dcferreira

I have a $100 claude subscription and a deepseek account though I haven't used it much. Just today I published a blog post about my setup [0], but the TLDR is vanilla claude code (and sometimes oh-my-pi) with an Obsidian vault.

This week I started using Herdr also, and while I don't yet have it setup as I'd want, it's slightly better than me previous pure tmux setup to manage multiple agents. I also always turn on remote control on claude so I can control it via the app.

I suspect the gas towns and similar are no longer a necessity: I think as the models get better, the organization structures need less attention. I've been doing fine with, for each project, spinning up just one manager to manage all the relevant sessions and ping me when I'm needed. Or if I want to be hands off I get a pm and a tech lead claude sessions up, and the pm moves the work forward while the tech lead makes decisions on what to work on next and how to tackle it. This is in spirit a similar thing to gas town, but incredibly simplified. I don't think this would work with claude 3.5, but with 5.5 I'm quite happy!

[0] https://dcferreira.com/post/2026-10-how-i-use-agents/

I'm curious where you got those numbers from. I did a quick search and find wildly different numbers (depending on method and source, from ~60% to >98%).

However I don't find anywhere claiming anywhere near <5%. Can you back that up?

Example source of manufacturers claiming >95% [0].

0: https://climate.mit.edu/ask-mit/how-well-can-electric-vehicl...

Cool post, I've also been using more and more type hints and it really makes writing python more enjoyable. Although unfortunately it's still quite common for libraries to not have type stubs.

On the section of construction functions, I've always used/seen `@classmethod` instead of `@staticmethod`. I just had a quick look at some big python libraries (pandas, transformers), and they do use class methods. Is there any particular reason you'd want to use static methods instead?

> This way you can write functions that e.g. take only a URL as argument, load the content from that site, and then summarize it with the LLM.

I'm currently doing a project where this would be very helpful, but I can't think of what I'd need to send the LLM. In my case I'm scraping headlines from many news websites. I'm doing it manually with xpath currently.

What would be the way to use LLMs here? Just sending the HTML wouldn't work, as it's too many tokens. Probably I could send all <a> tags, but then how could I be sure the LLM doesn't choose too many/few?