Anthropic AI Updates: July 2, 2026

1. Claude Code v2.1.198 Brings Claude in Chrome to General Availability

Anthropic. Claude Code v2.1.198, tagged July 1, added general availability for Claude in Chrome, letting users run the coding agent directly within the browser. The release also introduced background agent notifications through the Notification hook with new agent_needs_input and agent_completed events, plus a /dataviz skill for chart and dashboard design that includes a color-palette validator. Source

2. Anthropic Publishes Guide on Building Agentic Loops in Claude Code

Anthropic. Anthropic published an engineering post explaining how the Claude Code team designs agentic loops, defined as agents repeating cycles of work until a stop condition is met. It categorizes four loop types, turn-based, goal-based, time-based, and proactive, each suited to different scenarios. The post stresses that reliable loops require clear verification criteria, well-defined success conditions, and careful token management to balance capability against cost. Source

3. Anthropic Details How Dust Cut Model Spend While Deepening Agent Runs

Anthropic. Anthropic published a customer story on Dust, a multiplayer platform that lets teams build and deploy no-code AI agents connected to company knowledge, tools, and workflows. By optimizing a three-tier prompt caching structure and adopting the Model Context Protocol for tool integration, Dust reduced overall model spend by 18 percent, saving roughly $10,000 daily, and lifted cache efficiency from 30 to 65 percent of input tokens. The changes also expanded autonomous tool-calling depth from four or five steps to more than eight per agent run. Source