopped due to budget limits, bot‑blocking, quality rules, and partial output, highlighting the need for logged stops for human review.

Problem

Automated AI agents sometimes refuse to complete jobs that involve cost, external system constraints, quality rules, or partial outputs, requiring human oversight.

Outcome

The system successfully completed 6 jobs, identified 4 critical stop reasons (budget, bot block, quality rule, partial output), and highlighted the importance of logging stops for human review.

Workflow

• 1. AI builder wakes at 02:40 Helsinki time
• 2. Reads ranked list of jobs
• 3. Attempts to build each job
• 4. If a job triggers a stop condition, logs reason and stops
• 5. Writes one line per job into a ledger
• 6. Owner reviews ledger the next morning

Evidence

The ledger says: 6 built or done, 2 not reached for lack of time, and 4 where it stopped and wrote down why

Source

It wouldn't spend $4 without asking

Source

It refused: that is getting past a site's bot block by pretending to be someone else

Source

It killed 39 pages its own rule said not to publish

Source

It marked the job "medium", said so, and left it for a dedicated slot instead of shipping half an issue to subscribers

Source