Meta's Chief AI Officer Alexandr Wang reported that a swarm of autonomous agents outperformed a team of 100 human engineers on specific tasks. The announcement took place at Y Combinator's Startup School 2026 during a discussion with Garry Tan.
The infrastructure behind the system relies on simple, modular components rather than complex monolithic models. Agents utilize markdown files for persistent memory and are managed via standard cron jobs for task scheduling.
According to Wang, the performance gains stem from robust evaluation methods and continuous feedback loops that allow agents to iterate without human intervention. He noted that the evaluation system is more critical than the model's sophistication.
Wang, who founded Scale AI, joined Meta in June 2025 following the company's acquisition of his startup for $14.3 billion. He currently oversees the development of Meta's agentic systems.
Meta is also preparing to launch Muse, a personal AI agent previously known as Hatch. The tool is designed to manage emails, scheduling, and payments across various applications, with a public release scheduled for September 2026.
