Inside OpenAI's Plan For Your Future

A 13-page document landed on April 6th that I keep coming back to. OpenAI published it. The title is Industrial Policy for the Intelligence Age: Ideas to Keep People First.

Four days later, someone threw a Molotov cocktail at Sam Altman's house.

Altman wrote a blog post the next morning that connected those two events. He said the fear around AI is justified. He called for a society-wide response. He's not wrong to draw that line. I'm still working out whether it's the line I'd draw.

This isn't my usual lane. I spend my weeks heads-down in the build. Agents, skills, workflows. I'm not a policy writer. But the document landed the same week I was finishing a cohort of my Applied AI course — people learning how to use these tools. And I thought I knew what I'd find. A thinly disguised case for UBI, funded by the companies doing the displacing, wrapped in New Deal language.

That's not what I found. The document is more interesting than that. It's also more uncomfortable.

The paper proposes about a dozen things: robot taxes, a Public Wealth Fund modeled on Alaska's Permanent Fund, a 32-hour workweek pilot, auto-triggering safety nets, portable benefits, and — buried on page seven — a Right to AI.

That paragraph is, almost word for word, the argument I've been making to anyone who will listen. AI capability is the new literacy. If people don't learn it, they get left behind. An adjacent section proposes microgrants and "startup-in-a-box" support for workers turning domain expertise into new companies using AI. Also correct.

My first reaction: they get it. Whoever wrote this inside OpenAI understands that the real survival mechanism isn't a check. It's capability.

And then I kept reading.

On page six:

Lucia Velasco, a senior economist at the Inter-American Development Bank and former head of AI policy at the UN, summarized it in Fortune:

Alaska's Permanent Fund works because the state owns the oil rights. Citizens are owed a share because the resource belongs to all of them. In this proposal, the resource — AI capability — is owned by a handful of private companies, and those companies are invited to help decide how the fund is structured. That's a different arrangement. The displacer helps design the mechanism for the displaced.

Nathan Calvin of Encode AI called the document an improvement over OpenAI's previous policy papers, then noted that OpenAI's Leading the Future PAC spent over $2.2 million against a congressional candidate who authored the RAISE Act — the same kind of AI safety legislation the company now publicly endorses. California's SB1047, which OpenAI successfully lobbied to kill in 2024, contained almost every safety mechanism the new paper proposes, including a public compute cluster that would have operationalized the Right to AI.

A document calling for policies the publisher recently spent money to kill isn't a policy proposal. It's a positioning document.

The paper is also nearly silent on concentration. Strong on redistribution after AI capability concentrates in a small number of firms. Quiet on preventing the concentration. A wealth fund to socialize gains, yes. Nothing about dispersing control.

Tristan Harris, former Google design ethicist, sat down with Steven Bartlett on Diary of a CEO in November. He wasn't arguing about UBI. He was arguing about something that sits under the whole conversation.

His frame, borrowed from Munger: "If you show me the incentive, I will show you the outcome."

Apply that here. The companies building the AI that displaces labor have a financial incentive to make sure the replacement arrangement runs through them. That's not a moral failing. It's gravity. The structure tilts the same way regardless of the authors' sincerity.

The harder thing Harris said was about consent:

He was talking about AGI development. A policy framework published by one of those six companies, proposing the terms of the social arrangement after their technology reshapes the economy, is also a decision you didn't consent to. It doesn't matter how reasonable the ideas are. The question is whose table they were written at.

Harris made one more point. On UBI specifically, he said the math and political will, don't add up, even under abundance. The premise that AI companies could or would sustainably fund universal basic income at a scale that actually matters is, in his reading, not serious. Which leaves the thing UBI was supposed to solve - mass displacement without a transition plan, unsolved.

I worked at Microsoft for twenty years. Engineering, then sales, then Director of AI Sales. I spent the back half of that career selling cloud and AI into Fortune 500 companies.

Here's what I learned: the paycheck was a consequence. The thing underneath it was the capability - The ability to look at a problem, decompose it, propose a path, execute it. Lose the paycheck and the capability stays, which is why losing a job in your forties is survivable, even if it doesn't feel that way for the first year or so.

UBI, as the paper sketches it, hands you the paycheck and says nothing about the capability. It implicitly assumes the capability is gone.

I teach a course called Applied AI for Knowledge Work. Mid-career professionals, four weeks, no code. About four weeks in, something changes. They stop asking "will AI replace me." They start asking "what else could I build." One student restructured her research workflow and got three hours back per week. Another built stock analysis agent. Another started teaching her team.

Ethan Mollick at Wharton documents the same pattern. His rule: ten hours of serious use is the threshold. Below it, people bounce off. Above it, something clicks. In one BCG study he cites, the performance gap between top and bottom consultants shrank from 22% to 4% when AI entered the workflow. AI is a leveler. It gives the biggest boost to the people previously at the bottom.

That is, maybe, the actual story. Not "AI took my job." AI is changing what my job is. Which is the case for leaning in - not out of optimism, but out of practicality.

If OpenAI's Right to AI is real, the test is simple. They'd be advocating for public compute, open weights, free access for education, and antitrust guardrails. The company lobbied to kill CalCompute in California. The gap between the paper and the behavior is the story.

I don't know if UBI is a bad policy. I don't know if the Public Wealth Fund version is better. I'm not equipped to answer those questions definitively.

I'm more certain about this: whatever the answer is, it can't be outsourced. Not to OpenAI. Not to any one company. Not to a check. Not to a share of a fund.

If abundance is real, the only way it stays yours is if you learn to work inside it. The tools are public right now. Anyone with curiosity and a few hours a week can become genuinely capable with them. That isn't a guru promise. It's what I watch happen in every cohort I teach, and what Mollick documents in study after study.

Agency scales through participation. Picking up the tools, using them daily, building something small, teaching someone else. The people who do this become harder to displace over time. The people who wait for the dividend get whatever terms the dividend comes with.

Sam Altman's paper contains a version of this argument. I want to believe he means it. The evidence is mixed enough that I'm going to act as if he doesn't, and pick up the tools anyway.

I'd call that a reasonable hedge.

Written in conversation with Claude, across several passes. We discussed the document in detail, researched perspectives and sharpened our opinions in the process.

Thanks for reading through.
I’d love to know how you felt about today’s newsletter. This will help me make the newsletter better.