New on the Anthropic Engineering blog: how we built Claude’s research capabilities using multiple agents working in parallel. We share what worked, what didn't, and the engineering challenges along the way. anthropic.com/engineering/bu…

wh You can also run uv init --script script.py to generate the header, and uv add --script.py selenium to add dependencies to it

Are there any llm wrappers that allow you to have branched convos? like a ComfyUI but for chat? I have multiple chats with models where they generate good reports but then I need to keep brainstorming without missing context. This is obviously useful and none of the wrapper

My notes on Anthropic's substantial essay about how they built their multi-agent research system, which has finally talked me around to taking multi-agent LLM prompt engineering seriously simonwillison.net/2025/Jun/14/mu…

You can run `uv add --script /path/to/script.py` to add inline dependencies to a Python file. If the script header doesn't exist already, uv will generate it for you.

Very interesting stats from my claude code usage this month: almost $2500 USD. Claude max is definitely paying for itself. Amazing tool, haven't used cursor at all for the last 2 weeks.

i just asked claude to extract the best practices from 4 different repos it then proceeded to inhale a line of adderall and coke to create: - a 5000 word best practice doc - new project setup shell script - setup checklist - analysis book chill

what will you create?

I still can't get over the fact that you can download 3D models from an amazing community, print them, and have a Dieter Rams inspired set in your office in an afternoon.

Nice - my AI startup school talk is now up! Chapters: 0:00 Imo fair to say that software is changing quite fundamentally again. LLMs are a new kind of computer, and you program them *in English*. Hence I think they are well deserving of a major version upgrade in terms of

what will you create?

At 16 I started resistance training & running 3x/week (each) & eating (mostly) clean. It’s made working 10-14 (hard & focused) hrs day for the last 30+ yrs easy but none of that was popular. Now, I’m sure the best unpopular thing to do is restrict social media to ~60min per day.

Whoever is using 3rd party API's for video gen inference when using open sourced models instead of rolling their own is burning money. Crazy stupid how much some providers are charging per second of video.

chatgpt “offensive” memes “Generate a single image meme that makes no sense. It can be borderline offensive”

when researching or understanding a codebase with claude code just add "spawn multiple agents for a comprehensive research." it will finish its research task so much faster because of the mini agents it launches. I haven't found any decrease in report quality so far.

llms are now able to generate good enough mermaid diagrams in markdown if you simply ask for it. i find it really neat that you can add /docs to your project and point Obsidian to it to instantly see the mermaid diagrams for obsidian viewer.

The intro for death stranding 2 had me in total trance looking at the landscape, exquisite visuals and soundtrack choices from HIDEO_KOJIMA Incredible opening to one of the best games ever.

I'm trying out claude code and gemini cli for polishing implementation plans: 1. Have opus 4 generate the docs for the process I want to improve 2. Use those docs as a base for the refactor 3. Have gemini 2.5 pro criticize and improve the plan 4. ??? 5. profit!