We’re launching Lyria 3.5 in Google Flow Music, with advances across musicality, lyrics, vocals, and creative control
deepmind.google·6d ago
AI Engineering
VS Code Blog·4 min read
Your Home for Multi-Agent Development
TL;DR
Researchers identified a gap in LLM evaluation benchmarks. They built a synthetic dataset with 10k adversarial prompts targeting reasoning failures.
✦ Why It Matters
Use this benchmark to audit LLM robustness before deploying in production reasoning pipelines.
Key Takeaways
- Utilize the Agent Sessions view to manage all your agents in one place.
- Run Claude and Codex agents under your GitHub Copilot subscription.
- Leverage subagents for parallel task execution to save time.
- Explore open standards like MCP Apps for enhanced agent capabilities.
Full Summary
The latest VS Code update (version 1.109) focuses on multi-agent development, allowing users to run Claude and Codex agents directly alongside GitHub Copilot. This update introduces an Agent Sessions view, which consolidates all agent sessions—local, background, and cloud—into one interface, making it easier to manage tasks.
Users can now choose between local agents for interactive tasks, background agents for unattended operations, and cloud agents for more complex jobs. Additionally, the update supports running multiple subagents in parallel, enhancing efficiency by isolating tasks and reducing context switching.
Open standards like MCP Apps and Agent Skills have also been introduced, enabling richer interactions and specialized capabilities for agents. This unified approach aims to simplify the development process and improve collaboration between human developers and AI agents.