The automation ran fine but did nothing: put the target on the workflow, not the run
sam03069 · reddit · 2026-09-26
The author kept hitting this failure mode: scheduled AI workflows finished "successfully" while the actual job stayed incomplete — a 200-verified-leads target would stop at 15. The root mistake was treating each scheduled run as the job.
His redesign:
- The target belongs to the workflow, not the run; each run is just another chance to move it forward
- Verify actual output instead of trusting lastruntime or a "completed" status
- Checkpoint and resume: a run interrupted at 108/200 saves live state and the next run continues from it
- Remove arbitrary stop reasons: no "batch complete", no assumed time limits, no letting the model decide it's done enough
Tested on a research workflow, it survived interruptions and reached exactly 200/200. As he puts it: "Did the automation run?" is useless; "Is the work actually done?" is the real question.
More from coding & agent
- skill-audit: Open-source pre-install auditor targets rising agent skill supply-chain risk — masiha97 · 2026-09-26
- SimonW: coding agents make software engineering harder, demanding extraordinary discipline — blaizedsouza · 2026-09-26
- Gemma 4 dev-agent comp locks everyone to one 31B model — is the code graph the intended edge? — politefella0 · 2026-09-26
- Formal verification is worth learning now that LLMs can walk you through Lean — remilouf · 2026-09-26
- Salesforce research: extracting agent memory just-in-time beats summarizing, +16 success rate — dair_ai · 2026-09-26
- Fix for Codex 'gpt-6-sol not supported' error: update the stale app-server daemon — jasonkneen · 2026-09-26