Hacker News Reader

Comment by mrcwinn | Hacker News Reader

I went from years of vscode to "Cursor is the future" to never using Cursor at all. Claude Code, even with new limits, is just too good. If I were to switch to gpt-5, why wouldn't I just use Codex? I'm struggling to understand the value of what they're presenting.

View on HN

I find the Codex CLI to be the worst of the CLI tools I’ve used (including, but not limited to, Claude Code, Gemini, Aider). There’s something about it that makes it clunky. Haven’t tried Cursor CLI yet though.

We (Codex) shipped a pretty large CLI update today and have many more improvements coming. Give it a try if you haven't.

https://x.com/OpenAIDevs/status/1953559797883891735 (0.19 now)

Thanks for the heads up. I’ll check it out!

I don’t visit Twitter links. Why not a link to the GitHub changelog?

Also, as an aside since you are on the team - the organization verification is frustrating in that the docs indicate:

>You must not have recently verified another organization, as each ID can only verify one organization every 90 days.

I champion OpenAI at my work, so naturally I’d be the one to verify there. But I apparently can’t, because I verify for my personal-led org. That gets in the way of me proselytizing gpt-5 based coding tools (such as, possibly, Codex CLI).

Huge +1 to this. We have two orgs at work (for separate budget/rate limit blast radii) and had to get two people to verify this morning…

Another +1 to this. Some of us are unwilling to click on a Twitter link. Link to changelog would be more appropriate.

Besides being unwilling, it is unusable if you don’t have an account. It’s as bad as linking to Facebook.

nice! auto-updates like open code could help to not have to remember to update

loving the animations and todos so far

also gpt-5 is just great at agentic stuff

Tried your latest version - thanks for posting about it.

Codex needs plan mode (shift-tab in Claude Code)

And Codex needs the prompt to always be available. So you can type to the model while it’s working & have it eventually receive the message and act on it, instead of having to Ctrl-C to interrupt it before typing. Claude Code’s prompt is always ready to type at - you can type while it is working. That goes a long way towards it feeling like it cares about the user.

Thanks for mentioning this. These are the kinds of features that are 100% required for me to even consider Codex.

> Use with your ChatGPT plan

It's asking me to buy credits but I'm already on Plus?

Got past that but now hitting the bad request error that lots of people seem to have

codex login does not seem to work with ChatGPT Plus.

Getting this error:

Error response Error code: 500

Message: Token exchange failed: <urlopen error [SSL: CERTIFICATE_VERIFY_FAILED] certificate verify failed: unable to get local issuer certificate (_ssl.c:1006)>.

Error code explanation: 500 - Server got itself in trouble.

https://github.com/openai/codex/issues/1243

I am and many other similar users would love to move from Claude Code to Codex, but this is preventing us from doing it.

Hmm, I just upgraded my local codex-cli to 0.19.0 and had to smoke the ~/.codex directory before I could again try to login. Once past that, everything worked fine (so far at least), and managed to get it to connect to my ChatGPT Pro account.

I was actually eager to try it today, but it've just failed with some issue which is open since July 8 (https://github.com/openai/codex/issues/1481) and I gave up on it.

I tried it out and it's definitely much better. A couple requests: - Let me change my setting for command running approval, or make approval fine-grained by command - Support vim motions

Why is Claude Code better than Cursor?

In my experience, it is much better at tool-calling, which is huge when we're talking about agentic coding. It also seems to do a better job of keeping things cleaning and not going off on tangents for anything that isn't accomplished in one shot.

I have had the exact opposite experience. Claude Code in any meaningful codebase for me gets stuck in loops of doing the wrong thing. Then when that doesn't work it deletes files and makes its own that don't have the problem it's encountering.

Cursor on the other hand, especially with GPT-5 today but typically with Sonnet 4.1, has been a workhorse at my company for months. I have never had Claude Code complete a meaningful ticket once. Even a small thing like fixing a small bug or updating the documentation on the site.

Would love any tips on how to make Claude Code not a complete waste of electricity.

Better prompts?

> Better prompts?

I think you're right.

People getting really poor results probably don't recognize that their prompts aren't very good.

I think some users make assumptions about what the model can't do before they even try, so their prompts don't take advantage of all the capabilities the model provides.

I don’t really have a problem prompting cursor with the same models. But I have no doubt my prompts could be improved

If you don’t know how to divide a problem up given a toolset you won’t be able to solve it regardless of what those tools are. Maybe Cursor’s interface is more intuitive for you.

The problems I’ve given CC are things that are incredibly simple and basic. Things I knew how to fix immediately. I would tell it the gilt to change and how to change it. And it will get lost when the types are incorrect, or when it causes a test to fail. It will like just delete the test.

I don’t doubt I could improve my prompts but I don’t have those same prompting problems with cursor.

Don’t know what to say. Those problems are exactly why I left Cursor behind—I don’t really encounter those issues with Claude Code (despite only using Anthropoc models in either case).

> Cursor on the other hand, especially with GPT-5 today but typically with Sonnet 4.1

You probably mean Opus 4.1; there's no Sonnet 4.1 yet.

Yes that’s correct.

Opposite experience. I worked with Claude code a lot, then switched to Cursor and then tried to switch back and discovered that CC often gets stuck in loops. Cursor just works. It definitely helps that I can switch the foundational models in Cursor when it gets stuck.

Because iterating multiple sessions through multiple terminals is obviously more efficient and seamless than interacting thought a scuffed IDE side panel ui.

Is this sarcasm? I used Claude Code via a VSCode side panel.

That's a funny response I have to admit. No 100% serious, I don't see the advantage.

My company has a huge codebase, for me cursor would freeze up / not find relevant files. Claude code seems able to find the right files by itself.

I seem to always have better outcomes with Claude code.

Cursor and these tools evolve quite fast.

The Cursor you used a month ago is not the one you get now.

Just saying that because in this space you should always compare latest X with latest Y.

I too switched weeks ago to Claude Code. Then between the times I am out of tokens I launch Cursor and actually find it...better than I remember if not on par with Claude Code (the model and quality of prompts/context matters more than the IDE/CLI tool used too).

How does Claude Code handle editing multiple files? My understanding is that it's CLI based, so it edits a bunch of files on it's own, so how do you accept/reject and rollback changes?

I use git and when necessary, git worktrees to keep parallel Claudes from bothering/actively-undoing each other’s edits (esp when an edit requires multiple files of changes, so the tree is temporarily broken).

> how do you accept/reject and rollback changes?

Git

What I have found Claude Code is extremely good at is that it makes one change at a time, gives you a chance to read the code its changing, and lets you give feedback in real time and steer it properly. I find the mental load with this method to be MUCH lower than Cursor or any of the other tools which give you two very different options: "Ask" mode which dumps a ton of suggestions on your and then requires semi-manual implementation, or "Agent" mode which dumps a ton of actual changes on you and requires your inspection and feedback and roll-backs, etc.

This may not work for everyone, but as a solo dev who wants to keep a real mental model of my work (and not let it get polluted with AI slop), the Claude Code approach just works really well for me. It's like having a coding partner who can iterate and change direction as you talk, not a junior dev who dumps a pile of code on your plate without discussion.

+1 to this. Cursors Agent feels too difficult to wrangle. CC is easier to monitor.

I set rules for cursor that when i need cursor to make changes, it will do it with plan-scheme-execute mode. Everything is clear especially when it prompts questiones for you to make scheme as you wish. Today cursor's gpt-5-fast-high model exploits this working style to its extend. This model gives the most detailed scheme for me to customize and i benefit a lot.

CC just feeds the whole codebase and entire files into the model, no RAG, nothing in the way. It works substantially better because of that, but it's $expensive$.

That's not true. It uses CLI tools (e.g. find, grep) to find the relevant code from the codebase.

The more stuff you put in the context the worse models perform. All of them.

Larger context is a bonus sometimes, but in general you're degrading the quality of the output by a lot.

Precise prompting and context management is still very important.

Claude Code has some non-LLM magic in it that just makes it better for code in general, despite (or because of) having minimal IDE integration.

The values is that, to use cursor, you don't need anymore to switch your IDE (unless you were already using vscode). You can keep your preferred IDE and run the agent in the terminal. IDE Is for humans, agents need only a terminal for running.

I find it really difficult to take this stuff seriously when the proponents of these tools are constantly flitting from tool to tool. There's tools and paradigms I've used for years at a time. Some things like vim have been used for decades.

This all just feels profoundly immature. You tell us one thing then two months later you're on to the next thing. There's no sense of mastery here.

is the multi-file editing as easily understood in the Claude Code cli? or am I missing something? I always felt like Claude Code CLI was OK as a tool yet powered by one of the best AI models. I stuck with Cursor b/c of the multi-file editing, agent mode, etc. I'm very very willing to leave Cursor because they've recently been scummy about their pricing changes.

Reply on news.ycombinator.com