Andrew Ng’s recent fireside chat offered sharp insights for AI builders and businesses, emphasizing thinking about “agenticness” as a spectrum rather than a binary, and highlighting that current business opportunities often involve translating linear workflows into agentic ones. He underscored the critical importance of evaluation (evals) for agent builders, alongside strong LLM understanding, broad tool knowledge, and adaptability, while also pointing to underrated areas like the voice stack and AI-assisted coding. Ng also noted that while Model-driven Coordinated Programming (MCP) is promising for standardizing agent connections, true agent-to-agent communication is still nascent, and for startups, speed and technical knowledge are the top predictors of success.