Post-AGI institutions are institutions designed on the assumption that cognitive labor is abundant and inexpensive rather than scarce. Most existing institutions have the opposite assumption built somewhere into their design, usually without saying so.
The question is a design question rather than a forecasting one. It stays worth answering across a wide range of views about how capable AI systems become and how quickly.
What is actually being priced
A university degree certifies that a person can perform a set of cognitive tasks. That certificate has value because those tasks are difficult and the people who can perform them are scarce. Research grants allocate researcher time, which is the expensive input. An accelerator compresses the interval between an idea and a product, because that interval is where startups die.
Each of these designs is a reasonable response to scarce human thinking. When the scarcity moves, the design keeps operating and stops producing the outcome it was built for. Universities keep issuing credentials that certify less. Grant committees keep allocating an input that is no longer the constraint. Accelerators keep optimizing a bottleneck that has shifted downstream.
Four things that need rebuilding
Education. If knowledge transfer is cheap, the remaining scarce capacity is judgment: knowing which questions are worth asking, recognizing when an answer is wrong in a way that matters, and holding a problem long enough to understand it. Judgment has always been formed through apprenticeship and sustained contact with practitioners rather than through curriculum, which makes it expensive and difficult to scale, which is why institutions have mostly stopped trying.
Credentialing. If a degree certifies capabilities a model also has, employers will substitute something else, and the something else is currently portfolio, network and reputation. Those are far more sensitive to who a person already knows than a degree is, so the default path here concentrates opportunity rather than distributing it.
Research funding. Grant structures assume the scarce resource is qualified researcher time. If analysis becomes cheap, the binding constraints become problem selection, physical experiment, and the patience to run something for a decade. Almost no funding instrument is designed around those.
The link between wage labor and standing. Employment currently does far more than distribute income. It provides structure, social position, and an answer to what a person is for. This is the largest of the four and the one with the least credible work behind it, and pretending otherwise would be dishonest.
What Apollo works on
Apollo Innovation Commons works on alignment through its AI and Intelligence cluster paired with Metatheory and Moral Inquiry, an approach its white paper calls alignment beyond code: research grounded in moral philosophy rather than preference optimization. The questions it takes as primary are the ones technical labs tend to defer. What counts as a good outcome. Whose values end up embedded in a system. How competing goods are adjudicated when the stakes are civilizational.
It also works on the layer around alignment, which is the subject of this page. Which structures should exist for research during this period, who governs them, how they are funded, and where the output goes. Apollo does not run an interpretability lab, and the technical safety agenda is well served by organizations built for it. The gap Apollo works in is the one between alignment research and the institutions that determine which research happens at all. That sits alongside AI governance and policy work, and it is closest in spirit to differential technology development, the principle of deliberately advancing protective technologies faster than dangerous ones instead of treating technological progress as a single dial. The idea originates in Nick Bostrom's work on existential risk and was developed into an innovation governance principle in a 2022 paper by Jonas Sandbrink, Hamish Hobbs, Jacob Swett, Allan Dafoe and Anders Sandberg.
Differential development is an institutional problem before it is a technical one. Which work gets funded, on what horizon, and who decides, are all determined by institutional design. That is the layer Apollo operates on.
The coordination gap
There is a specific class of work that no existing mechanism funds well. It is too applied for academic grants, too long-horizon for venture capital, and too cross-sectoral for conventional philanthropy, whose program areas the work tends to cross rather than fit inside.
Most of the institutional design work described on this page falls into exactly that gap, which is a large part of why so little of it exists. Apollo is built in the gap: a mixed-capital structure in which philanthropy carries early discovery risk, nonprofit operations carry formation and coordination, and regenerative capital scales proven models while recycling returns back into the commons.
Where this work happens at Apollo
The Metatheory and Moral Inquiry cluster holds this question directly. It operates at three levels: working with the other labs on the ethical questions inside their projects, shaping Apollo's own governance, and publishing research on coordination and responsibility at civilizational scale.
It works next to AI and Intelligence, Civics and Governance, and Human Flourishing, which is the arrangement the argument requires. Institutional design done at a distance from the technology it governs tends to produce recommendations that the people building the technology cannot use.
Apollo also treats its own structure as a test of the argument. It has no exit event, is designed to be permanently self-owned, and holds its mission through a purpose trust and a mission-use restriction recorded against its property. Whether those instruments hold over decades is an empirical question that only time answers, and the honest position is that nobody has run this experiment long enough to know.
Common questions
What are post-AGI institutions?
Institutions designed on the assumption that cognitive labor is abundant and inexpensive rather than scarce. Most existing institutions price scarce human thinking somewhere in their design, and stop producing their intended outcome when that scarcity moves.
Does Apollo work on AI safety?
Yes, approached through moral philosophy rather than preference optimization. Apollo's AI and Intelligence cluster works alongside Metatheory and Moral Inquiry, and that pairing produces alignment research concerned with what counts as a good outcome, whose values end up embedded in a system, and how competing goods are adjudicated when the stakes are civilizational.
Apollo also works on the layer around alignment: which research institutions should exist during this period, who governs them, how they are funded, and where the output goes. It does not run an interpretability lab.
What is differential technology development?
The principle of deliberately advancing protective and beneficial technologies faster than dangerous ones, rather than treating the rate of technological progress as a single dial. The idea originates in Nick Bostrom's work on existential risk and was developed into an innovation governance principle in a 2022 paper by Sandbrink, Hobbs, Swett, Dafoe and Sandberg.
Which institutions are most affected?
Education and credentialing, research funding, the accelerator and venture model, and the link between wage labor and social standing. The last is the largest and has the least credible work behind it.
Does this depend on AGI arriving?
Less than it appears. The design question holds across a wide range of views about capability and timelines, because the institutions in question are already showing strain from systems that exist today.
Last reviewed 31 August 2026. Maintained by Apollo Innovation Commons, San Francisco.