AI News - September 20, 2026

GitHub Copilot Workspace Adds Multi-Repository Context

GitHub announced Copilot Workspace now supports simultaneous context from up to 10 repositories, enabling developers to work on microservices architectures more efficiently. The update includes automatic dependency tracking across codebases and intelligent suggestion routing based on which repository you're actively editing. Early access users report 40% faster development time on distributed systems projects.

Anthropic Releases MCP 2.1 with Streaming Context Updates

The Model Context Protocol specification has been updated to version 2.1, introducing real-time streaming context capabilities that allow AI assistants to receive incremental updates without full context reloads. This reduces latency by up to 75% for long-running coding sessions and enables more responsive pair programming experiences. Major implementations from Cursor, Continue, and Zed have already merged compatibility updates.

Open-Source AgentKit Framework Reaches 1.0 Milestone

The community-driven AgentKit project released its stable 1.0 version, providing a batteries-included framework for building autonomous AI agents with built-in tool calling, memory management, and error recovery. The framework supports all major LLM providers and includes pre-built integrations for file systems, databases, and web APIs. Over 2,000 production agents are now running on the platform according to telemetry data.

Claude Code Introduces "Deep Review" Mode for Security Audits

Anthropic's Claude Code editor added a specialized Deep Review mode that performs comprehensive security analysis on codebases, identifying vulnerabilities, supply chain risks, and compliance issues. The feature combines static analysis with LLM-powered reasoning to catch subtle security patterns that traditional tools miss. Initial testing shows it identifies 3x more genuine security issues while reducing false positives by 60%.