Hey folks,

It was another big week for Anthropic with many separate launches that look unrelated if you only read the headlines. Claude merged Chat and Cowork into one app. Claude Code got a Projects mode that keeps building after you close your laptop. And Anthropic published a number on how much of its own agent research Claude now does by itself.

I’ve built some great things with Claude but that’s not the point. The tools are all finally connecting with each other.

Let's dive in..

Rollout is staged. Pro and Max plans get it first over the following weeks.

I pointed Claude at the iPhone Duo launch page and asked for a slide deck. Inside about 5 minutes I had a fully designed deck, built in the ordinary Claude desktop app.

Every slide was editable in place. Text, graphics, layout, the kind of thing you'd normally open Canva/Powerpoint/Google Slides for. Export sits at the top of the screen: click it and pick PDF or PowerPoint, done.

Here's my honest verdict. Making a deck out of a prompt isn't new, plenty of tools do that already. I expected a decent deck. I didn't expect to get it without leaving the Chat window.

Before this week, that job meant a few stops. Claude Code for the technical build. Claude Design for a mockup. Google Docs for the write-up collab. Canva or PowerPoint for the deck. Copy here, paste there, keep 4 tabs open.

Claude just closed most of those gaps. One conversation now produces the design, document and the deck with edits in place, and hands you a designed output without switching tabs.

Then, a few days later, Claude Code Projects went into beta where you describe the work, a coordinator scopes it and hands it worker threads, each its own cloud session running on its own branch.

My bet and Anthropic hasn't said this: the same always-on pattern lands on the documents and decks before too long. Right now Claude Docs and Slides are faster because you're still there driving. The next obvious step is handing one off and finding it finished when you get back.

One more piece from the same week. Anthropic put out its own figures on how much of its research Claude now does.

In February, Claude led under 1% of Anthropic's AI research and development, meaning it could complete most of a task end to end from a prompt with a human only supervising. By August, that was 26%.

Above that, over 90% of research work now involves Claude doing large chunks of it under close direction. Around 30,000 agents were doing research and engineering work at Anthropic at any one time as of August.

One thing caught my eye in Anthropic's own post explaining why they're publishing it: "We want to illuminate the pace of progress for the public" and to minimize the gap between what frontier labs know and what the public knows.

I'm not going to pretend that number tells you anything special. But it's the same shape as everything else this week. Work that used to need someone sitting there now mostly doesn't, whether that someone is an Anthropic researcher or you building a deck.