Agent swarms slash AI costs 8x

Cursor

Cursor's new agent swarm system uses hierarchical task decomposition with expensive models for planning and cheap models for execution, achieving major improvements in software engineering task quality while reducing costs up to 8x. The system successfully built SQLite from scratch in Rust, demonstrating that most AI agent work doesn't require frontier intelligence once properly decomposed.

Kimi

Kimi Work launches as an automation platform with built-in scheduling that runs LLM agents and Python scripts continuously in the background for repetitive workflow tasks.

Axios

The U.S. government is considering restrictions on Chinese AI models including liability rules and trade blacklists, as Chinese systems close the capability gap with American frontier models, escalating AI competition policy.

The Information

Google's custom Frozen chip embeds Gemini's architecture into hardware for improved efficiency, exemplifying the industry shift toward proprietary AI-optimized silicon to gain competitive advantages in model deployment and inference performance.

TechCrunch

Anthropic's $1.5 billion copyright settlement is finalized and the largest in U.S. copyright law history, but the case fails to establish industry-wide legal precedent on AI training and fair use since it was settled rather than appealed.