September 25th 2025
At the All-In Summit 2025, former Google CEO Eric Schmidt sketched a blunt, high-energy map of the decade ahead: AI systems racing toward agentic autonomy; a U.S.–China contest splitting between closed vs open-weight model strategies; and a defense revolution where cheap, swarming drones upend trillion-dollar doctrines. The conversation doubles as a playbook for founders, policy makers, and operators trying to build—and defend—in the intelligence age.
Schmidt’s cultural take landed first: early-career builders need in-person osmosis to learn fast and ship faster. Silicon Valley shouldn’t benchmark against cushy norms when its true competitor is a China tech workforce still living the spirit of “996.” In his view, the West wins by turning up the intensity—not by lowering the bar.
Contrary to the usual “who has the biggest model?” narrative, Schmidt argues China—constrained by chip export controls and shallower capital markets—is optimizing for application breadth and open-weight ecosystems that can spread globally like a “Belt & Road for models.” The West, by contrast, is largely building closed systems at frontier scale. The risk: global developers default to Chinese open weights if U.S. offerings don’t meet them in the open. His remedy: keep pushing open-weight options on small, fast models that run on edge devices, while the supercomputer-scale AGI push continues.
Expect the agentic revolution—useful, multi-step assistants that execute plans under human prompts—to transform businesses in the near term, with increasingly tight human-in-the-loop workflows (“middle-to-middle” collaboration).
On AGI, Schmidt isn’t in the “imminent” camp but sees a credible path within five to seven years to domain-level savants (e.g., physics or chemistry) before anything resembling self-set objectives (true generality).
The technical barriers today are significant. One is the problem of non-stationary objective functions: humans constantly change their goals depending on context, but AI systems struggle to adapt when the “target” keeps moving. Another is the absence of robust analogy-making across domains: geniuses often take insights from one field and apply them in surprising new areas, but current AI systems are still trapped within narrow domains of expertise. Until models can flexibly shift objectives and reason by analogy, Schmidt argues, they remain powerful tools, not independent intelligences.
Translation for operators: optimize for agent productivity now; keep a fast-follow strategy on reasoning breakthroughs, because once those blockers fall, the leap toward true AGI could come quickly.
Schmidt’s most concrete—and unsettling—section tackled warfare. He describes a battlefield where $5k drones can neutralize $30M tanks, shifting power from heavy platforms to algorithms, autonomy, and logistics at scale. Expect drone vs anti-drone dynamics, resilient comms, synthetic training data, and reinforcement-learning (RL) battle planning. Paradoxically, when each side fields millions of autonomous systems executing opaque strategies, unpredictability itself may raise deterrence—because neither side can reliably model the other’s plan. It’s a lose-lose landscape that still demands human presence to hold ground after machine attrition.
Schmidt’s civic thesis is simple: American exceptionalism isn’t tidy—it’s a noisy allocation engine that converts deep capital markets, elite universities, immigrant talent, and founder density into world-class companies. The mission: stoke it, not suffocate it—especially as aging demographics threaten growth across the developed world. Immigration, abundant compute, and energy become strategic levers to keep the flywheel spinning.
Schmidt’s Summit message isn’t about fear—it’s about focus. Build useful agents, broaden the West’s open-weight footprint, and harness America’s messy but unmatched innovation system. Do those three things, and the “chaotic advantage” should hold.
