The question isn't whether to regulate AI — it's which problems actually need government intervention versus which ones markets and people can handle on their own.
The Summary
- Economist Jason Furman breaks down AI regulation into three buckets: problems requiring government action, problems markets can solve, and moral questions without clear answers
- The bioweapon-to-job-loss spectrum demands different regulatory approaches — not one-size-fits-all policy
- The hardest AI governance questions aren't technical or economic, they're moral
The Signal
Furman is doing something rare in the AI regulation debate: drawing actual boundaries. Most policy conversations treat "AI governance" like a single problem. His framework splits it into three distinct categories, each requiring fundamentally different responses.
The first bucket is existential risk and clear harm prevention. Bioweapons, critical infrastructure attacks, autonomous weapons systems. This is where government regulation isn't just appropriate, it's necessary. Markets don't price in civilization-ending risks well. The private sector optimizes for profit, not species survival. No startup's terms of service can prevent a bad actor from training an AI to design novel pathogens.
"From bioweapons to job losses, not every AI problem should be solved the same way."
The second bucket is economic adjustment, where Furman argues markets and social systems should do the heavy lifting. Job displacement from AI follows the same pattern as previous technology waves. The speed might be faster, but the playbook exists: retraining programs, social safety nets, labor market flexibility. Government's role here isn't to stop the technology, it's to smooth the transition for people caught in the shift.
This is where the agent economy gets real. When your job becomes a prompt, government blocking that technology doesn't help you. But portable benefits, skills training, and income support during transition periods do. The Fourth Web isn't going to wait for policy consensus.
The third bucket is the thorniest: moral questions without clear answers. Should AI systems make life-or-death medical decisions? Who owns the output when an AI agent creates something valuable? What counts as "fair" when algorithms allocate opportunities? These aren't problems you solve with regulation or markets. They require ongoing social negotiation about values.
The Implication
If you're building in the agent economy, understand which bucket your product sits in. Compliance theater won't save you from real risks, but overcautious self-regulation might kill you before necessary rules even exist. The companies that thrive will be the ones that can distinguish between "this needs government rails" and "this needs market discipline" and "this needs society to decide what we actually want."
Watch for jurisdictional arbitrage. Different countries will make different calls on which bucket each AI application belongs in. That creates opportunities and risks. The regulation gap between the EU's precautionary approach and the US's market-first stance is already shaping where certain AI products can launch first.