Agent task runs 90+ minutes as dev highlights the endless loop of better models enabling harder tasks
BraceSproul · x · 2026-09-12
Developer BraceSproul describes the perpetual loop of agent development: agent tasks take forever → models improve and finish faster → you assign harder tasks → tasks take forever again.
The task he shared has been running for over 90 minutes and still chugging along — illustrating how rising model capability keeps raising the ceiling on task complexity, making long-running agents the norm.
More from coding & agent
- Developer: Codex fast mode with astra shows no usable speed gain — madhavsinghal_ · 2026-09-12
- Claude Code Ships 'claude plugin eval' as Hamel Husain Details Four Key Shortcomings — sh_reya · 2026-09-12
- Founder is skeptical of 'software factories': AI can't yet ship features end-to-end — petergyang · 2026-09-12
- Hermes Desktop Masterclass Part 2: Sessions, Subagents, and Code Review in Action — alexcovo_eth · 2026-09-12
- Full-document context adds no gain over free structural context for code chunk descriptions — donk8r · 2026-09-12
- Sourcegraph CEO Praises Meta's Muse Spark Coding Model and CLI — zeeg · 2026-09-12