upvote
I don’t even trust myself to not mess up my git repo
reply
I'm not GP, but I have backups, plus I always make sure I've committed and pushed all code I care about to the remote. I do this even when running a prompt in an agent. That goes for running most things actually, not just CC. If claude code runs a git push -f then that could really hurt, but I have enough confidence from working with the agents that they aren't going to do that that it's worth it to me to take the risk in exchange for the convenience of using the agent.
reply
> how do you trust the code claude wrote?

If that's something you're worried about, review the code before running it.

> don't you get anxiety "what if there's an error in tui code and it would mess up my git repo"?

I think you might want to not run untrusted programs in an environment like that, alternatively find a way of start being able to trust the program. Either approaches work, and works best depending on what you're trying to do.

reply
> If that's something you're worried about, review the code before running it.

It takes more, not less, time to thoroughly review code you didn't write.

reply
Depends. If I was the one coming up with the implementation anyways, it's basically just the "coding" part that was replaced with "fingers hitting keyboard" and "agents writing to disk", so reviewing the code certainly is faster, you just have to "check" it, not understand it from scratch.

If we're talking receiving random patches where first you have to understand the context, background and so on, then yeah I agree, it'll take longer time probably than what it took for someone to hammer with their fingers. But again, I'm not sure that's how professionals use LLMs right now, vibe-coding is a small hyped world mostly non-programmers seem to engage in.

reply
> you just have to "check" it, not understand it from scratch.

How can you "check" that which you don't "understand"?

> I'm not sure that's how professionals use LLMs right now

I'm a professional and I can tell you how I use LLMs: I write code with their assistance, they don't write code for me.

The few times I let Claude or Copilot loose, the results were heartbreaking and I spent more time reviewing (and then discarding) the code than what it took me to later write it from scratch.

reply
> How can you "check" that which you don't "understand"?

??? I do understand, since I literally just instructed it, how would I otherwise? I'm not letting the LLM do the design, it's all me still. So the "understand" already exists before the LLM even finished working.

> I'm a professional and I can tell you how I use LLMs: I write code with their assistance, they don't write code for me.

Hey, welcome to the club, me too :) I don't write code, I write English prose, yet nothing is vibe coded, and probably I'll end up being able to spend more time than you thinking about the design and architecture and how it fits in, because actual typing is no longer slowing me down. Yet again, every line is reviewed multiple times.

It's more about the person behind the tools, than the tools themselves I think ultimately. Except for Copilot probably, the times I've tried it I've just not been able to produce code that is even slightly up to my standards. It's a breeze with Codex though (5.2), and kind of hit and miss with Claude Code.

reply
I push my branches daily, so I wouldn't lose that much work. If it breaks then I ask it to fix it.

But I do quickly check the output what it does, and especially the commands it runs. Sometimes it throws all code in a single file, so I ask for 'good architecture with abstractions'.

reply
I see this regularly: "I use GitHub to backup my local repos."

If `gh repo ...` commands get run you can lose everything instantly. You can force push and be left with a single blank commit on both sides. The agent has full control of everything, not just your local data.

Just set up Rclone/restic and get your stuff into a system with some immutability.

reply
Force pushing doesn't actually remove anything from the remote repository, only changes some references for which commits the branches point to. Plus, any forks on github will be completely unaffected. It's not perfect, since Github doesn't seem to offer any history of such reference alterations (a la the reflog), but it's still a valuable offsite backup from a developer's perspective.
reply
Okay, fair enough re force pushing (though `gh repo delete` is still an option). I suppose for a sufficiently active codebase copies of it will exist elsewhere. Just seems odd to me that people aren't backing up anything else on their computers otherwise they could trivially just include their git-based projects.
reply
Isn't it this case no matter who wrote the code? How do you ever run anything if you're worried about bugs?
reply
When I write the code myself, I'm not worried that I snuck a `git reset --hard` somewhere.
reply
Different type of creator, different type of bugs. I'd assume a human giving me a way to delete merged branches has probably had the same issue, solved the same problem and understands unspecified context around the problem (e.g protect local data). They probably run it themselves so bugs are most likely to occur in edge cases around none standard use as it works for them.

Ais are giving you what they get from common patterns, parsing documentation etc. Depending what you're asking this might be an entirely novel combination of commands never run before. And depending on the model/prompt it might solve in a way any human would balk at (push main to origin, delete .git, re-clone from origin. Merged local branches are gone!)

It's like the ai art issues - people struggle with relative proportions and tones and making it look real. Ai has no issues with tones, but will add extra fingers or arms etc that humans rarely struggle with. You have to look for different things, and Ai bugs are definitely more dangerous than (most) human bugs.

(Depends a little, it's pretty easy to tell if a human knows what they're talking about. There's for sure humans who could write super destructive code, but other elements usually make you suspicious and worried about the code before that)

reply
It makes a difference whether an AI or a human wrote it. AIs make more random, inconsistent errors or omissions that a human wouldn’t make. AIs also don’t dog-feed their code the way human developers of tools usually do, catching more errors or unfit/missing logic that way.
reply
I assume that whatever I type can be also flawed and take precautions like backups etc
reply
It's a git repo. What's sort of mess-ups are you worried about that you can't reflog your way out of (or ask claude code to fix)? It's certainly possible to lose uncommitted work, but once it's been committed, unless claude code goes and deletes .git entirely (which I've had codex do, so you'd better push it somewhere), you can't lose work.
reply