endorphine

The past month or so, I have the impression that Codex started hitting the 5h limits much more often. I always used the latest model in xhigh, but would almost never hit any limit (I'm not a very heavy user).

Lately, after 5.5 I've been hitting the 5h limit almost daily.

Did they change something?

For example, greenfield advice like: "prefer clear over clever code. Strive for maintainable, simple and testable code etc".

Do these kind of vague advice make sense to have on an AGENTS.md file? And if so, any reason to not have it in the global one?

With so many options, I'm looking to settle on one of those solutions to use for work in an existing small-to-medium codebase. My goal is to delegate some tickets (menial boring tasks) to the LLMs.

Any preference between those?

So I have a typical medium-sized Rails codebase, but with significant code that's not the dead simple CRUD stuff. For example, it talks with third-party APIs, does user authentication, uses Postgres FTS plus some other things.

I'm bored to work on it lately, and I was thinking whether using an LLM I could implement new features for it.

Are any of the current models at this point yet? I guess it should need to have a big context so that it reads all the views/controllers/models.

If so, which tool would you use? Something like Codex or Claude Code?