Anthropic AI Updates: July 13, 2026

1. Claude Code Gains a Built-In Browser That Reads, Clicks, and Types on Websites

Anthropic. Claude Code added a built-in browser that lets the agent read pages, click elements, and type into forms on external sites, with guardrails that require user approval before actions like transactions or account creation. The feature pushes agentic coding tools further into full web automation, letting developers hand off tasks that span code and live web interfaces. It also raises the familiar tradeoff between autonomy and control that teams must manage when granting agents access to real accounts. Source

2. Anthropic Says Claude Cowork’s Top Use Is the Office Work Nobody Wants

Anthropic. Analyzing about 1.2 million sessions, Anthropic reported that roughly half of Claude Cowork usage is mundane administrative work such as status reports and slide decks rather than flashy creative tasks. The data suggests the near-term enterprise value of agents lies in absorbing low-status routine work that employees are happy to offload. It is a useful reality check for teams sizing where agentic tools actually save time today. Source

3. Anthropic Hosts a Live Session on Production-Grade Agents Built on Claude Sonnet 5

Anthropic. Anthropic hosted a webinar featuring Zed and ClickHouse demonstrating how to run production-grade agent applications on Claude Sonnet 5 at scale, with live product demos, a look at design decisions behind sustained high-volume agent work, and Q&A with an Anthropic technical staff member. The session focused on price-to-performance optimization for agent workloads and practical deployment strategies for real-world results. As Sonnet 5 becomes a default model for agentic coding and knowledge work, Anthropic is leaning on partner practitioners to show developers how to move from prototypes to sustainable, high-throughput production systems. Source