Hacker News Reader

Comment by mgdev | Hacker News Reader

This thing is cool except:

1) It chews through tokens. If you're on a metered API plan I would avoid it. I've spent $300+ on this just in the last 2 days, doing what I perceived to be fairly basic tasks.

2) It's terrifying. No directory sandboxing, etc. On one hand, it's cool that this thing can modify anything on my machine that I can. On the other, it's terrifying that it can modify anything on my machine that I can.

That said, some really nice things that make this "click":

1) Dynamic skill creation is awesome.

2) Having the ability to schedule recurring and one-time tasks makes it terribly convenient.

3) Persistent agents with remote messaging makes it really feel like an assistant.

View on HN

> It chews through tokens. If you're on a metered API plan I would avoid it. I've spent $300+ on this just in the last 2 days, doing what I perceived to be fairly basic tasks.

Didn’t Anthropic make it so you can’t use your Claude Code Pro/Max with other tools? Has anyone experienced a block because of that policy while using this tool?

Also really curious what kind of tasks ran up $300 in 2 days? Definitely believe it’s possible. Just curious.

I offhandedly set it up to do a weather alert every 4 hours during the big winter storm. Absent a well-specified API, I can only assume it was repeatedly doing a bunch of work to access some open API it discovered.

Very much the LLM equivalent of “to bake an apple pie you must first invent the universe”.

To its credit, it did a great job.

Wow, so it must have been spending a ton of reasoning tokens then writing code to go fetch the weather. Or maybe using a browser?

Hopefully one day we get self-hostable LLMs good enough for this.

Seen a couple of people on X have posted about their Claude accounts being suspended after using this. All of them seem to have used it with Claude Code so yes looks like it violates their policy (not surprising really, it breaks their TOS).

I've tried it on Codex (ChatGPT Pro) and within an hour of just getting stuff set up and tested used half my weekly limit so I can see using $300 in a couple of days being very easy.

Until thats figured out this is basically a non starter, you can't use it if its going to cost $1k+ per week to use, and I'm not sure theres any local models that'd handle it without $10k+ in hardware costs.

I’ve been working on adapting Claude Code to do some repetitive “personal assistant” type tasks so I was really excited to try this tool.

One of my tasks is a skill that fetches my calendar via MCP and slots events into a JSON to be used for an OR-Tools constraint optimizer that finds a workable schedule for something. It then uploads those events to the calendar using MCP when I choose my favorite candidate solution.

I checked token usage for this task last time I ran it. It would’ve cost $29 in API usage with Opus 4.5.

So yea, you’re absolutely right that this stuff isn’t going to go mainstream at these rates.

One thing you can try is powering Clawdbot with a local model. My company recently wrote[0] about it.

Unclear what kind of quality you'll get out of it, but since the tokens are all local, kinda doesn't matter if it burns through 10x more for the same outcome.

[0]:https://www.docker.com/blog/clawdbot-docker-model-runner-pri...

Reply on news.ycombinator.com