Sessions That Resume Themselves

I'm Shannon, and this is the Claude Notes Brief -- your weekly rundown of Claude Code updates and Anthropic news for the week of Monday, August twenty-fourth. Sessions now resume on their own when your usage limit resets. A new Concise output style is available. And the New York Times reports on Anthropic's potential public listing.

Let's start with Claude Code, where the change most people will notice is automatic session continuation -- if you hit a usage limit partway through a task, Claude Code now picks the session back up on its own once that limit resets. Before this, the turn just sat there stranded until you came back to it. Now a long-running task can carry on without you babysitting the clock. It's on by default, and you can switch it off in your configuration under the option to continue automatically at the usage limit.

The usage limit message itself got more useful too -- it now tells you when your session and weekly limits reset, not just the monthly one. Alongside that, there's a new built-in output style called Concise. It makes Claude lead with the result and skip the preamble and the running commentary, while doing exactly the same work underneath. If you've ever wanted less narration in your terminal, that's the one to pick, and you'll find it under output style in your configuration.

There's also a new environment variable that sets the model new sessions start on. The distinction matters -- the older variable locked you to a model, while this one just sets a starting point. Pick a different model from the interface and that choice still wins, and it persists across restarts. That theme of staying in control without interrupting yourself shows up in the in-session commands too.

You can now open the permissions panel and add a directory while Claude is still working, and permission rule changes take effect for the rest of the current turn rather than waiting for the next one. In the fullscreen interface, the auto-compact, theme, help, configuration and advisor panels also open mid-turn, and those panels no longer cover up your most recent messages. Two smaller additions round it out. There's a readline keybinding flavor that makes word deletion and word-wise cursor movement behave the way they do in Bash, if that's the muscle memory you have.

The default behavior is unchanged. And the Visual Studio Code extension now announces replies, permission requests, errors, and status changes to screen readers, with per-turn heading navigation through the transcript.

On the fixes side, this was a heavy week, and memory is the headline. Long interactive sessions no longer grow without bound -- subagent tool results are released once they scroll out of the display window, and expired web fetch content is now actually dropped the way it was supposed to be. Related to that, the built-in Claude API skill costs roughly twenty-five thousand tokens to load, down from over two hundred thousand. That's context you get back for the actual work.

The same cleanup effort reached billing and caching. Prompt caching works again when you're running behind a gateway or a custom base URL, and there was a Bedrock streaming bug that silently doubled your billed API calls behind certain proxies -- that's fixed. Startup is also faster on macOS, because the automatic update check is now deferred about ten seconds so it isn't competing for processor time right when you launch. Beyond that, there were fixes across remote control, fullscreen rendering, and cross-session messaging.

Turning to the broader Anthropic news. The New York Times reports that Anthropic could be preparing a public listing aiming to raise as much as one hundred billion dollars, which would put it among the largest offerings on record. That matters here because the company's funding structure shapes how aggressively it can invest in products like Claude Code. On a related note, the Wall Street Journal has an exclusive comparing quarterly sales figures, and the reported numbers put Anthropic ahead of OpenAI on enterprise growth this quarter, driven largely by coding and agent workloads.

Useful context if you're making the case for Claude Code inside your own organization. Elsewhere, The Verge has two pieces worth your time. One explains how Anthropic's invisible text watermarking will work, built on SynthID -- relevant if you ship generated text or documentation into downstream systems. The other covers expanded Gmail and Google Drive connectors, plus broader availability for Claude Cowork.

And on the Claude blog, there's a guide for startups covering five working practices small teams use with Claude Code, and a piece from an Anthropic CI engineer on the agent that triages their build and test failures internally. All linked in the show notes. That's it for the brief. I'm Shannon, and we'll see you next week.