I'm not likely to install it again in my latest configuration, instead applying some specific tricks to things like `make test` to spit out zero output exit on unsuccessful error codes, that sort of thing. Anecdotally, I see GPT-5.5 often automatically applying context limiting flags to the bash it writes :shrug:

It's hard for me to see this not becoming a pile of slop, but hey, maybe I'm wrong

A lot of Buddhist practice is basically trying to train against immediately collapsing reality into self/other, right/wrong, craving/aversion.

Practicing this with Elon Musk is effectively ultra hard mode.

--

Though I do think there’s a subtle irony here too — the original commenter may simply be describing their own emotional reaction/disillusionment, while your response risks collapsing them into "part of the problem."

Feels like everybody in the thread is pointing at the same tendency from different angles.

To get an idea of what I'm talking about, you could install https://github.com/obra/superpowers/ into both Codex and Claude Code -- You'll find that the behavior is remarkably similar if you A/B compare them on the same problems. CC occasionally misses things that Codex gets and vice versa.

Overall the output structure and final code is remarkably similar... Which is pretty different than if you just run them with their default system prompts. I'd throw codex out the window with its default outputs.

*But* I don't work with the defaults -- I work with my own prompt framework based off of superpowers.

Given sufficient prompt scaffolding, I've found the models relatively interchangeable -- _I might_ be getting some of this for free by basing my own system off of superpowers which is used across various harnesses -- In other words achieving this kind of portability may be a lot harder than it looks and I'm benefiting from other people's work.

From my perspective, Claude Code is decidedly not better than Codex. They’re slightly different and work better together. I would have no issues dropping CC entirely and using codex 100%.

If you’re working off of “defaults”, in other words no custom prompting, Claude Code does perform a lot better out of the box. I think this matters, but if you’re a professional software developer, I’d make the case that you should be owning your tools and moving beyond the baked in prompts.

How is it different?

Sorry about all the broken plastic on the trim -- That's also very familiar...

I work in healthcare RCM. I have no trouble believing the staff here that nothing in their system works.

I know the authors of Skyvern are around here sometimes -- How do you think about code generation with vision based approaches to agentic browser use like OpenAI's Operator, Claude Computer Use and Magnitude?

From my POV, I think the vision based approaches are superior, but they are less amenable to codegen IMO.

CV and direct mouse/kb interactions are the “base” interface, so if you solve this problem, you unlock just about every automation usecase.

(I agree that if you can get good, unambiguous, actionable context from accessibility/automation trees, that’s going to be superior)

If you care about understanding relative performance between models for solving known problems and producing correct output format, it's pretty useful.

- Even for well-known problems, we see a large distribution of quality between models (5 to 75% correctness) - Additionally, we see a large distribution of model's ability to produce responses in formats they were instructed in

At the end of the day, benchmarks are pretty fuzzy, but I always welcome a formalized benchmark as a means to understand model performance over vibe checking.