Subagent Forking On by Default, Plus Anthropic's $11.5 Billion Quarter

I'm Shannon, and this is the Claude Notes Brief -- your weekly rundown of Claude Code updates and Anthropic news for the week of August seventeenth. Subagent forking is now on by default. You can mention other live sessions by name. And Anthropic reported quarterly revenue over eleven and a half billion dollars.

Let's start with Claude Code, where subagent forking is now on by default, and it changes how agents pick up where you left off. When you fork a subagent now, it inherits your full conversation and your prompt cache, so instead of starting cold, it begins with everything your session already knows. That saves setup and keeps the forked agent aligned with what you're actually working on. Related to that, non-teammate agent spawns in interactive sessions now run in the background by default, so your main session stays free while they work rather than blocking on them.

That theme of connecting sessions carries into the next change -- you can now type the at symbol in your prompt to mention another live Claude session by name, and Claude reaches out to it directly. To make that reliable, sessions on the same machine now keep unique names, so if two names collide, one gets a distinct variant. Next, there's a change worth flagging if you lean on task tracking. The built-in todo and task tracking tools are no longer available on the newer models -- Opus four point eight, Sonnet five, Fable five, Mythos five, and anything newer.

If you rely on them, you can restore them with an environment variable, but by default they're gone on those models. On the version control side, GitLab support is now broader. The worktree flag and the agents view both handle GitLab merge request URLs, showing them with an exclamation mark and the number, and bare GitLab repository URLs, including nested subgroups, now clone as plugin marketplaces the same way GitHub URLs already did. And one more that pairs with the model changes -- newer models can now overwrite an existing file with the Write tool without having read it first this session, which matches how the Edit tool already behaves.

Older models still require the read.

On to what changed under the hood, and this week leaned heavily on security. Several permission and sandbox bypasses were closed. That includes a Windows device-prefix path that leaked NTLM credentials, a PowerShell parameter override, Git Bash following Cygwin symlinks, and a Linux sandbox escape that got around a protected path. Alongside those, nested git repositories now require their own trust confirmation, so opening one doesn't inherit trust from the parent.

There's also a new memory limit on Linux -- you can cap how much memory a Bash command uses, so a runaway build can't stall your session. The other focus was responsiveness. Long responses no longer partly disappear or print twice while streaming, fullscreen sessions stay smooth because the conversation isn't re-normalized on every update, and those multi-second stalls after editing files with heavy IDE diagnostics are resolved.

Now to the broader Anthropic news. First, according to TechCrunch, Anthropic is turning Claude Code's auto mode on by default. That's a change you'll notice right away, since it affects how your daily sessions handle approvals and background work. On the business side, Bloomberg is reporting that Anthropic's second-quarter revenue topped eleven and a half billion dollars, ahead of a planned public offering -- a signal of continued investment in the products behind Claude Code.

That success is drawing competition. Bloomberg also reports that DeepSeek is publicly targeting Claude Code's lead in coding, part of a wider push by Chinese labs into agentic coding. And connecting back to this week's forking and multi-session changes, TechCrunch has a piece on what happened when Anthropic set multiple agents loose on the same task -- they started what the outlet calls a turf war. It's useful context as multi-agent workflows become the default.

We'll link all of these in the show notes. That's it for the brief. I'm Shannon, and we'll see you next week.