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

I was a top 0.01% Cursor user, then switched to Claude Code 2.0

blog.silennai.com

Post reply on HN
Print view View original

211–220 of 236 posts

Re: I was a top 0.01% Cursor user, then switched to Claude Code 2.0

#211

Post by fennecfoxy »

sailorganymede wrote: post #110

Meanwhile here’s me still just using the ChatGPT web chat asking it for code snippets.

Not only that but I find ChatGPT (and LLMs in general) a great partner to discuss my existing ideas with, with the general idea of getting it to challenge me on my decisions and provide ideas that I can shoot down (or make me think "actually...that could work".

I think LLMs are much more effective when you have a good understanding of a field and can work in the above way; their assistance is useful because I don't have time or brainpower to be an _expert_ in many of the fields I'm _knowledgable_ in.

Re: I was a top 0.01% Cursor user, then switched to Claude Code 2.0

#212

Post by Havoc »

kayson wrote: post #11

Do we really need to qualify our power user level down to 100ppm percentiles...?

It’s weird how it only shows up in LLM space. A lot of technical articles historically attempt to establish credibility by pointing at prior work but pointing at token usage is a new trend.

My heaviest token use thus far was on my first attempt to vibecode a project. It was a lot of code but complete trash

Re: I was a top 0.01% Cursor user, then switched to Claude Code 2.0

#213

Post by dspillett »

ramon156 wrote: post #171
ramon156 wrote: post #170

- AI generated article - Overconfident claims (which are based on solo dev) - Spending an absurd amount on an LLM subscription - No actual details, just buzzwords and generic claims If AI-hype was a person

Also, what do you mean "you do not need to review the code". Why did you even start coding in the first place if this is a positive thing? Did we just stop caring about the art of programming altogether?

> > "you do not need to review the code"

> Did we just stop caring about the art of programming altogether?

Yes. Decades ago for some.

I'm sure there have been a number of significant bugs caused by someone taking work from an outsourced team into production without sufficient review. Or even work from a local junior. Heck, even a local senior!¹

Outsourcing work to GlorifiedPredictiveText and friends should be treated the same was as passing it on to other humans, but at the moment it too often isn't as many have fallen for the marketing. Always remember: the models were trained on public code, and public code is far from always right. And the models hallucinate³ on top of that.

--------

[1] Around this time last year, I was that senior… Fun times. Luckily no permanent damage done², but that'll teach people not to trust me too much!

[2] It wasn't as smooth as I would have hoped, but the roll-back plan worked.

[3] Going back to the analogy of outsourcing to other humans: this is akin to “making shit up as they go along and hoping for the best”, which also very much happens and has happened for decades.

Re: I was a top 0.01% Cursor user, then switched to Claude Code 2.0

#214

Post by xenodium »

forgotpwd16 wrote: post #185
xenodium wrote: post #174

For most of 2025, I ignored popular agents because I wanted to stay within my preferred text editor (Emacs). Thanks to ACP ( https://agentclientprotocol.com ), I no longer live under a rock ;) I built https://github.com/xenodium/agent-shell and now get a native experience. Claude Code works great. If curious what that looks like, I made a video recently https://xenodium.com/bending-emacs-episode-10-agent-shell

This is awesome. Thanks for chatgpt-shell.el too. Could use authinfo for api-keys so don't have to put them inside config (saved in ~/.authinfo, format eqv. to ~/.netrc).

Great to hear it! Thank you =)

Re: I was a top 0.01% Cursor user, then switched to Claude Code 2.0

#215

Post by sroerick »

Alupis wrote: post #153
sunrunner wrote: post #148

Earlier quoted context omitted.

I’m so glad we’ve got domain experts to write those tricky things like left-pad for us. On a more serious note, I do think that the maintenance aspect is a differentiator, and that if it’s something that you end up committing to your codebase then ownership and accountability falls to you. Externally sourced libraries and frameworks ultimately have different owners.

I'm reminded of the recent "vibe coded" OCaml fiasco[1]. In particular, the PR author's response to this question: > Here's my question: why did the files that you submitted name Mark Shinwell as the author? > > Beats me. AI decided to do so and I didn't question it. The same author submitted a similar PR to Julia as well. Both were closed in-part due to the significant maintenance burden these entirely LLM-written P…

This exchange is so funny. That much time around LLMs and no human feedback really seems to have broken that guy's brain.

Re: I was a top 0.01% Cursor user, then switched to Claude Code 2.0

#216

Post by PunchyHamster »

ramon156 wrote: post #170

- AI generated article - Overconfident claims (which are based on solo dev) - Spending an absurd amount on an LLM subscription - No actual details, just buzzwords and generic claims If AI-hype was a person

the oversaturation of articles about AI and often written by AI in tech space is so tiring

Re: I was a top 0.01% Cursor user, then switched to Claude Code 2.0

#217

Post by stkdump »

arowthway wrote: post #189

Influenced by many comments on this site, I recently switched from $10 Github Copilot to $20 Claude Code and so far haven't seen any benefits. Maybe it's because I don't know how to 'close the agentic loop' in this project. I thought it being more agentic meant I can just tell it to research a subsystem on its own and plan the changes I want, but that caused it to spawn 3 subagents and consume the entire 4-hour token…

Ah, that makes sense. After all, Anthropic is selling tokens, so they would like you to consume more of them.

Re: I was a top 0.01% Cursor user, then switched to Claude Code 2.0

#218

Post by lysace »

arowthway wrote: post #189

Influenced by many comments on this site, I recently switched from $10 Github Copilot to $20 Claude Code and so far haven't seen any benefits. Maybe it's because I don't know how to 'close the agentic loop' in this project. I thought it being more agentic meant I can just tell it to research a subsystem on its own and plan the changes I want, but that caused it to spawn 3 subagents and consume the entire 4-hour token…

The $100 Claude plan is the minimum, I feel. Otherwise you run out of tokens way too often.

Re: I was a top 0.01% Cursor user, then switched to Claude Code 2.0

#219

Post by jermaustin1 »

ben_w wrote: post #209
bandrami wrote: post #179

Earlier quoted context omitted.

What's the model for integration and maintenance in that case, then? Just re-create the app every time?

Ask your manager how they get humans to do those things, and copy that process. So, get a bug tracker, track those bugs, tell Claude to pick tasks off it etc. I'm not claiming this actually works, I've not tried it, I don't know how good it is for large brownfield projects, but that's the general sentiment I see.

I've tried it. It can work. My prompt was "Use the gh commandline tool to get the issues for the current repository, and work on them in order, with bugs taking priority."

Elsewhere there are steps for how to develop: 1. Create new branch for the feature you are working on; 2. implement the feature fully, thinking hard when you need to (toolcall think(low, med, high) switches the reasoning level);

That said, it also failed a lot.

Re: I was a top 0.01% Cursor user, then switched to Claude Code 2.0

#220

Post by HNisCIS »

SilenN wrote: post #166
HNisCIS wrote: post #152

Earlier quoted context omitted.

My brain kinda shuts off when I read this stuff because I know it's more of a "text shaped object" than actual text. I find LLM text to be too fluffy and information sparse much of the time so I subconsciously start skimming it more than actually reading it.

Outside of the story at the start I intentionally tried to make it information dense. Would appreciate feedback!

The story itself is fine it's more the trained response to LLM voice that causes problems
Post reply on HN