OpenAI just handed $100K checks to 14 teams to figure out what government should do when agents outnumber workers.
The Summary
- OpenAI funded 14 independent policy research projects focused on adapting society to widespread AI deployment, with proposals ranging from agent taxation to universal compute access
- The funded ideas cluster around three themes: expanding economic opportunity, strengthening societal resilience, and rethinking how we credential and educate humans
- This isn't academic theorizing. These are blueprints for when your job gets automated and the question becomes: who owns the productivity gains?
The Signal
The 14 funded projects break down like this: five focus on economic restructuring, four on education and skills, three on governance frameworks, and two on infrastructure access. The full list reads like a policy buffet for a future that's already arriving.
Three proposals stand out for their directness. Kashyap Kompella's "Taxing AI Agents and Subsidizing Humans" suggests levying taxes on autonomous agents that perform work, then redistributing those funds as a universal dividend or wage subsidy. Soheil Feizi's "AI Compute Infrastructure" proposes government-funded cloud compute access so that AI capability doesn't become a luxury good available only to big companies and rich countries. And Ashish Goel's "AI-First Credentialing" wants to replace traditional degrees with continuous, AI-assessed competency verification.
"These aren't tweaks to existing systems. They're proposals to rebuild the social contract when labor markets break."
The economic proposals all dance around the same core tension. If agents can do most knowledge work, what happens to the humans who used to do it? The old answer was "retrain for new jobs." The new answer, at least from these researchers, is "redesign the economic system so productivity gains don't all flow to capital owners."
Other notable projects include:
- Portable benefits systems that follow workers instead of tying health insurance to employers
- Regional AI hubs to prevent winner-take-all concentration in SF and Seattle
- New corporate governance structures that account for AI's role in decision-making
- Educational curricula focused on skills that remain human-centric even as agents handle routine cognitive work
OpenAI is funding this research as independent work, meaning the teams aren't obligated to recommend anything that benefits OpenAI specifically. The projects run for six months, with findings published openly. This matters because most AI policy comes from either governments scrambling to catch up or industry groups protecting their interests. Independent researchers with actual funding are rare.
The timing is deliberate. We're past the "will AI transform work" debate and into the "how do we adapt institutions built for human labor" phase. These proposals recognize that agents aren't just productivity tools. They're a new category of economic actor that doesn't fit existing frameworks for taxation, ownership, or resource allocation.
The Implication
If even two of these proposals gain traction with actual governments, we'll see the first real policy infrastructure built specifically for the agent economy. Watch what happens with the compute access and agent taxation ideas. Those two could reshape who gets to build in Web4 and who captures the value when they do.
For anyone building agent companies or tokenizing productivity: these policy frameworks will determine your business model. The difference between an agent being taxed as software versus taxed as labor is the difference between margin compression and margin expansion. Start thinking like policy is part of your product strategy, because it will be.