Another late-night Claude Code post

https://twitter.com/Steve_Yegge/status/1898993080931611112

by tosh • 1 year ago

42 26 1 year ago

evalstate2• 1 year ago

I've spent more on Claude Code than I'm willing to admit, and I'd estimate about 50% of the spend is "written off". With CC it can be tough to judge when to stop on a particular path.

Prior to CC I was using Goose, which is similar - and it's hard to tell how much better CC it than Goose as Sonnet 3.7 was released at the same time I switched. One of the nice features of Goose that CC doesn't have is loadable history/resumable sessions.

The workflow I've actually found most effective (and cheaper) now is to use the GitHub integration in the Claude.ai to get started, then use Claude Code to fill in the bits. The GH integration is much better than I expected, and worth a try if you have a Claude plan.

yodon• 1 year ago

Which GitHub integration are you referring to?

evalstate2• 1 year ago

The GitHub logo between the File Attach, Screenshot and Google Drive icons.

milesrout3• 1 year ago

I dont think HN needs to syndicate Steve Yegge's twitter feed

relaxing• 1 year ago

Yeah this is pretty content free as hype goes.

nullstyle3• 1 year ago

Ythis community needs more people to be accountable for the nonsense they post. Still want to talk about how there’s no territorial threat to canada?

milesrout3• 1 year ago

Yes. There is zero territorial threat to Canada.

theshrike79• 1 year ago

He should stop tweeting and get back to blogging. Or even cross-post to a platform not controlled by a megalomaniac billionaire.

inciampati2• 1 year ago

I'm impressed by how many people who are working with Claude Code seem to have never heard of its open source inspiration, aider: https://aider.chat/

It's exactly what Yegge describes. It runs in the terminal, offering a retro vision of the most futuristic thing you can possibly be doing today. But that's been true since the command line was born.

But it's more than Claude Code, in that it's backend LLM agnostic. Although sonnet 3.7 with thinking _is_ killing the competition, you're not limited to it, and switching to another model or API provider is trivial, and something you might do many times a day even in your pursuit of code that vibes.

I've been a vim and emacs person. But now I'm neither. I never open an editor except in total desperation. I'm an aider person now. Just as overwhelmingly addicted to it as Yegge is to Claude Code. May the future tooling we use be open source and ever more powerful.

edit: Another major benefit of aider is its deep integration with git. It's not "LLM plus coding" it's really like an interface between any LLM (or LLMs), git, and your computer. When things go wrong, I git branch, reset to a working commit, and continue exploring.

cadamsdotcom• 1 year ago

Do you happen to know how they compare in actual usage? Steve Yegge is gushing proudly, so it makes me wonder if, while of course aider and CC are of the same rough variety of tool, Claude Code is way better at something, or a little better at a few things, or something that's causing him to be so excited, and makes me wonder if aider is quite "up there" with Claude Code at doing those things.

I guess what I'm saying is, if Steve Yegge tried aider would he gush about it the same way?

ookblah• 1 year ago

maybe my settings were off but i tried aider when it first came out and kept coming back to it and it ended up being super slow and generating subpar stuff compared to just chat. actually turned me off to CLI based style code assistants.

claude code blew my mind, however. really great for refactoring existing stuff with a proper CLAUDE.md file or one-shotting small features. i mainly use it for scaffolding, sounding board and refactoring but it's been a game changer (albeit more expensive).

still like to use boring chat-based and pasting back and forth as it helps me approach and understand issues better for later on. i find generating huge blocks of code/files to feel like magic but then u dig into a bug later and find the LLM wrote or structured a bunch of semi-garbage that you have to undo.

skerit• 1 year ago

i tried aider when it first came out and kept coming back to it and it ended up being super slow and generating subpar stuff

I had very similar results. I discovered Aider & Claude-code at the same time. I even tried Aider with Claude 3.7 (with thinking), and it just wasn't as good. I think it's a lot more annoying to use and set up, a bit more conservative in how it tries to understand your project, where claude-code has no problems scanning your entire codebase.

still like to use boring chat-based and pasting back and forth as it helps me approach and understand issues better for later on. i find generating huge blocks of code/files to feel like magic but then u dig into a bug later and find the LLM wrote or structured a bunch of semi-garbage that you have to undo

Same. When an LLM generates code in a simple chat, it's always a bit more abstract, which is a good thing. Some of the things claude-code generated turned out to be garbage.

j45• 1 year ago

Aider seems to have fed or inspired a lot of the coding tools out there with some of Aider's breakthroughs.

leptons3• 1 year ago

Because Claude Code keeps doing stuff. It keeps solving massive problems, one after another. I throw larger and larger things at it, and it is unfazed. Chomp. Chomp. Chomp.

There's very little substance to this post. What "massive problems"? What "larger and larger things"? Am I supposed to guess? Sorry, I'm not impressed by vagaries.

I'm also personally not very impressed by AI, as it won't solve any problems I have because I have very specific coding problems that are based on undocumented APIs, doing things that literally nobody has done before. I've tried AI, but it's practically always wrong, and when it's correct it's something so trivial that I could have easily written the code. Maybe the problems I'm trying to solve are too "massive" for AI to guess at, or too obscure.

kledru2• 1 year ago

I have some 20-year-old codebases, so I do know what massive problems are. In such codebases AI assistants have finally made it possible to solve problems nobody had time to solve for decades and Claude Code seems to be best fit for that work so far.

_steve_yegge_• 1 year ago

Couldn't have put it better myself. It's great for the problems nobody has ever had time to solve. It's the problems that fall just below the ROI threshold, that tend to pile up until you have hundreds to thousands of them, each with icky weird dependencies that are _exactly_ the reason nobody has time to work on them.

Now you can chip through a dozen of these complex issues a day, and hope has finally arisen for getting through the backlog. That's a life-changing difference for anyone with a legacy code base.

leptons3• 1 year ago

Again, grand claims with zero substance. What are these supposed "massive problems"? I think your definition of "massive" and mine could be entirely different things.

eknkc• 1 year ago

I have been throwing it on to our codebase last couple of days.

I found it to be good in reviewing code. I ask it to check if we missed anything. If the diff is sound etc. Works really good. Caught a couple of issues and provided decent feedback.

I asked it to implement new things and.. I’m not sure. Depends on what you are working on I guess but it did “fine” overall. I’d need to provide feedback, ask it to consider some other stuff etc. Overall it was like steering an extremely intelligent junior developer. Not that revolutionary.

I don’t know how people actually utilise these though. Maybe I’m doing it wrong.

zurfer• 1 year ago

Claude code is surprisingly good. It's easily another 2x speed up over my previous AI enabled software development (copilot + chatgpt).

It's not a replacement for a SWE today, but it allows a SWE to get a lot more done.

I feel the current system is constrained in 2 dimensions: overall integration ( writing a feature and then using/testing it to make sure all the relevant cases are covered) and still limited context ( in a bigger code base there are utility functions and conventions that it doesn't always follow and so it introduces subtle bugs).

However, it's not like a new software engineer would get all of these things right as well.

The future of coding seems really to be a lot more review and test based vs. hand crafting all the details. That being said, you should keep your eye on what makes your app unique. Chances are high that it doesn't get these things right without enough guidance.

Another late-night Claude Code post

Related Stories