The Game Theory of Running a Company With Zero Humans

You can now run a company where every employee is an AI agent, and someone just wrote the game theory playbook for how that company should behave.

That's not hypothetical anymore. A new paper from arXiv applies formal game theory to autonomous corporations, treating them like strategic players that need to make rational decisions about cooperation, competition, and resource allocation. The research moves beyond simple agent-to-agent handoffs and asks a harder question: how should a swarm of bots act in its own self-interest?

This is the difference between a fragile bot swarm and a resilient entity. Most people building with agents today are focused on task execution. This paper focuses on strategy.

But strategy means nothing if your agents can't coordinate. Another piece of research quantifies what they call the coordination tax in multi-agent systems. The finding is blunt: adding more agents doesn't always increase output. There's a real cost to social loafing in bot swarms, where agents duplicate work, lose context, or just sit idle waiting for instructions. For anyone building a zero-human company, this is a direct warning against naively scaling headcount just because you can.

So what does the tooling actually look like for running persistent agent swarms? AutoGPT just shipped version 0.6, and it reads more like a production checklist than a feature release. The focus is on memory and coordination, which are the two hardest problems when you want agents to work together over days and weeks instead of minutes. If you're building anything that needs to run autonomously and stay coherent, this release is worth a close look.

Then there's the financial plumbing. WalletConnect is expanding its agent treasury infrastructure across more blockchains. This is boring work, but it matters because autonomous companies need to actually hold and move capital. An AI agent that can write code and make decisions is interesting. An AI agent that can pay for its own compute, receive revenue, and manage a treasury across multiple chains is a company.

And if you needed confirmation that this whole category is real, YC's latest batch keeps reinforcing the zero-human company thesis. Their blog now consistently highlights agents that can own assets and operate independently. This isn't a research curiosity anymore. It's a fundable category with real infrastructure, real theory, and real production tooling behind it.

The stack for autonomous companies is moving fast. Game theory for strategy. Production checklists for coordination. Treasury rails for capital. The pieces are fitting together, and the question isn't whether zero-human companies will exist. It's whether you'll build one or compete with one.