Epoch AI researchers Philip Trammell and Anson Ho map four post AGI redistribution designs and show that each one encodes a distinct theory of what citizens owe the state, and what it owes them.
If artificial general intelligence does what its builders hope, it will eventually perform most paid work at a lower cost than the workers it displaces. Two outcomes then become possible at once. Wages can collapse for huge fractions of the population, leaving people with no market income at all. Or, even where work remains, the share of national income going to capital owners can climb, because the machines that replace labor are owned by someone, and the dividends flow somewhere.
A new Epoch AI Gradient Updates piece dated June 10, 2026, by researchers Philip Trammell and Anson Ho, lays out that fork clearly and then does something more useful than forecasting it. The piece is a simple taxonomy of the main proposals for post-AGI universal redistribution, and the authors are upfront that the mapping is the point. "This post does not advocate for any policy," they write. "It aims simply to explain the landscape."
The four mechanisms that dominate the conversation, as Trammell and Ho describe them, look at a glance like variations on giving people money. Each actually encodes a different theory of what a citizen is for, and what the state owes them in return.
Universal basic income is the simplest. The state cuts a regular cash payment to every resident, no conditions attached. The recipient decides what to do with it. The value encoded is freedom: cash is fungible, and the recipient is treated as the best judge of their own life. The cost is exposure. A check that arrives every month does not insulate a household from a medical bill, a rent shock, or a price spike in food or energy. People with thin savings and weak financial infrastructure ride out the transition on their own balance sheet.
Universal basic services flips the design. The state does not hand out cash. It provides a defined bundle of services for free, covering basic needs: housing, healthcare, transit, food, and connectivity. The value encoded is provision. The state guarantees the outcome (you will be housed, you will be seen by a doctor), and asks the citizen to take what is offered rather than choose. It is also the design where the political fight collapses to a single question: how generous is the bundle, not whether it exists.
Universal basic capital is the least familiar of the four. The state grants every adult a one-time endowment of financial or productive assets, often held in a trust, and the returns on those assets flow to the holder as a dividend. The value encoded is stakeholding. Everyone owns a slice of the productive economy, in the same way, so the citizen's relationship to capital is not mediated by an employer or a landlord. The risk is also distinctive. A capital endowment that arrives once, at age 18, has to last a working life. If the underlying assets are illiquid, risky, or politically captured, the holder is stuck with the consequences. The mechanism rewards patience and punishes haste in ways that cash does not.
Sovereign wealth funds are the oldest design on the list. The state itself owns a portfolio of domestic and foreign assets, and the returns pay for public goods, services, or transfers. The value encoded is collective ownership. The citizen owns a share of the national portfolio indirectly, through the state, rather than directly. The historical track record is mixed. State-owned portfolios have compounded for decades in some countries and remained politically stable; in others they have been raided, politicized, or diluted. The mechanism works when the fund is treated as a long-duration institution, and tends to break when it is not.
Trammell and Ho are careful to scope their argument. The piece brackets two questions that any real redistribution policy will have to answer: how the state raises revenue after AGI, and what share of GDP it spends. The taxonomy is about the third question only: given a pool of money, what is the cleanest way to push it to everyone? That separation matters, because it keeps the four mechanisms comparable on their own terms rather than tangled up with the separate fight over the size of the pot.
Two readers of the same article can walk away with opposite views of which mechanism is best, depending on what they think the post-AGI settlement is supposed to accomplish. The Trammell-Ho framing makes those commitments explicit. UBI asks the state to get out of the way of the individual. UBS asks the state to take responsibility for outcomes. UBC asks the state to give every citizen ownership of the productive economy. SWF asks the state to be the productive economy's largest owner on the citizen's behalf.
None is dominant. Each is a defensible answer. And the cost of choosing wrong is asymmetric. A UBS bundle that turns out to be too thin to cover real needs becomes a permanent feature of the welfare state until politics changes it. A SWF that is captured by the government of the day is harder to fix than a UBI rate, because the fund's losses are not visible in any single year's budget. A UBC endowment that gets captured at the moment of distribution cannot be redistributed later in life. A UBI payment that is too small to live on is the easiest mistake to correct, because the next budget cycle can change the number.
The piece is also clear-eyed that AGI may not arrive in a way that gives people time to prepare. Households without savings, without liquid assets, and without access to financial advice are the most exposed to a fast displacement shock. The four designs handle that transition differently. UBI handles it by paying continuously. UBS handles it by guaranteeing specific categories of consumption. UBC handles it by handing out capital once, which means households that cannot manage the endowment are immediately worse off. SWF handles it only indirectly, through whatever public spending the fund's returns pay for.
Trammell and Ho do not forecast which mechanism "wins." They note that some versions of the four can be combined. They also note, without resolving, that the post-AGI settlement will depend on choices that the piece deliberately does not make. The taxonomy is a map. The territory is not yet a country.
The value of the piece is that it makes those choices visible. The conversation about post-AGI redistribution is often framed as a binary over UBI. The Trammell-Ho taxonomy makes a more useful question visible: what theory of the citizen should the redistribution encode, and which design will work for the households least able to absorb a transition they did not choose.