Here’s how to easily set up a working playground for Anthropic’s new computer use API. Takes <10min. It gives Claude 3.5 Sonnet a tool that can use a computer just like a human which opens up endless possibilities - it’s wild. Absolute gamechanger for AI.

Here’s that link for the Anthropic Quickstart repo: This is some of the most fun you’ll have had in your time with AI. Fascinating to watch this thing work. Next few months are going to be insane - buckle up!

Whoever builds the best sandboxes for these things to run at scale is going to be extremely wealthy lol

People need to really start to think about the implications of what happens when: - You combine computer use + realtime - Add in a GPT-5/Claude-4 class model - Cost goes down by 10x - Speed goes up by 10x - Infrastructure gets to where it’s easy to spin entire teams of these up

Tangent on the updated 3.5 Sonnet… So far noticeably *worse*. Have to think it’s been optimized to be smaller or something, idk. Reverted back to the old one within Cursor and it was quite noticeably. I immediately got back to getting great results.

Add voice to this and it’s over

Oh boy do I have something for you

It will be slower at first, kind of like training a new group of employees. Nobody will be as fast as an experienced person... but at some point, you can build an AI team of 100+ agents, EACH doing a task 70% as well as you.

With Microsoft and Anthropic both coming out with their versions of Agents over the last couple of days - unless OpenAI does something similar soon (maybe the 2 year anniversary tease?) then they're falling quite far behind in the AGI race it seems.

Oh they’ve got plenty it’s just not public

Here's a script made specifically for MacOS with minimal setup