Any ideas why ripgrep is popular with GPT-5?

[lapplislazuli]

Hi,
bit of a random question.
I am a researcher working on automated software engineering, and that is at the moment a lot of LLMs and agentic systems.

One thing I noticed recently is that the same agentic system with GPT-5-mini will use ripgrep quite a lot, more so than other tools like find or grep, etc.
Has someone else mentioned this?
Is there maybe some known reason for this?

If there is no better guess, I would say either

• Ripgrep is somehow prominent in LLM training data as a repository / documentation
• There are other LLM training steps that use ripgrep

but I would be happy to hear opinions.

[BurntSushi]

I'm aware. But I haven't really used LLMs yet, so I probably don't have much insight.

With that said, I don't see that you've considered:

1. It's faster than grep in most cases. And find too.
2. It's cross platform.
3. It can print its results in a JSON format.

[lapplislazuli]

Thanks for the quick reply !
I see where you are coming from and why these are good 'functional' arguments that an LLM would choose ripgrep.

But then the question is how does the LLM choose it still - it cannot "think" or "know" and any arguments it may makes are repetitions of online arguments it saw 🤔

[BurntSushi]

That's a question for people who build LLMs. I don't think you'll get much insight on it here.