Thursday, May 21, 2026

12 articles — 🔴 3 critical , 🟡 8 important , 🟢 1 interesting

← Previous day Next day →

🤖 Models (2)

📦 Open Source (1)

⚖️ Regulation (1)

🤝 Agents (4)

🟡 🤝 Agents May 21, 2026 · 2 min read

Anthropic: MCP Tunnels and self-hosted sandboxes for Claude Managed Agents

Editorial illustration: Anthropic MCP Tunnels for private networks and self-hosted sandboxes for Claude Managed Agents

Anthropic presented MCP Tunnels in Research Preview on 19 May 2026 — a feature enabling Claude agents to connect to Model Context Protocol servers on a user's private network — and self-hosted sandboxes as an alternative to Anthropic's own infrastructure for tool execution. Updates also include dynamic MCP configuration changes within active sessions and automatic overflow of outputs larger than 100K tokens into a sandbox file.

🟡 🤝 Agents May 21, 2026 · 2 min read

Google DeepMind: Co-Scientist multi-agent AI partner for scientific research

Editorial illustration: Google DeepMind Co-Scientist multi-agent AI partner for accelerating scientific research

Google DeepMind announced Co-Scientist on 19 May 2026 — a Gemini-based multi-agent AI system that generates, debates, and refines scientific hypotheses using 6 specialised agents in a Tournament of Ideas debate. The system was developed in collaboration with more than 100 research institutions and has already produced concrete results in liver fibrosis, ALS, cellular ageing, and infectious disease research, with analysis time reduced from months to days.

🟡 🤝 Agents May 21, 2026 · 2 min read

Google: I/O 2026 round 2 — Antigravity 2.0, Gemini Spark and Universal Cart

Editorial illustration: Google I/O 2026 second wave — Antigravity 2.0, Gemini Spark and Universal Cart consolidate the agent-first strategy

At I/O 2026, Google announced the second wave of major AI launches — Antigravity 2.0 as an agent-first development platform with CLI and SDK, Gemini Spark as a persistent personal AI agent running in the background on-device, and Universal Cart as an AI shopping assistant integrated across Google services. The trio follows the Gemini 3.5 Flash and Omni announcements already covered in the previous run, and consolidates Google's agent-first ecosystem strategy.

🟡 🤝 Agents May 21, 2026 · 3 min read

LangChain: Deep Agents get QuickJS interpreters for code between tool calls

Editorial illustration: LangChain Deep Agents with QuickJS interpreters that preserve state between tool calls and reduce token consumption

LangChain introduced interpreters on 20 May 2026 — embedded QuickJS runtime environments in the Deep Agents framework that let agents write and execute code between LLM tool calls without serialising state into the message history. The company claims up to 35 percent lower token consumption on some tasks because state persists within the runtime instead of in the model context, with an explicitly controlled action space that by default has no access to the filesystem, network, or shell.

🔧 Hardware (1)

🏥 In Practice (2)

🛡️ Security (1)

← Previous day Next day →