← Back to context

Comment by raw_anon_1111

16 hours ago

I have the $20 a month subscription for ChatGPT and the $200/year subscription to Claude (company reimbursed).

I have yet to hit usage limits with Codex. I continuously reach it with Claude. I use them both the same way - hands on the wheel and very interactive, small changes and tell them both to update a file to keep up with what’s done and what to do as I test.

Codex gets caught in a loop more often trying to fix an issue. I tell it to summarize the issue, what it’s tried and then I throw Claude at it.

Claude can usually fix it. Once it is fixed, I tell Claude to note in the same file and then go back to Codex

18 comments

raw_anon_1111

Reply
strongpigeon  15 hours ago

The trick to reach the usage limit is to run many agents in parallel. Not that it’s an explicit goal of mine but I keep thinking of this blog post [0] and then try to get Codex to do as much for me as possible in parallel

[0]: http://theoryofconstraints.blogspot.com/2007/06/toc-stories-...

  • raw_anon_1111  15 hours ago

    Telling a bunch of agents to do stuff is like treating it as a senior developer who you trust to take an ambiguous business requirement and letting them use their best judgment and them asking you if they have a question .

    But doing that with AI feels like hiring an outsourcing firm for a project and they come back with an unmaintable mess that’s hard to reason through 5 weeks later.

    I very much micro manage my AI agents and test and validate its output. I treat it like a mid level ticket taker code monkey.

    • bonesss  15 hours ago

      My experience with good outsourcing firms is that they come back with heavily-documented solutions that are 95% of what you actually wanted, leaving you uncomfortably wondering if doing it yourself woulda been better.

      I’m not fully sure what’s worse, something close to garbage with a short shelf life anyone can see, or something so close to usable that it can fully bite me in the ass…

    • strongpigeon  15 hours ago

      I fully believe that if I didn’t review its output and ask it to clean it up it would become unmaintainable real quick. The trick I’ve found though is to be detailed enough in the design from both a technical and non-technical level, sometimes iterating a few time on it with the agent before telling it to go for it (which can easily take 30 minutes)

      That’s how I used to deal with L4, except codex codes much faster (but sometimes in the wrong direction)

      6 replies →

motbus3  16 hours ago

I will say that doing small modifications or asking a bunch of stuff fills the context the same in my observations. It depends on your codebase and the rest of stuff you use (sub agents, skills, etc)

I was once minimising the changes and trying to take the max of it. I did an uncountable numbers of tests and and variations. Didn't really matter much if I told it to do it all or change one line. I feel Claude code tries to fill the context as fast as possible anyway

I am not sure how worth Claude is right now. I still prefer that rather than codex, but I am starting to feel that's just a bias

  • girvo  14 hours ago

    I don’t think it’s bias: I have no love for any of these tools, but in every evaluation we’ve done at work, Opus 4.5 continually comes out ahead in real world performance

    Codex and Gemini are both good, but slower and less “smart” when it comes to our code base

650REDHAIR  15 hours ago

I hit the Claude limit within an hour.

Most of my tokens are used arguing with the hallucinations.

I’ve given up on it.

  • hnsr  15 hours ago

    Do you use Claude Code, or do you use the models from some other tool?

    I find it quite hard to hit the limits with Claude Code, but I have several colleagues complaining a lot about hitting limits and they use Cursor. Recently they also seem to be dealing with poor results (context rot?) a lot, which I haven't really encountered yet.

    I wonder if Claude Code is doing something smart/special

  • TuxSH  15 hours ago

    In my case I've had it (Opus Thinking in CC) hit 80% of the 5-hour limit and 100% of the context window with one single tricky prompt, only to end up with worthless output.

    Codex at least 'knows' to give up in half the time and 1/10th of the limits when that happens.

  • theshrike79  14 hours ago

    I don't want to be That Guy, but if you're "arguing with hallucinations" with an AI Agent in 2026 you're either holding it wrong or you're working on something highly nonstandard.