Agent View Lands, Anthropic Hits 900 Billion

I'm Shannon, and this is the Claude Notes Brief -- Claude Code updates and Anthropic news for the week of May eighteenth. A new agent dashboard ships as a research preview. Fast mode now defaults to a newer model. And Anthropic closes a funding round at a nine hundred billion dollar valuation.

Let's start with Claude Code -- the headline this week is a research preview called agent view, which gives you a single dashboard for every session you have running. The idea is simple. If you're juggling work across multiple repos, you can now pull up one list that shows which sessions are running, which are blocked waiting on your input, and which have finished. The same release adds a long set of configuration flags for sessions you dispatch from that dashboard, covering things like which model to use, which directories to include, which MCP servers to load, and which permission mode to run in.

Anthropic has a full walkthrough on the engineering blog, and we'll link it in the show notes. Tied to that multi-session theme, background sessions now properly preserve their configuration across retire and wake cycles. Previously a backgrounded worker could quietly drop its MCP servers or revert to default permissions after going idle. Now your model, your effort level, your permission mode, and your custom directories all stick.

There's also a new goal command worth knowing about. You give Claude a completion condition, and it'll keep working across turns until that condition is satisfied, with a live overlay showing elapsed time, turn count, and token usage. It works in interactive mode, in headless runs, and over Remote Control. Quieter but useful -- the rewind menu picked up a summarize up to here action, which compresses earlier conversation history while leaving your most recent turns verbatim.

Handy when a long session is bumping the context limit but you still want the last few exchanges intact. And on the plugins side, the system now understands dependencies. Disabling a plugin that another enabled plugin depends on will refuse, with a copy-pasteable hint showing the full disable chain. The marketplace browser and the plugin details view also show projected token cost per turn and per invocation, so you can see what a plugin will do to your context budget before you install it.

On the under-the-hood side, the theme this week is reliability for things that talk to the outside world. MCP timeouts got a real fix -- the tool timeout setting now actually raises the per-request fetch timeout for remote HTTP and SSE servers, which was previously capped at sixty seconds no matter what you configured. Servers returning a four oh three now correctly show as needing auth instead of just failing, and SSE responses are capped at sixteen megabytes per frame to stop misbehaving servers from running away with memory. There's also a worktree safety fix that's worth flagging.

When git worktree remove failed, cleanup used to fall back to a recursive delete, which could wipe gitignored or in-progress files. That fallback is gone. The background daemon also handles macOS sleep and wake correctly now, so clock jumps no longer look like elapsed idle time. And one more for Windows users on Bedrock, Vertex, and Foundry -- the PowerShell tool is now on by default, with execution policy bypassed so signed-script restrictions don't block tool calls.

There are environment variables to opt out if you need the old behavior.

On to the broader news, and the dominant story is funding. The Financial Times is reporting that Anthropic has agreed terms on a thirty billion dollar funding round at a nine hundred billion dollar valuation. That capital is what backs the compute commitments behind Claude's expanding usage limits. The Wall Street Journal has a companion piece framing how Anthropic went from challenger to front-runner in the AI boom, with Claude Code and enterprise adoption central to the story.

Worth a listen if you want context for the infrastructure and pricing moves of the last few months. On pricing specifically, The Information is reporting that Anthropic is raising prices and tightening subscription limits while customers keep paying. That tracks with the tighter caps subscribers saw land this week. And one more enterprise story -- The Verge reports that Microsoft is canceling internal Claude Code licenses for its developers as it pushes its own tooling.

Mostly a procurement story, but a useful signal if you work in a Microsoft-heavy org. We'll link all four pieces in the show notes. That's it for the brief. I'm Shannon, and we'll see you next week.