Cowork is now Claude

Hey 👋

Welcome back! The big talk last week was AI safety. Dario Amodei, CEO of Anthropic, published an essay arguing that the industry should pace the development of frontier AI. Sam Altman, Elon Musk and Demis Hassabis backed the idea, but not everyone agreed.

We came home from Build Day last Monday optimistic about how AI can help people solve more problems, faster. The risks are real too. For us this is not a binary question of AI or no AI. It is about how we roll it out, at what pace, with which safeguards, and how we make sure the benefits reach as many people as possible.

Last week also brought AI agents to everyone's phone. Meta's personal agent Muse reached number one on Apple's US App Store ten days after launch, ahead of ChatGPT.

The biggest change for the people we train is also on its way. Claude Cowork and chat are becoming one Claude, so what Cowork could do will be available from any conversation. The change is rolling out over the coming weeks, and we do not have it ourselves yet. Jump to the AI News section further down for all three stories.

In this issue

• Build Day at Mesh, and what people built in two hours
• NDC Oslo, with a fireside chat and a talk
• Three tips you can try today, from our bootcamps and guides
• AI News, from the Claude merge to the Agents API
• Coming up, with our bootcamps, a Claude Conversation and NDC Porto

Build Day at Mesh

On Monday 14 September around 175 people filled Mesh Youngstorget for Build Day on Fable 5.1, which we hosted together with Anthropic. Two hours, three tracks called Delight, Breakthrough and Everyday, and one goal, to build something. Most people arrived without a plan, and many opened Fable 5.1 for the first time that evening. Two hours later there were working apps and games on the big screen. See what people built in Simon's post.

Daisy Hollman from the Claude Code team at Anthropic joined us on stage. She talked about her work on the harness behind Claude Code, how she builds with the models, and took questions from the room for the rest of the session.

Two things worked well for the builders, and both are worth trying on your next project.

Sketch before you build. Type /design in Claude Code and ask for three HTML sketches of what you want to make. Pick one, then build.

Ask for more. This was the point of our talk that evening, and it fits in one sentence.

How can I be more ambitious in this project?

Ask it once the plan is in place, and let the model suggest what the project could become.

NDC Oslo

The week did not stop at Mesh. The day after Build Day, Aleksander and Daisy sat down for a fireside chat on stage at Community Tuesday at NDC Oslo.

On Wednesday he gave his talk, 10 tips to level up your AI-assisted coding, to a packed room at Oslo Spektrum, and the audience voted it green. If you missed it, the developer bootcamp on 30 September goes deeper, in your own codebase. Read Aleksander's post about the week on LinkedIn.

The magic powers of HTML

Ask an AI agent for a document and you typically get a wall of text in the chat. Ask for an HTML page and you can get a page with controls, steps and figures. The difference between an answer you skim and a page you keep is often the last few words of the prompt.

Make this an HTML page. Put the assumptions on sliders and keep it to one screen.

Other words that work are make it a wizard, give me three directions side by side and one file I can send. Claude, ChatGPT, Copilot and Gemini can all build the page. The tool only decides where it lands.

The magic powers of HTML has the full list of words and phrases to try.

Give the AI agent a goal and a way to check it

An AI agent is an AI that is given a goal instead of a question. It gathers context, acts, checks its own result and goes round again. What differs between tools is how many rounds it runs without you.

So describe the finished result and how to check it, then let it work.

Write the client summary from the notes in this folder. Open with the decisions. Check every number against the notes, and ask me about anything you cannot place.

What is an AI agent? walks through the loop with an example from a folder of invoices.

From chat to agent is a scale

There is no clean divide between chat and agent any more. The difference is how many rounds it runs without you, and what it is allowed to touch while it runs.

The same tool can sit in several places on that scale. Claude answers in a chat, searches and reads files before it answers, works through a folder on its own, and runs a morning brief on a schedule. With Cowork becoming part of Claude, the scale is about how you set the task up, not which app you open.

Here are five things I asked you for this week. For each one, tell me whether you could have done more of it on your own with a folder, a connector or a schedule, and what you would have needed from me.

The spectrum in the agent guide lines up the five steps, from a chat that answers to several AI agents working together.

AI News

A lot happened since last time. Here are the updates worth a look.

Claude Cowork and chat are now one Claude. What Cowork and Claude Design could do is now available from any conversation, with your context, skills and connectors, and Claude Docs and Claude Slides are new. It rolls out to Pro and Max first over the coming weeks, with Team and Free to follow, so you may not see it yet. Existing chats, projects, connectors and skills stay where they were, so what we teach still applies.

Introducing Muse. Meta's personal AI agent, running in its own secure browser on the Muse Spark model. It reached number one on Apple's US App Store on 18 September, and a Mac app followed the same day. US only for now.

Introducing Grok Bot. xAI launched Grok Bot on 11 August as a team of always-on AI agents, each with its own cloud computer. Last week it got a voice mode on desktop and mobile, and it has had new updates almost every week since launch, including an Android app in early September.

The era of the personal agent. Axios on Muse, Grok Bot, Instinct and Siri, and on the data each of them wants in return.

We Must Pace the Frontier. Dario Amodei, CEO of Anthropic, argues for slowing frontier AI development, with independent evaluators working inside the labs and a ladder of steps up to a full pause.

Bringing Salesforce into Claude. A plugin in beta on all paid Claude plans, with 37 skills for account executives, working within the Salesforce permissions you already have. Skills packaged for one role, the same pattern we build with teams.

The Agents API. OpenAI opened cloud agents built on the Codex harness as a public beta for all developers, with no fee beyond tokens and tools.

A Data agent in ChatGPT Work. It connects to Snowflake, BigQuery, Databricks, Redshift and more, plus files in Google Drive and SharePoint. Administrators decide whether it is on.

Novo Nordisk and Anthropic. Novo will test Claude Science on research workflows in drug discovery.

Gemini 3.8 Live. Google's new real-time voice model, in the Gemini API and AI Studio, in 97 languages and more.

Cursor Projects. A coordinator agent that hands work to subagents running on cloud computers, rolling out in beta.

Anthropic's September threat report. Case studies of misuse the team disrupted between December 2025 and August 2026, across seven areas of harm.

Coming up

Claude Code Bootcamp for developers, Wednesday 30 September at Mesh. A full day in your own codebase, building the repo you take home with you.

Claude Conversation, Wednesday 30 September at Mesh, 16:00 to 19:00. A small, hosted conversation on the big open questions in AI, where everyone in the room takes part. What comes up becomes the brief for an Impact Lab in October. Registration needs approval.

Claude Code Bootcamp for everyone, Tuesday 27 October at Mesh. A new date, since our last issue said 21 October. A full day with your own documents and research. No coding experience required.

NDC Porto, 17 to 20 November. Aleksander is speaking, and we are running a workshop there. Another follows at NDC AI in December.

We also run company bootcamps built around your team's tools, workflows and priorities. Get in touch if you want to bring us in this autumn.

If you try one of the tips, reply and tell us how it went. We read every answer.

Best, Aleksander and Simon