Hacker News Reader

Comment by resonious | Hacker News Reader

As a counter example (re: agents), I routinely delegate simple tasks to Claude Code and get near-perfect results. But I've also had experiences like yours where I ended up wasting more time than saved. I just kept trying with different types of tasks, and narrowed it down to the point where I have a good intuition for what works and what doesn't. The benefit is I can fire off a request on my phone, stick it in my pocket, then do a code review some time later. This process is very low mental overhead for me, so it's a big productivity win.

View on HN

Sounds like a slot machine. Insert api tokens, get something that's pretty close to right, insert more tokens and hope it works this time.

Except the tokens you insert have meaning, and some yield better results than others. Not like a slot machine at all, really. Last I checked, those only have 1 possible input, no way to improve your odds.

Ok so it's poker rather than a slot machine

Yes I accept this analogy!

Not really, it's not a zero-sum game. You're not competing against anything, you're working with something. It's just a tool that takes practice, has some variability and isn't free. Like most things in life. More like buying corn or having friends.

Poker takes practice, has variability and isn't free. In fact it's the only game I know of that's pointlessly boring without money on the table.

LLM workflow is competing with other ways of writing code. DIY, stack overflow, paired, offshored...

> pointlessly boring without money on the table.

I bought a bunch of poker chips and taught Texas Hold'em to my kids. We have a fantastic time playing with no money on the line, just winning or losing the game based on who wins all the chips.

Give them enough time and they'll realize they can trade poker chips for other things.

That’s fine if your expectations are consummate.

How's that different from a human developer? Give the same task to different developers and you'll get different levels of correctness and quality. Give the task to the same developer on different days and it is the same.

Its a lot faster to give a task to an ai agent than a developer. The agent is always at their desk, always listening, and will immediately prioritize whatever you tell it to do.

An ai agent always has capacity, does not have competing priorities, nor does it have ideas about what does or does not fall within their "scope of work".

Thats cool, how are you integrating your phone with your Claude workflow?

I don't know how to do it with Claude Code, but I was at a beach vacation for the past few days and I was studying french on my phone with an webapp that I made. Sometimes I'd notice something bug me, and I used cursor's "background agents" tool to ask it to make a change. This is essentially just a website where you can type in your request, and they allocate a VM, check out your repository, then run the cursor LLM agent inside that VM to implement your requested changes, then push it and create a pull request to your repo. Because I have CI/CD setup, I then just merged the change and waited for it to deploy (usually going for a swim in-between).

I realized as I was doing it that I wouldn't be able to tell anyone about it because I would sound like the most obnoxious AI bro ever. But it worked! (For the simple requests I used it on.) The most annoying part was that I had to tell it to run rustfmt every time, because otherwise it would fail CI and I wouldn't be able to merge it. And then it would take forever to install a rust toolchain and figure out how to run clippy and stuff. But it did feel crazy to be able to work on it from the beach. Anyway, I'm apparently not very good at taking vacations, lol

Is this Terragon's offering, or does Cursor also have web-based background agents?

Cursor has web-based background agents, they launched them recently

You can set up hooks: https://docs.anthropic.com/en/docs/claude-code/hooks-guide

And use something like ntfy to get notifications on your phone:

https://ntfy.sh/

I’ve also seen people assign Claude code issues on GitHub and then use the GitHub mobile app on their phone to get notifications and review PRs.

My dev environment works perfectly on Termux, and so does Claude Code. So I just run `claude` like normal, and everything is identical to how I do it on desktop.

Edit: clarity

Do you use it on a phone or on a tablet?

Phone. One of those foldy ones though so pretty big screen.

I just SSH into my CC machine from the phone, then use CC.

The cost is in the context switching. Throw 3 tasks that came 15, 20 and 30 min later. The first is mostly ok, you finish by hand. The second have some problems, ask for a rework. Then came the other and, while ok, is have some design problems. Ask another rework. Comes back the second one, and you have to remember the original task and what things you asked for change.

What are you using to fire up requests on your phone?

Reply on news.ycombinator.com