AI News for April 15, 2026

GitHub Copilot Workspace Adds Multi-Repository Context Understanding

GitHub announced that Copilot Workspace can now maintain context across up to 10 linked repositories simultaneously, enabling more sophisticated monorepo and microservices development workflows. The update uses a new contextual embedding system that reduces token usage by 40% while improving code suggestion accuracy. Early access users report significant improvements when working on distributed systems architectures.

Anthropic Releases MCP 2.1 with Bidirectional Tool Streaming

The Model Context Protocol 2.1 specification now supports bidirectional streaming for tool execution, allowing AI assistants to receive real-time feedback during long-running operations. This enables more responsive agent behaviors for tasks like file processing, database migrations, and API integrations. The update is already supported in Claude Code and has been adopted by major MCP server implementations including mcp-server-postgres and mcp-server-filesystem.

Open-Source Agent Framework "Conductor" Reaches 1.0

Conductor, a lightweight Python framework for building multi-agent systems, has released version 1.0 after 18 months in development. The framework emphasizes deterministic agent orchestration with built-in observability, making it easier to debug complex agent workflows. It features native MCP integration and has already been deployed in production by several enterprise teams for automated code review and infrastructure management tasks.

Cursor Introduces "Shadow Mode" for Risk-Free AI Refactoring

Cursor's latest update includes Shadow Mode, which runs proposed refactoring operations in an isolated environment and presents a diff-based preview before applying changes. The feature uses static analysis to detect potential breaking changes and automatically generates rollback scripts. Developer beta testers report 65% faster adoption of AI-suggested refactoring compared to traditional approve/reject workflows.