AI takes the mouse, checkout moves out & more

Google released an open-source command-line interface for Workspace that includes more than 40 pre-built agent skills, enabling developers to automate tasks across Gmail, Docs, Sheets, and other apps. This tool facilitates integration of Workspace functions into broader AI-driven workflows.

Cursor introduced an automation platform that lets developers create AI agents that run continuously or trigger on specific events. Agents execute in isolated cloud sandboxes, verify their outputs, and retain memory across runs to improve tasks such as code reviews and pipeline monitoring.

Perplexity expanded its Computer product with Skills, allowing users to craft markdown-based reusable workflow snippets for consistent task automation. A forthcoming "Final Pass" mode will enhance document review capabilities.

OpenAI is developing an internal code repository platform to replace GitHub. Project started after repeated GitHub outages during Azure migration. May eventually be offered to external paying customers.

Gemini 3.1 Flash-Lite is 2.5× faster to first token than Gemini 2.5 Flash Throughput of 363 tokens / second and a 1‑million‑token context window Costs $0.25 per 1M input tokens and $1.50 per 1M output tokens (1/8 the price of Gemini 3.1 Pro)

A perspective argues that AI labs have limited time to embed safety measures before competitive pressures make them impossible to enforce.

Vietnam passed comprehensive AI law covering regulation, ethics, and enforcement. Marks one of the most extensive AI policy frameworks in Southeast Asia.

Microsoft releases Phi‑4‑reasoning‑vision‑15B, a compact multimodal model that can reason over text and images while being efficient in compute and data usage.

Google explains the technology behind Lens and Search to Image. New system can detect and search multiple objects within a single picture. Improves visual discovery for outfits, rooms, and complex scenes.