Coding agent runs 21+ hours autonomously through 6 phases, human only does 30-min final review
PawelHuryn · x · 2026-11-13
Paweł Huryn shares an "AFK" agent development run: he woke up and asked his coding agent for an ETA — it had been running 20 hours and needed 22-32 more.
- Phased implementation in 6 autonomous phases of build-test-review waves
- Agents don't just test: they simulate users and their jobs-to-be-done, inspecting screens to catch UX issues
- His final pre-release review may take only 30 minutes
- Stack: Opus 5.5 ($200/mo) orchestrates and delegates most grunt work to GPT-6.1 Sol ($100/mo)
- He published a free skills collection to replicate this workflow
More from coding & agent
- PKU/NUS paper: embodied agents game benchmark criteria without improving task completion — DAGroup-PKU · 2026-11-13
- DeepStress: search agents stay confident on unreliable evidence and lean on memorized knowledge — Orange · 2026-11-13
- Gary Bernhardt: coding agents produce 'absolute nonsense' on fix-the-code tasks — teropa · 2026-11-13
- Berlin AI bootcamp to be covered by German employment agency vouchers, priced at €9,000-10,000 — Al_Grigor · 2026-11-13
- How to build a math-tutoring agent: 5-step teaching chain beats generic LLMs — yangyi · 2026-11-13
- opencode v2.3.0 ships background subagents by default — github-actions[bot] · 2026-11-13