What makes Claude Code so damn good • Hacker News Board

Re: What makes Claude Code so damn good

#41

Post by SXX » Sat, Aug 23, 2025, 8:16 PM UTC

diego_sandoval wrote: post #17

It shocks me when people say that LLMs don't make them more productive, because my experience has been the complete opposite, especially with Claude Code. Either I'm worse than then at programming, to the point that I find an LLM useful and they don't, or they don't know how to use LLMs for coding.

This heavily depends on what project and stack you working on. LLMs are amazing for building MVPs or self-contained micro-services on modern, popular and well-defined stacks. Every single dependency, legacy or proprietary library and every extra MCP make it less usable. It get's much worse if codebase itself is legacy unless you can literally upload documentation for each used API into context.

A lot of programmers work on maintaining huge monolith codebases, built on top of 10-years old tech using obscure proprietary dependencies. Usually they dont have most of the code to begin with and APIs are often not well documented.

Re: What makes Claude Code so damn good

#42

Post by koakuma-chan » Sat, Aug 23, 2025, 8:17 PM UTC

dingnuts wrote: post #15

ebzlo wrote: post #9

Earlier quoted context omitted.

Yes technically it is RAG, but a lot of the community is associating RAG with vector search specifically.

it does? why? the term RAG as I understand it leaves the methodology for retrieval vague so that different techniques can be used depending on the, er, context.. which makes a lot more sense to me

> why?

Hype. There's nothing wrong with using, e.g., full-text search for RAG.

Re: What makes Claude Code so damn good

#43

Post by cpursley » Sat, Aug 23, 2025, 8:18 PM UTC

diego_sandoval wrote: post #17

It shocks me when people say that LLMs don't make them more productive, because my experience has been the complete opposite, especially with Claude Code. Either I'm worse than then at programming, to the point that I find an LLM useful and they don't, or they don't know how to use LLMs for coding.

I feel like I could have written this myself; I'm truly dumbfounded. Maybe I am just a crappy coder but I don't think I'd be getting such good results with Claude Code if I were.

Re: What makes Claude Code so damn good

#44

Post by ec109685 » Sat, Aug 23, 2025, 8:18 PM UTC

OtherShrezzing wrote: post #25

I think it’s just that the base model is good at real world coding tasks - as opposed to the types of coding tasks in the common benchmarks. If you use GitHub Copilot - which has its own system level prompts - you can hotswap between models, and Claude outperforms OpenAI’s and Google’s models by such a large margin that the others are functionally useless in comparison.

Anthropic has opportunities to optimize their models / prompts during reinforcement learning, so the advice from the article to stay close to what works in Claude code is valid and probably has more applicability for Anthropic models than applying the same techniques to others.

With a subscription plan, Anthropic is highly incentivized to be efficient in their loops beyond just making it a better experience for users.

Re: What makes Claude Code so damn good

#45

Post by amelius » Sat, Aug 23, 2025, 8:18 PM UTC

gervwyk wrote: post #35

We’re considering building a coding agent for Lowdefy[1], a framework that lets you build web apps with YAML config. For those who’ve built coding agents: do you think LLMs are better suited for generating structured config vs. raw code? My theory is that agents producing valid YAML/JSON schemas could be more reliable than code generation. The output is constrained, easier to validate, and when it breaks, you can act…

How do you specify callbacks?

Config files should be mature programming languages, not Yaml/Json files.

Re: What makes Claude Code so damn good

#46

Post by syntaxing » Sat, Aug 23, 2025, 8:19 PM UTC

StephenAshmore wrote: post #31

syntaxing wrote: post #16

I don’t know if I’m doing something wrong. I was using Sonnet 4 with GitHub Copilot. Recently a week ago switched to Claude Code. I find GitHub Copilot solves problem and bugs way better than Claude Code. For some reason, Claude Code seems very lazy. Has anyone experience something similar?

It may be a configuration thing. I've found quite the opposite. Github Copilot using Sonnet 4 will not manage context very well, quite frequently resorting to running terminal commands to search for code even when I gave it the exact file it's looking for in the copilot context. Claude code, for me, is usually much smarter when it comes to reading code and then applying changes across a lot of files. I also have it i…

I do agree with you, Github Copilot uses more tokens like you mentioned with redundant searches. But at the end of the day, it solves the problem. Not sure if the cost out weights the benefit though compared to Claude Claude. Going to try Claude Code more and see if I'm prompting it incorrectly.

Re: What makes Claude Code so damn good

#47

Post by socalgal2 » Sat, Aug 23, 2025, 8:19 PM UTC

diego_sandoval wrote: post #17

It shocks me when people say that LLMs don't make them more productive, because my experience has been the complete opposite, especially with Claude Code. Either I'm worse than then at programming, to the point that I find an LLM useful and they don't, or they don't know how to use LLMs for coding.

I’m trying to learn jj. Both Gemini and ChatGPT gave me incorrect instructions 4 of 5 times

https://jj-vcs.github.io/jj/

Re: What makes Claude Code so damn good

#48

Post by SXX » Sat, Aug 23, 2025, 8:19 PM UTC

tjr wrote: post #21

diego_sandoval wrote: post #17

It shocks me when people say that LLMs don't make them more productive, because my experience has been the complete opposite, especially with Claude Code. Either I'm worse than then at programming, to the point that I find an LLM useful and they don't, or they don't know how to use LLMs for coding.

What do you work on, and what do LLMs do that helps? (Not disagreeing, but most of these comments -- on both sides -- are pretty vague.)

For once LLMs are good for building game prototypes. When all you care is to check whatever something is fun to play it really doesn'a matter how much of tech debt you generate in process.

And you start from the stratch all the time so you can generate all the documentation before you ever start to generate code. And when LLM slop become overwhelming you just drop it and go to check next idea.

Re: What makes Claude Code so damn good

#49

Post by exe34 » Sat, Aug 23, 2025, 8:19 PM UTC

diego_sandoval wrote: post #17

It shocks me when people say that LLMs don't make them more productive, because my experience has been the complete opposite, especially with Claude Code. Either I'm worse than then at programming, to the point that I find an LLM useful and they don't, or they don't know how to use LLMs for coding.

it makes me very productive with new prototypes in languages/frameworks that I'm not familiar with. conversely, a lot of my work involves coding as part of understanding the business problem in the first place. think making a plot to figure out how two things relate, and then based on the understanding trying out some other operation. it doesn't matter how fast the machine can write code, my slow meat brain is still the bottleneck. the coding is trivial.

Re: What makes Claude Code so damn good

#50

Post by Aurornis » Sat, Aug 23, 2025, 8:19 PM UTC

diego_sandoval wrote: post #17

It shocks me when people say that LLMs don't make them more productive, because my experience has been the complete opposite, especially with Claude Code. Either I'm worse than then at programming, to the point that I find an LLM useful and they don't, or they don't know how to use LLMs for coding.

I’ve found LLMs useful at some specific tasks, but a complete waste of time at others.

If I only ever wrote small Python scripts, did small to medium JavaScript front end or full stack websites, or a number of other generic tasks where LLMs are well trained I’d probably have a different opinion.

Drop into one of my non-generic Rust codebases that does something complex and I could spent hours trying to keep the LLM moving in the right direction and away from all of the dead ends and thought loops.

It really depends on what you’re using them for.

That said, there are a lot of commenters who haven’t spent more than a few hours playing with LLMs and see every LLM misstep as confirmation of their preconceived ideas that they’re entirely useless.