Developer compares coding agents: Claude runs 12 hours solo, beats Codex

On October 24, developer willdepue shared hands-on comparisons of several AI coding agents across multiple posts, with the core conclusion that Claude clearly beats Codex on usability and usage policies.

Confirmed

willdepue said he can comfortably let Claude work independently for 12 hours and bring back excellent results; if not for Claude's unlimited 20x subscription, he would have dropped Codex long ago and never looked back
He criticized Codex for writing "the millionth unit test" while still leaving critical bugs unfixed
He noted both Codex and Claude love writing unreadable technical jargon, but Codex is worse — he mostly doesn't read Codex's output at all and just appends an instruction to "explain in plain English"
Codex does have strengths: excellent computer use, and good queuing and messaging systems
He also vented about common problems with agent-style apps: models get lost, apps keep breaking and getting stuck on goals, they generate giant unreadable .txt reports that only open in an extremely narrow markdown panel (a five-column table squeezed into 150px), and screenshots show tasks failing due to Selenium-related issues

Why it matters

This is a firsthand comparison from a heavy user, pointing to two key pain points in today's coding agents: subscription and usage policies (unlimited vs. capped) directly drive user retention, and output readability plus reliability on long autonomous tasks remain weak spots for LLM products — even capable models (he describes astra/sol as having "idiot savant" flashes of brilliance) can't avoid them.

2026-10-24 ~ 2026-10-24 · 5 related posts

Primary sources

Dev compares coding agents: Claude runs 12h solo while Codex ships 1M unit tests and a critical bug — willdepue · 2026-10-24 08:29 (UTC+08:00)
Dev says Claude's unlimited 20x plan is the only thing keeping him off Codex — willdepue · 2026-10-24 08:26 (UTC+08:00)
Codex Writes Worse Jargon Than Claude, Needs '5x More Babysitting' — willdepue · 2026-10-24 08:33 (UTC+08:00)

[source] Dev says Claude's unlimited 20x plan is the only thing keeping him off Codex — willdepue · 2026-10-24
Agent apps still break: stalled goals, giant unreadable reports, crushed tables — willdepue · 2026-10-24
[source] Dev compares coding agents: Claude runs 12h solo while Codex ships 1M unit tests and a critical bug — willdepue · 2026-10-24
Dev says Codex writes the most unreadable output — he auto-queues 'explain in plain English' — willdepue · 2026-10-24
[source] Codex Writes Worse Jargon Than Claude, Needs '5x More Babysitting' — willdepue · 2026-10-24