Rebuttal: Software Is Complex and Agents Can't Build Agents Yet
theo · x · 2026-09-12
A response to the claim that agent task times have doubled: the author attributes it mainly to software being extremely complex and to agents not yet being able to build agents (which is most of his current work), plus his reliability requirements. He attaches Codex records but notes they tell little of the story — he uses many devices, different accounts (corp vs personal), and many harnesses, using Claude Code primarily early in the year and Codex exclusively late in the year plus an internal harness. He suspects the measurement is wrong, noting he is at 0% quota remaining yet has had dozens of multi-hour hands-free tasks in the last two weeks, one lasting over a day. He argues the data must be dissected by task type, project, model, and reasoning level, and speculates he simply writes more or more complex software with different requirements.
More from coding & agent
- Anthropic's Jacalazu shares her agent-minutes formula for maximizing AI agent output — jacalulu · 2026-09-12
- Dev splits workflow: local scheduled tasks on Codex, cloud tasks ported to Grok Bot — petergyang · 2026-09-12
- OpenAI ships Docs skill for Codex plus a Developers plugin for building and managing agents — gabrielchua · 2026-09-12
- OpenAI Agents API ships with sandboxes, compaction, and multi-agent support — gabrielchua · 2026-09-12
- OpenAI hackathon tips: match Astra/Sol/Terra/Luna to tasks, lean on Agents API — gabrielchua · 2026-09-12
- Team runs OpenClaw in dev mode: multiplayer agent coding via Slack with mixed models — heyneighbor · 2026-09-12