Third Post-AGI workshop finds no plausible future that keeps humans in charge

A series that starts after AGI arrives

Most debate about artificial general intelligence (AGI) focuses on whether it will be built, when, and by whom. The Post-AGI workshops begin later, with what happens next and whether any of the resulting arrangements would be worth living in, according to the Schwartz Reisman Institute.

David Duvenaud, Schwartz Reisman Chair in Technology and Society and an associate professor of computer science at the University of Toronto, convened the third meeting at Lighthaven in Berkeley, California. He co-organized it with Jan Kulveit of Charles University and researcher Raymond Douglas. Both co-authored the 2025 paper "Gradual Disempowerment" with him, which argues that humanity could lose its grip on its own institutions without a single dramatic failure.

The first workshop took place on July 14, 2025, alongside ICML in Vancouver, and asked whether any good post-AGI equilibria exist. The second, on culture, economics and governance, was held with NeurIPS in San Diego on December 3, 2025.

Three lines of inquiry and a broad speaker list

The Berkeley program was built around three questions:

• Political economy: how political systems adapt when humans are no longer economically necessary.
• The role of AI: whether AI systems will act as tools, delegates, or economic actors with standing of their own.
• Cultural and moral change: where values come from when humans are no longer the only source of new ideas.

The speakers came from beyond the usual AI safety community. They included Paul Christiano of the Alignment Research Center, Erik Brynjolfsson of the Stanford Digital Economy Lab, Samuel Hammond of the Foundation for American Innovation, Michael Muthukrishna of the London School of Economics, Helen Toner of CSET, and Sean Lavelle of the U.S. Navy. The day before, Ryan Lowe and Joe Edelman of the Meaning Alignment Institute ran a satellite event on aligning post-AGI institutions.

What Duvenaud says changed, and what did not

Duvenaud told the institute that the discussion has moved from diagnosis toward proposals. "There were finally some relatively concrete post-AGI scenarios presented," he said. One was an embargoed document called "Plan A," a follow-up to "AI 2027."

His broader assessment remains bleak. "Amongst people who've thought seriously about it, there doesn't seem to be any plausible world in the long run in which something all that similar to existing humans remains in charge," he said.

He described the debate as consolidating into two main camps: those who do not believe AI will ever be that effective, and those who conclude that humans as important beings are finished. A smaller third group, economists employed by AI labs, argues that things will be fine even if AGI ends human economic competitiveness. The biggest remaining disagreement, he said, is that some speakers still do not accept the plausibility of radical change due to AGI.

Recordings and what comes next

Most Berkeley talks were recorded, and select videos are available on the Post-AGI website. Recordings from the first two workshops are on the Schwartz Reisman Institute's YouTube channel. Christiano's Lighthaven talk is not among those released, but he will give three public talks at the 2026 Hinton Lectures in Toronto from November 9 to 11, streamed online for free.

A fourth workshop is not confirmed. "We've now hosted most of the speakers that we were really excited about," Duvenaud said. "The next iteration might be something more policy-adjacent, but there's nothing in the works yet."

For now, the series has gathered economists, political scientists and technologists around the question, but by its organizer's own account it has not yet produced a plausible account of a future worth wanting.