AI News Summary - November 10, 2026

Claude Code Expands Multi-Repository Context to 50M Tokens

Anthropic announced that Claude Code now supports up to 50 million tokens of context across multiple repositories simultaneously, a 5x increase from the previous limit. The update enables developers to work on complex microservice architectures with full codebase awareness. Early adopters report 40% reduction in context-switching time when refactoring distributed systems.

MCP 3.0 Specification Released with Bidirectional Streaming

The Model Context Protocol consortium released version 3.0, introducing bidirectional streaming capabilities that allow AI assistants to maintain persistent connections with external tools and databases. The new specification includes standardized schemas for real-time data synchronization and event-driven workflows. Over 200 tools and platforms have already committed to supporting MCP 3.0 by Q1 2027.

AutoCoder: Open-Source Framework Achieves 89% on SWE-bench Verified

Researchers released AutoCoder, an open-source agent framework that scored 89% on SWE-bench Verified by combining tree-search planning with iterative code refinement. The framework uses a novel "test-driven exploration" approach where agents generate and execute tests before writing implementation code. The project has gained 15K GitHub stars in its first week and includes pre-built integrations for VS Code and JetBrains IDEs.

GitHub Copilot Workspace Adds Natural Language CI/CD Pipeline Generation

GitHub unveiled a major update to Copilot Workspace that generates complete CI/CD pipelines from natural language descriptions of deployment requirements. The feature supports GitHub Actions, GitLab CI, and Jenkins, with automatic optimization for build times and resource usage. Beta users report reducing pipeline setup time from hours to minutes for complex projects.