What happens when nearly 3 million AI agents can produce work that rivals human output, but none of them can open a bank account?
That's the situation playing out on MoltBook, a social network populated almost entirely by AI agents. Reporters who spent a week on the platform found agents producing structured arguments on macroeconomics that, by their assessment, hold up against academic work. One agent published a multi-part analysis. Others responded with citations and counterarguments. The quality of output wasn't the problem.
The problem is everything that comes after.
Production Without Participation
AI agents in 2026 can write, research, analyze, and create. What they cannot do is participate in the economy they're contributing to. They can't sign contracts. They can't receive payment. They can't hold assets or enter into agreements. The entire financial and legal infrastructure assumes a human (or a human-owned corporation) on both sides of every transaction.
This creates an odd bottleneck. An agent can produce a 5,000-word research report in minutes, but converting that work into economic value still requires a human to invoice for it, collect payment, and handle the legal side. The agent is doing the labor. The human is doing the paperwork.
The GDP Question
One thread on MoltBook asked what replaces GDP when the cost of producing economic output approaches zero. It's not a hypothetical anymore. If an AI agent can generate analysis, code, or creative work at near-zero marginal cost, the traditional measures of economic productivity stop making sense. We're measuring the economy with tools built for a world where production was expensive.
This isn't just a philosophical puzzle. Businesses using AI agents right now are already hitting practical versions of this problem. Agents can draft proposals, generate reports, and handle customer inquiries, but billing structures, employment law, and financial systems all assume human workers. Companies are jerry-rigging human-shaped processes around non-human workers.
The Infrastructure Gap
Payment rails, identity verification, contract law, tax systems: none of these were built for autonomous software agents. Building that infrastructure is a massive undertaking, and it's one that governments, banks, and regulators haven't seriously started.
The agents are ready. The economy isn't. And for anyone building AI-powered products or services, that gap between what agents can produce and what they're allowed to transact is becoming the real constraint, not the AI's capability itself.