A swarm of AI agents built by Meta outperformed a team of 100 human engineers on specific tasks, according to Chief AI Officer Alexandr Wang. Wang described the system at Y Combinator's Startup School 2026, pointing to markdown files, cron jobs, and rigorous evaluation rather than a more advanced model as the reason it worked.
A swarm of AI agents built by Meta outperformed a team of 100 human engineers on specific tasks, Chief AI Officer Alexandr Wang said. Wang made the disclosure during a conversation with Y Combinator president Garry Tan at YC's Startup School 2026, where he laid out the company's approach to autonomous AI systems.
Markdown files and cron jobs, not a bigger model
Wang described infrastructure that runs on old, unglamorous tools rather than novel architecture. The agents store persistent memory in markdown files and get scheduled through cron jobs, the same Unix task scheduler that has existed since the 1970s.
The system breaks tasks into discrete loops, letting agents evaluate their own output, correct course, and iterate without human intervention. According to Wang, what pushed the swarm past 100 engineers wasn't the underlying model's sophistication but the combination of a robust evaluation method, the ability to run continuously, and a feedback loop that let agents self-improve. Wang framed the claim narrowly, noting the swarm beat engineers "on specific tasks" under the right evaluation system.
From Scale AI to Meta's AI chief
Wang founded Scale AI, the data-labeling company Meta acquired for $14.3 billion, before joining Meta as Chief AI Officer in June 2025. He carried Scale AI's evaluation-first mindset into the new role, where measuring and validating AI output was central to the business model.
Meta is also building a consumer-facing agent called Muse, internally known as Hatch, designed to handle email, scheduling, and payments across applications through a single agent that keeps context between tools. Muse is scheduled for public release in September 2026.
Source: Crypto Briefing
Trading involves risk.