AI News Summary - September 18, 2026

Cursor Launches Multi-Repository Context Engine

Cursor has released version 0.45 with a groundbreaking multi-repo context feature that allows AI coding assistants to maintain awareness across up to 10 linked repositories simultaneously. The update dramatically improves accuracy for microservices development and monorepo workflows. Early benchmarks show a 40% reduction in context-switching errors compared to single-repo approaches.

Anthropic Expands MCP with Real-Time Collaboration Protocol

Anthropic announced MCP 2.1, introducing standardized real-time collaboration primitives that allow multiple AI agents to coordinate on shared tasks through a unified message bus. The update includes built-in conflict resolution, state synchronization, and a reference implementation for multi-agent coding sessions. Over 15 major AI development tools have committed to implementing the new protocol by Q4 2026.

Open-Source Agent Framework "Conductor" Reaches 1.0

The Conductor framework, designed for building production-grade AI agents with memory persistence and tool orchestration, officially hit version 1.0 after 18 months of development. Key features include native MCP support, built-in observability, and a plugin system that's already spawned 200+ community extensions. The framework is backed by a consortium of AI startups and has been adopted by three Fortune 500 companies.

GitHub Copilot Workspace Adds Automated Testing Generation

GitHub announced that Copilot Workspace now automatically generates comprehensive test suites when reviewing pull requests, covering unit, integration, and edge cases. The feature uses static analysis combined with Claude 4.5 to achieve 85% code coverage on average. Developers can accept, modify, or reject suggested tests directly in the PR review interface.