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

6 weeks of Claude Code

blog.puzzmo.com

Post reply on HN
Print view View original

421–430 of 603 posts

Re: 6 weeks of Claude Code

#421

Post by shortrounddev2 »

exe34 wrote: post #403
shortrounddev2 wrote: post #293

Earlier quoted context omitted.

It is not coding if you use natural human language

I think after all the goalpost moving, we have to ask - why the bitflip does it matter what we call it? Some people are getting a lot of work done using LLMs. Some of us are using it on occasion to handle thing we don't understand deeply but can trivially verify. Some of us are using it out of laziness because it helps with boilerplate. Everyone who is using it outside of occasional tests is doing it because they fin…

If my boss came to me and said "hey we're going to start vibe coding everything st work from now on. You can manually edit code but claude code needs to be your primary driver from now on" I would quit and find a new career. I enjoy coding. I like solving puzzles using the specifics of a language syntax. I write libraries and APIs and I put a great deal of effort into making sure the interface is usable by a human being.

If we get to the point where we are no longer coding, we are just describing things in product language to a computer and letting it do all the real work, then I will find a more fulfilling career because this ain't it

Re: 6 weeks of Claude Code

#422

Post by linuxhansl »

It seems to be great at writing tests, spitting out UI code, and many other things where there are many examples around.

Among other things I work on database optimizers and there Claude fails spectacularly. It produces wrong code, fails to find the right places where to hook up an abstraction, overlooks affects on other parts of the code, and generally confidently proposes changes that simply do not work at all (to put it mildly).

Your mileage may vary... It seems to be depend heavily on the amount of existing (open) code around.

Re: 6 weeks of Claude Code

#423

Post by itsalotoffun »

pmg101 wrote: post #398
ChuckMcM wrote: post #328

Earlier quoted context omitted.

An observation. If we stipulate that this is true that a 'senior developer' benefits from Claude Code but a junior developer do not. Then I'm wondering if that creates this gap where you have a bunch of newly minted '10x' engineers who are doing the work that a bunch of junior devs helped with, and now you're not training any new junior devs because they are unemployable. Is that correct?

It already was the case wasn't it, that you could either get one senior dev to build your thing in a week, or give them a team of juniors and it would take the whole team 4 weeks and be worse. Yet somehow companies continued to opt for the second approach. Something to do with status from headcount?

n=1 but my experience is the ratio of what'd I'd class "senior" devs (per the example given) to everyone else is comfortably 10:1.

Re: 6 weeks of Claude Code

#424

Post by Archer6621 »

ako wrote: post #412
nosianu wrote: post #411

Earlier quoted context omitted.

Brains are not computers and we don't learn by being given abstract rules. We also don't learn nearly as well from class room teaching as we do from doing things IRL for a real purpose - the brain always knows the difference and that the (real, non-artificially created) stakes are low in a teaching environment. That's also the huge difference between AI and brains: AI does not work on the real world but on our commun…

And that is the best thing about AI, it allows you to do and try so much more in the limited time you have. If you have an idea, build it with AI, test it, see where it breaks. AI is going to be a big boost for education, because it allows for so much more experimentation and hands-on.

By using AI, you learn how to use AI, not necessarily how to build architecturally sound and maintainable software, so being able to do much more in a limited amount of time will not necessarily make you a more knowledgeable programmer, or at least that knowledge will most likely only be surface-level pattern recognition. It still needs to be combined with hands-on building your own thing, to truly understand the nuts and bolts of such projects.

Re: 6 weeks of Claude Code

#425

Post by itsalotoffun »

zmmmmm wrote: post #279
phito wrote: post #185

Earlier quoted context omitted.

It's alright until you have a bug the LLM can't solve, then you have to go in the code yourself and you realize what a mess it has made.

Perhaps you set a very high quality bar, but I don't see the LLMs creating messy code. If anything, they are far more diligent in structuring it well and making it logically sequenced and clear than I would be. For example, very often I name a variable slightly incorrectly at the start and realise it should be just slightly different at the end and only occasionally do I bother to go rename it everywhere. Even with a…

> hey are far more diligent in structuring it well and making it logically sequenced and clear than I would be

Yes, with the caveat: only on the first/zeroth shot. But even when they keep most/all of the code in context if you vibe code without incredibly strict structuring/guardrails, by the time you are 3-4 shots in, the model has "forgotten" the original arch, is duplicating data structures for what it needs _this_ shot and will gleefully end up with amnesiac-level repetitions, duplicate code that does "mostly the same" thing, all of which acts as further poison for progress. The deeper you go without human intervention the worse this gets.

You can go the other way, and it really does work. Setup strict types, clear patterns, clear structures. And intervene to explain + direct. The type of things senior engineers push back on in junior PRs. "Why didn't you just extend this existing data structure and factor that call into the trivially obvious extension of XYZ??".

"You're absolutely right!" etc.

Re: 6 weeks of Claude Code

#426

Post by wordofx »

lukaslalinsky wrote: post #337
wordofx wrote: post #327

Earlier quoted context omitted.

I don’t know why amp isn’t talked about more. It’s better than Claude code.

It burns through credit too quickly. As a previous Sourcegraph Cody user, I was trying Amp first, but I've spent tens of dollars every day for the trial, and that was with an eye on the usage. It felt horrible seeing how I pay mostly for it's mistakes and the time it takes debugging. With CC, I can let go of the anxiety. I get several hours a day out of the Claude Pro plan and that's mostly good enough for now. If it…

My AMP bill is less than Claude Code but I’m getting more work done.

Re: 6 weeks of Claude Code

#427

Post by d4rkp4ttern »

arrowsmith wrote: post #101

The real power of Claude Code comes when you realise it can do far more than just write code. It can, in fact, control your entire computer. If there's a CLI tool, Claude can run it. If there's not a CLI tool... ask Claude anyway, you might be surprised. E.g. I've used Claude to crop and resize images, rip MP3s from YouTube videos, trim silence from audio files, the list goes on. It saves me incredible amounts of tim…

Beyond just running CLI commands, you can have CC interact with those, e.g I built this little tool that gives CC a Tmux-cli command (a convenience wrapper around Tmux) that lets it interact with CLI applications and monitor them etc:

https://github.com/pchalasani/claude-code-tools

For example this lets CC spawn another CC instance and give it a task (way better than the built-in spawn-and-let-go black box), or interact with CLI scripts that expect user input, or use debuggers like Pdb for token-efficient debugging and code-understanding, etc.

Re: 6 weeks of Claude Code

#428

Post by mentos »

maille wrote: post #245
mentos wrote: post #12

Earlier quoted context omitted.

Cursor has made writing C++ like a scripting language for me. I no longer wrestle with arcane error messages, they go straight into Cursor and I ask it to resolve and then from its solution I learn what my error was.

Can you really use Cursor for CPP ? Would you mind describing your setup? How better is it than copilot or windsurf?

Open your C++ project in Cursor. Before anything else ask it to review the codebase and tell you what the codebase does so you can understand its power. Play around asking it to find sections of the code that handle functionality for certain features. It should really impress you.

Continue to work on it in your preferred IDE let’s say Visual Studio. When you hit your first compile error, just for fun even if you understand the error, copy and paste it into Cursor and ask it to help you understand the cause and propose a solution. Ask it to implement it, attempt to compile, give it back any further errors that its solution may have to review and fix. You will eventually compile.

Then before you go back to work writing your next task, ask Cursor to propose how it might complete the task. After the proposal review and either tell it to proceed to implement or suggest tweaks or better alternatives. For complex tasks try setting the model manually to o3 and rerunning the same prompt and you can see how it thinks much better and can one shot solutions to complex errors. I try to use auto and if it fails on more complex tasks I resubmit the original query with o3. If o3 fails then you may have to gather more context by hand and really hold its hand through the chain of reasoning. That’s for a future post.

More advanced: Create a build.bat script that Cursor can run after it has implemented code to run and see its own errors so you can avoid the copy paste round trip. (Look into Cursor rules for this but a rule prompt that says 'after implementing any significant code changes please run .\build.bat and review and fix any further errors') This simple efficiency should allow you to experience the real productivity behind Cursor where you’re no longer dying the death of 1000 cuts losing a minute here or a minute there on rote time consuming steps and you can start to operate from a higher natural language level and really feel the ‘flow’.

Typing out the code is just an annoying implementation detail. You may feel ‘competency leaving your fingers’ as DHH might say but I’d argue you can feel your ass filling up with rocket fuel.

Re: 6 weeks of Claude Code

#429

Post by itsalotoffun »

arevno wrote: post #55
jeswin wrote: post #19

Claude Code is ahead of anything else, in a very noticeable way. (I've been writing my own cli tooling for AI codegen from 2023 - and in that journey I've tried most of the options out there. It has been a big part of my work - so that's how I know.) I agree with many things that the author is doing: 1. Monorepos can save time 2. Start with a good spec. Spend enough time on the spec. You can get AI to write most of t…

> 1. Monorepos can save time Yes they can save you some time, but at the cost of Claude's time and lots of tokens making tool calls attempting to find what it needs to find. Aider is much nicer, from the standpoint that you can add the files you need it to know about, and send it off to do its thing. I still don't understand why Claude is more popular than Aider, which is by nearly every measure a better tool, and ca…

Because it works.

Honestly, it's just this. "Claude the bar button on foo modal is broken with a failed splork". And CC hunts down foo.ts, traces that it's an API call to query.ts, pulls in the associated linked model, traces the api/slork.go and will as often as not end up with "I've found the issue!" and fix it. On a one sentence prompt. I think it's called an "Oh fuck" moment the first time you see this work. And it works remarkably reliably. [handwave caveats, stupid llms, etc]

Re: 6 weeks of Claude Code

#430

Post by d4rkp4ttern »

Agreed. CC lets you attempt things that you wouldn’t have dared to try. For example here are two things I recently added to the Langroid LLM agent framework with CC help:

Nice collapsible HTML logs of agent conversations (inspired by Mario Zechner’s Claude-trace), which took a couple hours of iterations, involving HTML/js/CSS:

https://langroid.github.io/langroid/notes/html-logger/

A migration from Pydantic-v1 to v2, which took around 7 hours of iterations (would have taken a week at least if I even tried it manually and still probably wouldn’t have been as bullet-proof):

https://github.com/langroid/langroid/releases/tag/0.59.0-b3

Post reply on HN