11–20 of 74 posts
If AI agents are so underdeveloped and useless that they can’t parse out CLI flags, then the answer is not to rewrite the CLI.
You either give the agents an API layer or you don’t use them because they’re not mature enough for the problem space.
> Humans rarely typo a traversal. I don't think this is true?
You want me to hand type a file name? I’ll flip a letter or skip one!
1: Strictly speaking, there are ways to access some GUI programs on Linux with a screen reader. However, frankly, most are not really a joy to use. The speed of interaction I get from a TUI is simply unmatched. Whenever I work with a true GUI, no matter if Windows, Mac or Linux, it feels like I am trying to run away from a monster in a dream. I try to run, but all I manage to do is wobble about...
Are we reinventing RPC again? Calling CLI program with JSON sounds like RPC call. The schema feels likes something LSP can provided for such function. Maybe asking agent to write/execute code that wraps CLI is a better solution.
Everything old is new again...
The CLIs I’ve seen agents struggle with are those that wrap an enormous, unwieldy, poorly designed API under one namespace. All of Google Workspace apis, for example.
The pattern I used was this:
1) made a docs command that printed out the path of the available docs
$ my-cli docs
- README.md
- DOC1.md
- dir2/DOC2.md
2) added a --path flag to print out a specific doc (tried to keep each doc less than 400 lines).
$ my-cli docs --path dir2/DOC2.md
# Contents of DOC2.md
3) added embeddings so I could do semantic search
$ my-cli search "how do I install x?"
[1] DOC1.md
"You can install x by ..."
[2] dir2/DOC2.md
"after you install..."
You then just need a simple skill to tell the agent about the docs and search command.
I actually love this as a pattern, it works really well. I got it to work with i18n too.
Personally, I'm skeptical:
- Having the agent look up the JSON schemas and skills to use the CLI still dumps a lot of tokens into its context.
- Designing for AI agents over humans doesn't seem very future proof. Much of the world is still designed for humans, so the developers of agents are incentivized to make agents increasingly tolerate human design.
- This design is novel and may be fairly unfamiliar in the LLM's training data, so I'd imagine the agent would spend more tokens figuring this CLI out compared to a more traditional, human-centered CLI.
> Humans rarely typo a traversal. I don't think this is true?
The typos are a bit different, but that’s one reason I hate the command line as a human. You want me to hand type a file name? I’ll flip a letter or skip one!
> The real question: what does it actually look like to build for this? What was the not-so-real question? Or the surreal question? I know it's becoming tiresome complaining of slop in HN. But folks! Put a bit of care in your writing! It is starting to look as if people had one more agent skill "write blogpost", with predictable results, as we are not a Python interpreter putting up with meh-to-disgusting code but ac…
This feels completely speculative: there's no measure of whether this approach is actually effective. Personally, I'm skeptical: - Having the agent look up the JSON schemas and skills to use the CLI still dumps a lot of tokens into its context. - Designing for AI agents over humans doesn't seem very future proof. Much of the world is still designed for humans, so the developers of agents are incentivized to make agen…
A cli that is well designed for humans is well designed for agents too. The only difference is that you shouldn't dump pages of content that can pollute context needlessly. But then again, you probably shouldn't be dumping pages of content for humans either.