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

381–390 of 603 posts

Re: 6 weeks of Claude Code

#381

Post by ksahin »

AndyNemmity wrote: post #353
hansvm wrote: post #351

Earlier quoted context omitted.

I don't think that's true with current-gen models. You can even go so far as to write pseudocode for the LLM to translate to a real language, and for anything out-of-the-box my experience is that it will blatantly ignore your instructions. A baseline-competent junior at least has the context to know that if there are 5 steps listed and they only did 3 then there's probably a problem. Prompting an LLM is definitely a…

I don't believe it is good enough, it's also not as relevant. my prompts have gotten less and less. It's the hooks and subagents, and using the tools that matter far more. This is a thread about claude code. the other LLms don't matter. Nothing ever blatantly ignores my instructions in claude code. that's a thing of the past. Of course, not using claude code, for sure. But all my instructions are followed with my set…

What's your experience with sub-agents ? is it really improving the output?

Did you use the one suggested here https://docs.anthropic.com/en/docs/claude-code/sub-agents or have you created custom ones?

Re: 6 weeks of Claude Code

#382

Post by AndyNemmity »

ksahin wrote: post #381
AndyNemmity wrote: post #353

Earlier quoted context omitted.

I don't believe it is good enough, it's also not as relevant. my prompts have gotten less and less. It's the hooks and subagents, and using the tools that matter far more. This is a thread about claude code. the other LLms don't matter. Nothing ever blatantly ignores my instructions in claude code. that's a thing of the past. Of course, not using claude code, for sure. But all my instructions are followed with my set…

What's your experience with sub-agents ? is it really improving the output? Did you use the one suggested here https://docs.anthropic.com/en/docs/claude-code/sub-agents or have you created custom ones?

The suggested ones are terrible, and it's guidance is terrible.

Last weekend I ran head to head tests of agents against each other with a variety of ideas, and selected the best one, and did it again. It has caused me to have a very specific subagent system, and I have an agent who creates those.

Re: 6 weeks of Claude Code

#383

Post by aryonoco »

leptons wrote: post #342
awesome_dude wrote: post #244

Earlier quoted context omitted.

No, not at all. In both all you need is the ability to communicate to the machine in a way that the machine can convert your ideas into actions. The restricted language of a compiler is a handicap, not evidence of a skill - we've been saying forever that "Natural Language" compilers would be a game changer, and that's all that an AI really is Edit: It appears that this discussion is going to end up with a definition…

Writing code that gets compiled is deterministic, but asking an LLM to produce code is a non-deterministic guessing game. Notice how nobody is skeptical of compilers?

No one is skeptical of compilers?! I guess you haven’t met many old fashioned C systems programmers, who go out of their way to disable compiler optimisations as much as they can because “it just produces garbage”.

Every generation, we seem to add a level of abstraction conceding because for most of us, it enhances productivity. And every generation, there is a crowd who rails against the new abstraction, mostly unaware of all of the levels of abstraction they already use in their coding.

Re: 6 weeks of Claude Code

#384

Post by globular-toast »

M4v3R wrote: post #164
lbrito wrote: post #162

It's less that I'm a skeptic, but more that I'm finding I intensely abhor the world we're building for ourselves with these tools (which I admittedly use a lot).

Why abhor specifically?

Not the GP but we're descending into a world where we just recycle the same "content" over and over. Nothing will be special, there'll be nothing to be proud of. Just constant dopamine hits administered by our overlords. Read Brave New World if you haven't.

Re: 6 weeks of Claude Code

#385

Post by collingreen »

giancarlostoro wrote: post #169
lukaslalinsky wrote: post #57

I have about two weeks of using Claude Code and to be honest, as a vibe coding skeptic, I was amazed. It has a learning curve. You need to learn how to give it proper context, how to chunk up the work, etc. And you need to know how to program, obviously. Asking it to do something you don't know how to do, that's just asking for a disaster. I have more than 25 years of experience, so I'm confident with anything Claude…

If you are a Senior Developer, who is comfortable giving a Junior tips, and then guiding them to fixing them (or just stepping in for a brief moment and writing where they missed something) this is for you. I'm hearing from Senior devs all over thought, that Junior developers are just garbage at it. They product slow, insecure, or just outright awful code with it, and then they PR the code they don't even understand.…

I'm getting my moneys worth having claude write tools. We've reached the dream where I can vibe out some one off software and it's great; today I made two different (shitty but usable!) gui programs in seconds that let me visually describe some test data. The alternative was probably half an hour of putting something together if my first idea was good. Then I deleted them and moved on.

It still writes insane things all the time but I find it really helpful to spit out single use stuff and to brainstorm with. I try to get it to perform tasks I don't know how to accomplish (eg. computer vision experiments) and it never really works out in the end but I often learn something and I'm still very happy with my subscription.

Re: 6 weeks of Claude Code

#386

Post by enobrev »

lukaslalinsky wrote: post #57

I have about two weeks of using Claude Code and to be honest, as a vibe coding skeptic, I was amazed. It has a learning curve. You need to learn how to give it proper context, how to chunk up the work, etc. And you need to know how to program, obviously. Asking it to do something you don't know how to do, that's just asking for a disaster. I have more than 25 years of experience, so I'm confident with anything Claude…

I have a similar amount of engineering experience, was highly skeptical, and I've come to similar conclusions with Claude Code after spending two weeks on a greenfield project (TS api, react-native client, TS/React admin panel).

As I've improved planning and context management, the results have been fairly consistent. As long as I can keep a task within the context window, it does a decent job almost every time. And occasionally I have to have it brute-force its way to green lint/typecheck/tests. That's been one of the biggest speed bumps.

I've found that gemini is great at the occasional detailed code-review to help find glaring issues or things that were missed, but having it implement anything has been severely lacking. I have to literally tell it not to do anything because it will gladly just start writing files on a whim. I generally use the opus model to write detailed plans, sonnet to implement, and then opus and gemini to review and plan refactors.

I'm impressed. The progress is SLOW. I'd have gotten to the stage I'm at in 1/3 to 1/2 the time, likely with fewer tests and significantly less process documentation. But the results are otherwise fairly great. And the learning process has kept me motivated to keep this old side-project moving.

I was switching between two accounts for a week while testing, but in the end upgraded to the $100/month plan and I think I've been rate-limited once since. I don't know if I'll be using this for every-day professional work, but I think it's a great tool for a few categories of work.

Re: 6 weeks of Claude Code

#387

Post by conradfr »

js2 wrote: post #313
giancarlostoro wrote: post #169

Earlier quoted context omitted.

If you are a Senior Developer, who is comfortable giving a Junior tips, and then guiding them to fixing them (or just stepping in for a brief moment and writing where they missed something) this is for you. I'm hearing from Senior devs all over thought, that Junior developers are just garbage at it. They product slow, insecure, or just outright awful code with it, and then they PR the code they don't even understand.…

I've also found it good at catching mistakes and helping write commit messages. "Review the top-most commit. Did I make any mistakes? Did I leave anything out of the commit message?" Sometimes I let it write the message for me: "Write a new commit message for the current commit." I've had to tell it how to write commit messages though. It likes to offer subjective opinions, use superlatives and guess at why something…

How long are you commit messages if you still are ahead after typing all this prompt?

Re: 6 weeks of Claude Code

#388

Post by rhubarbtree »

epolanski wrote: post #190

At this point I am 99% convinced that AI coding skeptics are nothing short of Luddites. They would be like "but a robot will never ever clean a house as well as I would", well, no shit, but they can still do the overwhelming majority of the work very well (or at least as good as you instruct them to) and leave you with details and orchestration.

Genuinely not yet convinced. The CEO of windsurf said the goal for the next year was to get agentic coding to the reliability and maturity that it can be used in production for mature codebases. That rhymes true.

I use autocomplete and chat with LLMs as a substitute for stack overflow. They’re great for that. Beyond that, myself and colleagues have found AI agents are not yet ready for serious work. I want them to be, I really do. But more than that I want our software to be reliable, our code to be robust and understandable, and I don’t want to worry about whether we are painting ourselves into a corner.

We build serious software infrastructure that supports other companies’ software and our biggest headache is supporting code that we built earlier this year using AI. It’s poorly written, full of bugs including problems very hard to spot from the code, and is just incomprehensible for the most part.

Other companies I know are vibe coding, and making great progress, but their software is CRUD SaaS and worst case they could start over. We do not have that luxury.

Re: 6 weeks of Claude Code

#389

Post by awesome_dude »

aryonoco wrote: post #383
leptons wrote: post #342

Earlier quoted context omitted.

Writing code that gets compiled is deterministic, but asking an LLM to produce code is a non-deterministic guessing game. Notice how nobody is skeptical of compilers?

No one is skeptical of compilers?! I guess you haven’t met many old fashioned C systems programmers, who go out of their way to disable compiler optimisations as much as they can because “it just produces garbage”. Every generation, we seem to add a level of abstraction conceding because for most of us, it enhances productivity. And every generation, there is a crowd who rails against the new abstraction, mostly unaw…

Luxury! When I were a lad we didn't have them new fangled compilers, we wrote ASM by hand, because compilers cannot (and still to this day I think) optimise ASM as well as a human

Re: 6 weeks of Claude Code

#390

Post by rhubarbtree »

fdsf111 wrote: post #144

I recently tried a 7-day trial version of Claude Code. I had 3 distinct experiences with it: one obviously positive, one bad, and one neutral-but-trending-positive. The bad experience was asking it to produce a relatively non-trivial feature in an existing Python module. I have a bunch of classes for writing PDF files. Each class corresponds to a page template in a document (TitlePage, StatisticsPage, etc). Under the…

Your bad experience is because AI can’t really reason in general. It gets some kinda reasoning via a transformer, but that’s nothing like the reasoning that goes into the problem you described.

LLMs are great at translation. Turn this English into code, essentially. But ask it to solve a novel problem like that without a description of the solution, how will it approach it? If there’s an example in its training set maybe it can recall it. Otherwise it has no capability to derive a solution.

Post reply on HN