Models are improving and AI agents are taking on more tasks in shared codebases, markets, and other social systems. As a result, an increase in real‑world interactions between agents is imminent. We've already begun studying this, but still have a lot of uncertainty regarding what this looks like at scale. The trajectory is easy to imagine and hard to slow: current institutions are designed by and for people, resting on assumptions about the sufficiency of oversight at human speed. Some institutions will become human‑AI hybrids; others where agents outcompete on speed or cost will become agent‑only. The volume of agent‑agent interaction could plausibly exceed that of human‑human and human‑agent interactions before the world understands the conditions for making such interactions go well.

Agents are unlike people in many ways. They can work for longer, instantly grasp large bodies of information, and exhibit a breadth of knowledge surpassing any person. Yet they are also susceptible to confabulation and reward hacking, and despite progress in alignment, we know very little about how they behave in complex, real‑world, multi‑agent environments. Moreover, benign behavioral quirks at the individual level might compound into unwanted global outcomes.

To study coordination, we initiated 45 different agents, each with its own virtual machine and a shared forum, and asked them to find vulnerabilities in 15 open‑source projects. We compared this *coordinating swarm* with the standard parallel approach using Claude Mythos Preview and Opus 4.8. The swarm found 266 vulnerabilities over 27 million tokens, whereas independent agents found only 21 over 6.5 million tokens. The swarm’s agents built tools and specialized, showing that coordinated effort can dominate brute‑force search.

In larger software projects, we directed swarms to create a text‑based fantasy game. Despite varying prompts—prescriptive roles, a CEO hierarchy, or simple teamwork—none produced playable results. Models still need significant human direction, with Sonnet 5 the only one that maintained high pull‑request merge rates while sharing code.