
Barry Zhang and Mahesh Murag introduce AgentSkills, a new approach to enhancing AI agents by focusing on skills rather than rebuilding agents from scratch. They discuss how agents, while intelligent, often lack domain expertise, which AgentSkills aims to address by providing organized collections of files that package composable procedural knowledge. These skills, stored as simple folders with scripts, can be easily shared, versioned, and used by both humans and agents. The speakers highlight the growing ecosystem of skills, including foundational, third-party, and enterprise-specific skills, and how they complement existing MCP servers. They envision a future where agents have access to a vast, evolving knowledge base, enabling continuous learning and improved performance across various domains, and emphasize the importance of sharing and distributing skills to foster a collaborative environment.
Sign in to continue reading, translating and more.
Continue