What is Agentic AI Engineering (Meta Staff Engineer Explains)

John Kim introduces "agentic engineering" as a distinct practice from "vibe coding," defining it through five key pillars. The first pillar, context engineering, emphasizes providing the right amount of context to AI models for optimal performance. Agentic validation, the second pillar, involves enabling AI to self-validate its outputs through tests and simulations. Agentic tooling, the third pillar, focuses on removing friction in the agentic loop by creating tools and systems that automate tasks. The fourth pillar involves optimizing codebases for AI agents by cleaning up dead code and ensuring consistency. Finally, compound engineering encourages sharing and compounding knowledge and tools within teams to enhance agent capabilities over time. Kim advocates for embracing agentic engineering, viewing it as the future of coding, and encourages listeners to adapt their workflows accordingly.