Ask HN: How Do Swarms of Thousands of AI Agents Work Together?

Summary

The Hacker News post asks how swarms of thousands of AI agents could coordinate after citing Andrej Karpathy’s description of a widening gap in AI access. Karpathy’s figures suggest that about 75% of people have barely used large language models, around 20% use them casually, roughly 0.2% use frontier systems professionally for mathematics and coding, and about 0.00006% have internal access to frontier-grade systems. He describes the systems seen by that small group as potentially involving thousands of agents collaborating for weeks on large software projects, cyber operations, new scientific discoveries, and advances in mathematics. The author says coordinating even two or three agents reliably is difficult and asks what architecture or operating process could make much larger swarms work. The discussion also seeks to understand what practical benefits emerge when agent collaboration is scaled to thousands.