Anthropic AI Updates: June 25, 2026

1. Anthropic Shares Lessons on Building Effective Human-Agent Teams

Anthropic published a guide on transitioning from single-player human-AI interaction to multiplayer team environments where humans and agents work as equals. The piece draws on internal experience to identify four key practices: making information broadly accessible so agents can understand organizational context, assigning clear roles with appropriate tool access, giving agents ambitious north-star goals so they can proactively suggest valuable work, and gradually expanding agent autonomy based on demonstrated reliability. Anthropic frames these as the same dynamics that define healthy human teams, but notes that agents make applying them consistently far more critical for outcomes. Source

2. Vercel Builds an Open Skills Ecosystem on Claude

Vercel built skills.sh, an open registry for Claude agent skills, and developed roughly 100 internal skills that power their data science agent. Their flagship “Deploy to Vercel” skill lets developers deploy applications directly from Claude to production, and non-developers can now contribute domain expertise to the shared skills ecosystem. Andrew Qu of Vercel noted that skills have fundamentally changed how the company works with agents, enabling teams to encapsulate processes that any agent can then reuse. Source

3. Box Adds Document Creation to Its AI Agent Using Claude Skills

Box integrated Anthropic’s Skills API into its AI agent to enable generation of PowerPoint, Excel, Word, and PDF files without building document infrastructure in-house. The integration compressed development time from months to roughly four weeks, allowing Box to ship a customer-facing document creation feature in about a month. In a demonstration, Box’s team completed a contract redline in two minutes, a task that previously required an afternoon of manual review. Source

4. Anthropic Hosts Webinar on Building an SRE Incident Response Agent

Anthropic is hosting a live webinar on June 25, 2026, focused on building a site reliability engineering incident response system using Claude. The session is part of the ongoing “Cooking with Claude” series, which guides developers through hands-on agent construction for real operational use cases. Participants can register through the Anthropic events page for details on the agenda and speakers. Source