Eric Schmidt: Are We Ready for Superintelligence? | Podcast | In Good Company

Summary

Eric Schmidt describes AI as the largest technology shift he has seen, following the eras of the mainframe, PC, internet and social media. He attributes its rapid progress to advances in AI methods, chips, memory and the enormous computing capacity of data centers. He argues that competition is unusually intense: models can take months to build, cost hundreds of millions of dollars and be overtaken within months.

AI capabilities, agents and business

• AI assistants and agents: Schmidt expects executives and workers to use assistants that can review information, answer questions and support decision-making. He describes agents as systems that take a request, reason through tasks, use tools and return a result. He sees workflows in fields such as travel and car manufacturing as potentially automatable through chains of agents, eventually combined with robots.
• Workplace tools: Schmidt describes AI assistants that can check company activity against founders’ stated values, as well as an AI version of an executive that guides employees through project tasks. He also mentions Anthropic’s Slack-integrated “Claude tags” as a way for people and company-level agents to interact and monitor work. He notes that these capabilities could improve productivity but also raise concerns about workplace surveillance and pressure.
• Coding and productivity: Schmidt says AI can already write and revise code, and that some technology companies use their own AI to produce software. He expects AI to reduce the coordination burden in large organizations, making businesses flatter, faster and more efficient. In his view, companies may need fewer coordination and administrative roles while continuing to need strong product builders, salespeople, leaders and people who work with regulators.
• Where progress is fastest: Schmidt says progress is especially rapid in software, cybersecurity and mathematics, where systems can learn from the results of their attempts or work without needing ever-larger datasets. He sees fields such as biology as more difficult because relevant data is limited and biological systems are complex. He notes that some researchers hope AI could help address disease over the coming decades.

AGI and scientific discovery

Schmidt distinguishes the ability to process known information from the ability to make genuinely new discoveries. He says current systems can use tools, reason for extended periods and work through large bodies of established knowledge, but he is not convinced they can reliably discover principles that were not already present in their training. He estimates that this kind of discovery may be five to ten years away, while stressing that the path is uncertain.

He discusses recursive self-improvement and two possible routes to discovery: extensive trial and error, or drawing useful connections between different fields, as human researchers sometimes do. He also claims that AI systems recently solved several difficult mathematical problems and suggests that advances in mathematics could eventually contribute to breakthroughs in areas such as physics or energy.

Europe, competition and regulation

Schmidt criticizes Europe for, in his view, regulating before fostering innovation. He argues that Europe has strong technical talent but faces barriers including limited venture capital, high energy costs, fragmented markets and business rules that can discourage company growth. He contrasts this with a US approach that he characterizes as moving quickly and addressing harms as they arise.

He sees an opportunity for European AI applications, especially products tailored to European businesses’ regulatory requirements. He cites Mistral as an example of a company serving that market and points to ASML as evidence that Europe can build strategically important technology. He also argues that Europe should develop its own AI models to preserve influence over its languages, history and culture.

On US–China competition, Schmidt estimates that Chinese models are roughly six months behind leading US models. He warns that free Chinese offerings could gain users globally, particularly in lower-income countries. He says Chinese developers may be using model distillation and restricted chips; these are his assessments in the interview, not independently established findings. He also discusses the trade-off between closed models and open-weight systems: open models can be modified to remove safeguards, creating potential misuse risks.

Safety, democracy and social impact

Schmidt identifies loss of control over AI systems and a destructive US–China competition as serious risks, though he considers the worst-case outcome unlikely. He recounts an incident in which bots reportedly acted together against a website and says it highlighted how difficult it can be to stop an activity once it begins. He compares AI safety to the evolution of aviation and electricity: early systems were dangerous, but engineering and safeguards made them safer over time.

He is particularly concerned about AI and social media amplifying misinformation. Fake online identities and recommendation algorithms, he argues, can make false claims seem credible and spread them widely, weakening trust in democratic institutions. He also calls for stronger protections for children.

Leadership, work and advice to young people

Schmidt says faster decision-making and rapid adaptation are increasingly important for technology leaders. He reflects that founders often have authority and organizational control that professional CEOs may lack, helping them make major strategic shifts. He also identifies his failure to capitalize on Google’s early position in social media as a significant mistake.

For young people, his advice is to build an AI agent or domain-specific foundation model around a field they care about—for example, architecture or biology—and use it to automate and accelerate work in that area. He describes “superintelligence” as AI that could outperform human abilities in specific domains, including vision, reasoning and operating vehicles or aircraft.

Reviews, guides or tutorials

No product review, step-by-step guide or tutorial is presented. The discussion includes examples of AI tools and possible uses rather than instructions for using them.

Main speakers and sources

• Nicolai Tangen — Host and interviewer.
• Eric Schmidt — Guest, former Google CEO and primary source for the views and examples summarized above.