AI News Summary - October 14, 2026

Claude Code Introduces Multi-Repository Context Windows

Anthropic's Claude Code now supports simultaneous context awareness across up to 10 linked repositories, enabling developers to refactor code spanning multiple microservices. The update includes a new "dependency graph visualization" that shows how changes propagate across codebases in real-time. Early adopters report 40% faster cross-service refactoring compared to single-repo workflows.

MCP 2.3 Specification Adds Streaming Tool Results

The Model Context Protocol consortium released version 2.3, introducing bidirectional streaming for tool execution results and partial response handling. This allows AI agents to begin processing tool outputs before completion, significantly reducing latency for long-running operations like database queries or API calls. Over 15 major AI frameworks have committed to implementing the new spec within 30 days.

Open-Source AgentFlow Framework Reaches Production Readiness

AgentFlow 1.0, an Apache-licensed framework for building multi-agent systems, launched with native support for error recovery, agent supervision hierarchies, and built-in observability. The framework includes pre-built templates for common patterns like ReAct, Chain-of-Thought, and Tree-of-Thoughts reasoning. Initial benchmarks show 3x better task completion rates compared to custom agent implementations on the AgentBench suite.

GitHub Copilot Workspace Adds Live Collaboration Mode

GitHub announced Copilot Workspace now supports real-time collaborative AI pair programming, where multiple developers can simultaneously interact with the same AI context and see each other's prompts and results. The feature includes conflict resolution when developers request contradictory changes and maintains a shared conversation history. The update rolls out to Enterprise customers starting next week.