Claude Tag, the Agentic Workspace, and AI Policy Shifts

Anthropic decided your Slack wasn't noisy enough and dropped an AI colleague into it. Elsewhere, the company published a policy paper that reads like a lobbying document in a lab coat, launched a fellowship that's basically Peace Corps for prompt engineers, and extended its cyber-defense toolkit to more of the planet.

Claude Tag: The AI That Won't Shut Up in Your Slack

On June 23, Anthropic shipped Claude Tag — the ability to @Claude in any Slack channel and have it respond as if it's a full member of your team rather than a glorified search box. This goes well beyond "ask a question, get an answer." Claude now accumulates channel context over time, building a working model of your projects, jargon, and ongoing debates. It references past decisions, follows up on threads, and — in what Anthropic calls "ambient" behavior — will proactively surface information or flag issues without anyone asking. Your new colleague never sleeps, never takes PTO, and has read every message you've ever sent. Comforting.

The system operates asynchronously, meaning it can chew on complex requests in the background and deliver results hours later — much like a human teammate in a different timezone, minus the passive-aggressive "per my last message." Anthropic claims 65% of its own product code now originates from this workflow: engineers describe problems in natural language in Slack, Claude produces working implementations, humans review and merge. Whether this makes Anthropic's engineers impressively efficient or slightly redundant is left as an exercise for the reader.

Why it matters: This is a legitimate UX shift. Instead of alt-tabbing to yet another AI tool, Claude meets you where you already waste most of your day. If the ambient behavior actually works — big "if" — it collapses the line between "using AI" and "working with a team." Available now for Enterprise and Team plans, because of course it is.

Anthropic's Policy Paper: "Everything Is Exponential and You Should Be Worried (But Not Enough to Regulate Us)"

On June 10, Anthropic published "Preparing for the AI Exponential" — a detailed policy document arguing that AI is advancing at exponential speed while institutional policymaking remains stuck in linear gear. The paper doesn't call for slowing development (funny, that), but proposes mechanisms for governments and civil society to keep pace: standing "AI situation rooms" in executive branches, pre-negotiated international agreements that trigger at specific capability thresholds, expanded compute access for safety boffins, and "regulatory sandboxes" for testing novel applications before full deployment.

Translation: Anthropic wants guardrails, just not the kind that involve anyone saying "stop." The proposals are genuinely thoughtful in places — the threshold-based international agreements are clever policy design — but the document also conveniently positions Anthropic as the Responsible Adult in the room who should definitely be consulted on all future regulation.

What's next: Expect this to become a reference document in upcoming Congressional hearings and international governance chats. Whether you find it principled or strategically self-serving depends entirely on whether you think a company selling the thing in question should be writing the rules. (We don't have a strong opinion. We do have eyebrows, and they're raised.)

Claude Corps: Teach for America, But Make It Agentic

Announced June 11, Claude Corps is a national fellowship targeting early-career professionals who want to bring AI to underserved communities and public institutions. Fellows will embed in government agencies, nonprofits, and schools to deploy AI systems — essentially acting as translators between institutions that desperately need AI expertise and an industry that desperately needs good PR.

Credit where it's due: Anthropic is right that AI adoption is a people problem, not just a technology one. Having competent humans embedded in organizations that can barely manage their SharePoint is probably worth more than any API improvement. Applications open July 1, first cohort September 2026.

Quick Hits

• Project Glasswing expands — now covering 150+ organizations across 15+ countries (June 2). Anthropic's AI-powered cyber-defense kit reaches critical infrastructure globally. The bad actors presumably remain unimpressed.
• Slack integration pricing — Claude Tag ships within existing Enterprise and Team plan pricing, with usage counting against standard token quotas. In other words: no extra charge, just extra Claude.
• Fellowship applications for Claude Corps open July 1, first cohort starting September 2026. LinkedIn posts about "my AI fellowship journey" expected by October.
• Policy reception — the exponential paper drew praise from former OSTP directors and predictable skepticism from AI pause advocates who called it "self-serving incrementalism." Both sides have a point.