Anthropic says an internal Claude Tag creates 65% of its product team's code, and the week in Claude

PLUS: the US clears Mythos 5 for roughly 100 companies and agencies, per CNBC, a new Economic Index tracks Claude by the hour, and API rate limits go up

By SamJun 28, 2026

Hi, friends. Here's your week in Claude, from Jun 21 to Jun 27.

This week, Claude Tag brought a team-wide Claude to Slack, in beta. The US cleared Mythos 5 for a small group of cyber defenders. And API rate limits went up.

Plus 10 quick hits. Every claim has a link.

Claude Tag puts Claude in your team's Slack channels

Teams can now tag @Claude in Slack and hand it work.

Anthropic launched Claude Tag on Jun 23. It's in beta for Claude Enterprise and Team customers.

Admins pick its channels. They also pick the tools, data and codebases it can use.

Each channel has one Claude that works with everyone. It remembers relevant information from the channels it's in.

It can schedule its own tasks. It can work on a project over hours or days.

If "ambient" behavior is turned on, Claude flags things it thinks you might need to know. It also follows up on threads or tasks that went quiet without being resolved.

Anthropic says 65% of its product team's code is created by its internal version of Claude Tag.

Admins can set token spend limits for the whole org or per channel. They can see a log of what Claude did, and who asked. Claude's memories stay scoped to the channels admins define.

Claude Tag replaces the Claude in Slack app. Admins can opt in within 30 days to migrate. It works with Opus 4.8.

The US clears Mythos 5 for a small group of defenders

The US government will let Anthropic bring back Mythos 5 for some users. Fable 5 stays off for now.

On Jun 26, Commerce Secretary Howard Lutnick sent Anthropic a letter. It lets Anthropic release Mythos 5 to roughly 100 companies and federal agencies, CNBC reports. CNBC says it viewed the letter.

"I have determined that appropriate safeguards are in place to permit certain trusted partners to access the Claude Mythos 5 Model," Lutnick wrote, per CNBC. The letter didn't approve bringing back Fable 5.

Anthropic gave CNN a statement. It said Mythos 5 "can be redeployed to a small group of cyber defenders and infrastructure providers." It said it's working to restore their access "as quickly as possible."

On Jun 12, a US export control directive barred foreign nationals from Fable 5 and Mythos 5. That included Anthropic's own foreign national staff.

Anthropic said that meant turning both off for all its customers. It called the order a misunderstanding.

Anthropic's data shows Claude use gets more personal on weekends

Anthropic's new Economic Index report tracks Claude use down to the hour. It also adds a survey of users.

The report came out Jun 26. Per Anthropic, the share of chat and Cowork conversations that are personal rises from around 35% during the workweek to just under 50% on weekends. That's in its sample period.

People most often ask for sleep advice around 5 a.m. Recipe requests are 2.3 times more frequent than average at 6 p.m.

The survey links about 9,700 people's answers to their Claude use. Over a third expect AI to be able to do most or nearly all of their work tasks next year. 10% rated losing their own job in the next 12 months as likely or very likely.

Anthropic says the survey is not representative of the general population. It reaches a random sample of Claude users.

There may be selection in who completes the survey. And it filters out infrequent users.

Anthropic raises Claude API rate limits and cuts tiers to three

Sonnet and Haiku now get the same API rate limits as Opus.

On Jun 26, Anthropic raised rate limits across the Claude API, per the release notes. Sonnet and Haiku limits now match Opus at every usage tier.

Usage tiers went down to three: Start, Build and Scale. Most organizations move to a higher tier. No organization gets lower limits than before, and no action is required.

You can see your tier and current limits in the Claude Console.

Claude Desktop gets chat on AWS, Google Cloud and Foundry

Some companies get Claude Desktop through AWS, Google Cloud or Microsoft Foundry. They now get the full app.

Until Jun 22, those customers only had Claude Cowork and Claude Code, per Anthropic. Now one app covers chat, Cowork and Claude Code.

Inference runs on your cloud, in the regions you configure. Conversation history is stored locally.

Chat, Cowork and Claude Code each have their own policy key. So admins decide who gets what.

An offline installer covers air-gapped setups. A Microsoft 365 connector can run entirely on the device.

Claude Code now replies when you run a shell command

Claude Code now answers shell command output. MCP servers can sign in from your shell.

It shipped versions 2.1.186 to 2.1.195 from Jun 22 to 26, per its changelog. Its weekly digest covers the week.

Shell mode commands now get a reply from Claude once the output lands. It costs the same as a normal prompt. A setting turns it off.
claude mcp login and claude mcp logout sign MCP servers in and out from your shell. You don't need the /mcp menu.
/rewind can now resume a conversation from before /clear was run.
A new sandbox setting blocks sandboxed commands from reading credential files and secret environment variables.
Streaming responses use about 37% less CPU.

Quick hits

Fast mode for Opus 4.7 is deprecated (Jun 25). After it's removed on Jul 24, fast requests to Opus 4.7 will return an error. Anthropic points users to fast mode for Opus 4.8.
Team and Enterprise admins can now require Trusted Devices for Remote Control (Jun 25). Members must verify their device before viewing or steering local Claude Code sessions remotely.
Multiple Claude models had elevated errors on Jun 23. It ran from 14:08 to 15:33 UTC (9:08 to 10:33 am Central). Per Anthropic's status page, it affected claude.ai, the Claude Console, the Claude API, Claude Code and Cowork.
Claude is now an agent provider in GitHub Copilot for JetBrains IDEs through the Claude Code CLI (public preview, Jun 22). Per GitHub, it runs in bypass permissions mode for now, so all file edits and tool calls are approved automatically. On Copilot Business and Enterprise, an admin must first turn on the Editor preview features policy.
Claude Code 2.1.193 adds an OpenTelemetry log event with the model's response text (Jun 25). It's redacted unless turned on. Deployments that already log prompt content start receiving response content on upgrade.
Background subagents in Claude Code now ask for permission in the main session instead of auto-denying (2.1.186, Jun 22). The dialog shows which agent is asking.
Org-set model restrictions now apply to Claude Code's model picker, /model, --model and ANTHROPIC_MODEL (2.1.187, Jun 23). A restricted model shows a "restricted by your organization's settings" message.
/review in Claude Code now uses the same review engine as /code-review medium (2.1.186, Jun 22).
Claude Code hook matchers with hyphens now match exactly (2.1.195, Jun 26). Before, names like code-reviewer could accidentally match by substring.
The MCP tunnels management API moved from the Admin API to the Claude API (research preview, Jun 22). The old surface stays available during a migration window.

Worth your time

Agent identity in Claude Tag (Jun 24): Claude gets its own account in each system it touches. A channel member without repo access can still ask Claude to read that repo. That works if the channel's profile grants Claude that permission.
Building effective human-agent teams (Jun 24): Anthropic has spent months testing what makes teams of people and agents work. It shares what it learned. One lesson: "For an agent, if it's not written down and accessible, it doesn't exist."

That's the week. The next issue lands Sunday, Jul 5.

Reply and tell me what you're building. I read every reply.

— Sam