OpenAI published Industrial Policy for the Intelligence Age last week — thirteen pages of policy proposals for navigating the transition to superintelligence. Will Manidis wrote a blistering response that is also a worthwhile read; much of it lands, particularly on the commitment vacuum (the biggest concrete offering is API credits — "a coupon for OpenAI's own store, denominated in its own currency").
But Manidis is holding OpenAI to the standard of a document it could never have published. Expecting a company approaching a trillion-dollar IPO to name unions, propose specific tax rates, and acknowledge that the New Deal was won at gunpoint is like expecting BHP to write mining tax legislation that fairly distributes super profits gained from unforeseen resource price spikes back to the public.
The more interesting question is whether any of the proposals contain the seed of something real or useful. One paragraph does, and I think Manidis dismisses it too quickly.
Help workers turn domain expertise into new companies by using AI to handle the overhead that usually blocks entrepreneurship.
Manidis calls this "telling a factory worker to learn to code, couched in policy language". He paints a picture of a paralegal in Ohio attempting to compete with companies that have access to billions in compute.
While there is some truth to the analogy, it misidentifies who this paragraph is about. The paralegal with 25 years of experience is not the factory worker in this analogy; she is the senior practitioner whose domain knowledge, client relationships, and professional judgement are precisely what AI cannot replicate and what clients are actually paying for. The overhead that kept her tethered to a firm — the research team, the document production, the back-office machinery — is collapsing. She doesn't need to compete with companies that have billions in compute; she needs to compete with her old firm, using the same AI tools (but not necessarily with the same IP, since what was in her head left with her), minus the 60% revenue share she was paying for infrastructure that AI now provides at marginal cost.
And it is not just the paralegal. It is the senior partner who realises he doesn't need his firm as much as they need him; who can keep his client relationships, use AI to handle the machinery, and retain the full value of his billings. That partner will need people — not a full bench of associates at firm rates, but project-based professionals who know how to work in an AI-augmented practice. Maybe that very same paralegal? Not a displaced factory worker learning to code, but a domain expert whose knowledge has become more valuable in a world where the mechanical work is automated and the judgement work is what remains.
The 'capital' in a professional services firm are not the ones who know how everything is done. The paralegal is Joan in this scene from Mad Men.
I wrote about this dynamic a month ago in The Defragged Economy, and it is striking to see OpenAI arrive at the same structural conclusion from the opposite direction — the practitioner's leverage shift, reframed as a policy question about supporting micro-enterprise formation at scale. The microgrant isn't seed capital for a startup; it is bridge financing, and the product lens of packaging up an easy bundle of "all the admin things you need when going solo" is actually quite clever (although it pretty much exists toady, just without the neat bow around it).
Manidis is right that this won't help a junior customer support representative. The proposal has real gaps — it says nothing about how early-career professionals build competence when the firms that used to train them are hollowing out, and the $100,000 fellowship grants are a little bit laughable from a company with $24 billion ARR. But this is not nearly as ridiculous as you might be forgiven for thinking after reading Manidis's post.
The experienced professionals who are going to leave their firms, start independent practices, and capture the full value of their AI-augmented output are not waiting for OpenAI's permission or its microgrants. They are doing the arithmetic right now.