Z.ai's GLM-5.2 ships as a 744B open-weights MIT model with 1M context for coding agents — deployable with OpenClaw, nearly matching Claude Opus 4.8 performance.
Claude Code v2.1.185 doubles stream-stall timeout to 20s, reducing false retries in long-running and multi-agent sessions.
FINOS releases the AIGF MCP Server, embedding AI governance and regulatory compliance (EU AI Act, DORA) directly into agentic MCP workflows.
Headroom, a Netflix engineer's open-source tool, compresses agent context locally by up to 95% — now trending with a dedicated OpenClaw plugin.
OpenClaw 2026.6.10-beta.1 ships reliable agent state, Zalo channel support, CLI dry-run previews, and richer Discord/Telegram delivery.
Step-by-step operator checklist for upgrading OpenClaw from 2026.6.8 to 2026.6.9: Telegram HTML delivery, Codex auto-approvals, and agent recovery settings.
Ponytail forces AI coding agents to ask 'does this need to exist?' before writing code — cutting generated output 80-94% via a six-rung laziness ladder.
Google, Microsoft, Cisco, and 10+ industry partners launch ARD — a federated open protocol letting AI agents discover tools like a search engine.
Microsoft Build 2026: Azure Functions now lets you define AI agents in .agent.md markdown files — serverless, event-driven, MCP-ready.
Jensen Huang and Boris Cherny declare loops — not prompts — the unit of agentic AI value. Loop engineering is officially mainstream in 2026.