Metamate from Chatbot to Coworker by Hussein Mehanna

In this presentation, Hussein Mehanna from Meta discusses the evolution of MetaMate, Meta's internal agentic system, from a simple chatbot to a sophisticated AI coworker. He highlights its widespread adoption within Meta, with over 50% of employees using it daily for tasks like knowledge synthesis and integration with various internal tools, resulting in significant time savings. Hussein details the five phases of MetaMate's development, emphasizing the shift to model-driven orchestration (MDO) and the use of sub-agents to improve scalability, adaptability, and stability. He illustrates MetaMate's capabilities with real-world examples, such as identifying contacts, summarizing feedback, calculating user metrics, and automating meeting scheduling, and concludes by emphasizing its potential in shaping evaluation-based development within the company.