Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Claude Code on the web

anthropic.com

Post reply on HN
Print view View original

131–140 of 405 posts

Re: Claude Code on the web

#131

Post by hnidiots3 »

I wonder why people don’t just use Amp Code and use the Oracle.

It’s Sonnet 4.5 + GPT-5 working together.

Codex just isn’t as good as people make it out to be. OpenAI seems to train on a lot of JavaScript/Tailwind to make visuals look more impressive but when it comes to actual backend work it just fails more than it succeeds. Sonnet is much better at chewing through tasks and GPT 5 is great at consulting planning and analysis.

Using Amp and asking it to check everything with the oracle leads to superior results.

But no one on HN has heard of it. I’m guessing HN hates twitter?

Re: Claude Code on the web

#132

Post by maherbeg »

mmaunder wrote: post #80
maherbeg wrote: post #45

Earlier quoted context omitted.

Yeah this has been my experience as well. The Claude Code UI is still so much better, and the permissioning policy system is much better. Though I'm working on closing that gap by writing a custom policy https://github.com/openai/codex/blob/main/codex-rs/execpolic... Kinda sick of Codex asking for approval to run tests for each test instance

Ah the tension between cybersecurity best practices and productivity is brutal right now.

lol yeah, but mostly just want to allow more types of reads for getting context, and primarily for test running / linting etc. I shouldn't have to approve every invocation of `pytest` or `bazel test`.

Re: Claude Code on the web

#133

Post by mmaunder »

bgirard wrote: post #81
mmaunder wrote: post #74

Earlier quoted context omitted.

Yes. Sadly. And it really does make me sad. I was rooting for Anthropic. Still kinda am.

I have a very similar experience. I was heavily invested in Anthropic/Claude Code, and even after Sonnet 4.5, I'm finding that Codex is performing much better for my game development project.

It seems particularly good at high performance programming in low level languages.

Re: Claude Code on the web

#134

Post by intellectronica »

shmoogy wrote: post #118
intellectronica wrote: post #99

Earlier quoted context omitted.

Codex works much better for long-running tasks that require a lot of planning and deep understanding. Claude, especially 4.5 Sonnet, is a lot nicer to interact with, so it may be a better choice in cases where you are co-working with the agent. Its output is nicer, it "improvises" really well even if you give it only vague prompts. That's valueable for interactive use. But for delegating complete tasks, Codex is far…

Can / Does Codex actually check docker logs and other things for feedback while iterating on something that isnt working ? That is where the true magic of Claude comes for me. Often things cant be one shot, but being able to iteratively check logs, make an adjustment, rebuild the docker containers, send a curl, and confirm fixed is huge improvement.

Yes, in this regard it's very similar. It works as an agent and does whatever you need it to do to complete the task. In comparison to Claude it tends to plan more and improvise less.

Re: Claude Code on the web

#135

Post by losteric »

myko wrote: post #20
dingnuts wrote: post #11

Earlier quoted context omitted.

The person driving is the one that learns the most in pair programming. In the scenario you've described, that would be Claude. LLMs don't learn. Doesn't CC sometimes take twenty, thirty minutes to return an attempt? I wouldn't know, because I'm not rich and my employer has decided CC is too expensive, but I wonder what you would do with your pair programming partner while you wait. The bosses would like to think we'…

> Nah, just go play air hockey until your boss realizes Claude is what they need, not you. This is a depressing comment. I am apprehensive about the future of software development in this milieu. I've pumped out a ~15,000 line application heavily utilizing Claude Code over a few days that seems to work, but I don't know how much to trust it. Certainly part of the fun of building something was missing during that proj…

> I've pumped out a ~15,000 line application heavily utilizing Claude Code over a few days that seems to work, but I don't know how much to trust it.

Can I ask what you built?

Re: Claude Code on the web

#136

Post by luisml77 »

mdeeks wrote: post #89

I feel like these background agents still aren't doing what I want from a developer experience perspective. Running in an inaccessible environment that pushes random things to branches that I then have to checkout locally doesn't feel great. AI coding should be tightly in the inner dev loop! PRs are a bad way to review and iterate on code. They are a last line of defense, not the primary way to develop. Give me an is…

I agree and I also think the problem is deeper than that. It's about not being able to do most code testing and debugging remotely. You can't really test anything remotely really... Its in an ephemeral container without any of your data, just your repo. You can't have the model do npm run dev and browse to see the webpage, click around, etc. You can't compile or run anything heavy, you can't persist data across sessions/days, etc.

I like the idea of background agents running in the cloud but it has to be a more persistent environment. It also has to run on a GUI so it can develop web applications or run the programs we are developing, and run them properly with the GUI and requiring clicking around, typing things etc. Computer use, is what we need. But that would probably be too expensive to serve to the masses with the current models

Re: Claude Code on the web

#137

Post by Aeolun »

mmaunder wrote: post #70
bcrosby95 wrote: post #66

Earlier quoted context omitted.

Yeah, after correcting it several times I've gotten Claude Code to tell me it didn't have the expertise to work in one of my problem domains. It was kinda surprising but also kinda refreshing that it knew when to give up. For better or worse I haven't noticed similar things with Codex.

I've chosen problems with non-negotiable outcomes. In other words, problem domains where you either are able to clearly accomplish the very hard thing, or not, and there's no grey area. I've purposely chosen these kinds of problems to prove what AI agents are capable of, so that there is no debate in my mind. And with Codex I've accomplished the previously impossible. Unambiguously. Codex did this. Claude gave up. It…

I don’t think Claude is actually incapable, you just spend a lot of time telling it to yes, please actually do the difficult thing. Do not give up halfway through.

Codex says “This is a lot of work, let me plan really well.”

Claude says “This is a lot of work, let me step back and do something completely different that you didn’t ask for.”

Re: Claude Code on the web

#139

Post by didgeoridoo »

hmokiguess wrote: post #47
ea016 wrote: post #13

No relations to them, but I've started using Happy[0]'s iOS app to start and continue Claude Code sessions on my iPhone. It allows me to run sessions on a custom environment, like a machine with a GPU to train models [0] https://github.com/slopus/happy/

This seems to be the only solution still if using bedrock or direct API access instead of Pro / Max plan, the Claude Code for Web doesn't seem to let you use it that way.

You can log in to your CC instance however you like, including via Pro/Max. Happy just wraps it and provides remote access with a much better UI than using a phone-based terminal app.

Re: Claude Code on the web

#140

Post by simonw »

nextworddev wrote: post #138

Developers may want to deny this, but it's getting dangerously close to maybe replacing 30% of developers

I continue to believe that making developers 2-3x times more productive makes those developers 2-3x more valuable, and the smart thing for companies to do is to take on 2-3x times the amount of work, or hire MORE developers and finally start crunching through their inevitably years-long backlogs.
Post reply on HN