Beyond the Controversy: How Schmidt’s 2024 Forecasts on AI Agents and NVIDIA Are Playing Out in 2026

Eric Schmidt’s August 2024 appearance at Stanford was an informal classroom conversation with economist Erik Brynjolfsson in the course “The Age of AI” (ECON295/CS323). Stanford posted the video, it circulated quickly, then was set to private after Schmidt asked for its removal. The episode became famous less for its technical content than for two unfiltered remarks that leaked out.

The two remarks that got the video pulledSchmidt said Google had chosen “work-life balance and going home early and working from home” over winning the AI race, while startups succeed because “the people work like hell.” He later told the Wall Street Journal he “misspoke about Google and their work hours” and requested the video come down.

The second flashpoint was his hypothetical for students if TikTok were banned: tell an LLM to “Make me a copy of TikTok, steal all the users, steal all the music… produce this program in the next 30 seconds.” He then added that if the clone took off, “you’d hire a whole bunch of lawyers to go clean the mess up.” He immediately tried to walk it back on camera (“I was not arguing that you should illegally steal everybody’s music”) but the clip spread anyway.

Those lines overshadowed the rest of a fairly substantive discussion.Schmidt’s actual forecast: three capabilities landing togetherHe argued that the next 12–24 months would be defined by the combination of three things:Very large context windows (he cited Anthropic moving toward a million tokens and predicted others would follow). He compared them to short-term memory: you can now feed the model 20 books and ask it to reason across them.
AI agents that can read a domain (his example was chemistry), extract principles, run tests, and fold the results back into their own knowledge.
“Text-to-action” — the ability to turn natural language into executable steps, effectively giving every person “their own programmer.”

“When they are delivered at scale, it’s going to have an impact on the world at a scale that no one understands yet. Much bigger than the horrific impact we’ve had by social media.”He also expected longer chains of reasoning (“a thousand-step recipe”) that could be verified by running the recipe and checking the output.What else he discussedThe frontier-model gap was widening again after a period when he thought it was closing (he said he revises his view every six months).
NVIDIA’s lead was structural because of CUDA optimizations that competitors struggle to replicate.
Founders, not professional managers, needed to stay in charge during this phase.
He floated an “equal time” rule for TikTok-style platforms and worried that social media companies were poorly organized to handle election-year misinformation.
He was relatively skeptical that true AGI-style systems were imminent, though he did not treat the definition as sharp.

Looking back from 2026Several of the technical predictions landed. Context windows grew dramatically (Gemini 1.5 and later models moved into the multi-million-token range). Agent frameworks and “computer-use” or tool-calling systems became a central product category. The “everyone has a programmer” idea is now visible in coding agents and no-code/low-code tools, even if the quality and reliability remain uneven. The geopolitical framing—U.S.–China competition as the defining narrative—has only intensified.The work-culture and IP comments remain the parts most people remember. They revealed how bluntly some Silicon Valley veterans still talk when they think the cameras are off. The rest of the talk is a useful snapshot of how a prominent insider saw the 2024–2026 window: not as science-fiction takeoff, but as a sudden jump in practical capability that would still surprise most people.