Microsoft Research Unveils Agent Lightning for AI Training
Microsoft Research released Agent Lightning, an open-source framework to train AI agents using reinforcement learning without rewriting workflow code.
Short briefings on models, tools, agents, and the rules that govern them.
English edition
Systems that take a sequence of actions, and the question of what they are allowed to touch.
Microsoft Research released Agent Lightning, an open-source framework to train AI agents using reinforcement learning without rewriting workflow code.
Cursor launched Remote Control for iOS, letting developers manage local AI agents on their computers without sending code to the cloud.
Cohere launched a private beta of Compass Cloud, a managed enterprise search service with native MCP support, data connectors, and multimodal parsing.
OpenAI opened its Decisions API in public beta, using gpt-6-luna to deliver structured outputs in 150 ms at $0.10 per million input tokens.
AMD tests show Ryzen AI Embedded X100 handles 82% of agentic AI calls on CPU, while GPU still takes 89% of runtime in edge workflows.
OpenAI demonstrated Codex CLI running voice commands and isolating concurrent agents across Git worktrees to stop local repo file collisions.
Every launched its shared AI assistant in Slack, outsourcing orchestration to Claude Managed Agents and passing token costs with zero markup.
Visa, Mastercard, and Stripe add protections for AI agent purchases, but unresolved chargeback rules leave merchants holding the bill for mistaken orders.
Perplexity begins shifting agent sandboxes to NVIDIA Vera CPUs, aiming to cut single-threaded execution bottlenecks in automated reasoning loops.
OpenAI Agents API update introduces hosted browser automation, persistent sandboxes, and GPT-6.1 Sol support. Service remains in public beta.