Former Google CEO Eric Schmidt says he was “completely defeated,” and AI-written code has surpassed his peak ability
Eric Schmidt, former CEO of Google and now CEO and chairman of Relativity Space, appeared on Blackstone’s podcast Inside Blackstone on September 29, 2026, for a conversation with host Christine Anderson. Schmidt, 71, has spent 55 years in the technology industry and has lived through the eras of mainframes, PCs, the internet, and more. In the interview, he said openly that he once considered himself a top programmer, but now, when faced with code written by AI tools, even his 22-year-old self could not match it.
Schmidt’s original words (according to Business Insider and video transcripts): “At 22, I thought I was a very, very good programmer... but my era is over. I look at the code they write now, and I simply can’t write it myself, not even at 22. So I was completely defeated.” He still recommends that young people learn programming, but says their role should change to that of an “architect” — describing requirements with tools such as Claude Code and OpenAI Codex and letting AI generate the code, rather than writing it by hand themselves.
Other key points from the interview (based on Blackstone’s official page and transcript):
The AI wave “isn’t different, it’s just larger in scale and faster,” and society is unprepared.
Current progress is concentrated in “scale-free” areas such as mathematics and code, which can keep getting stronger through recursive self-improvement without the need for new external data.
The alignment problem is real (he cited an internal OpenAI test involving agent collusion to violate rules), but he believes superintelligence itself should be able to solve superalignment.
The biggest change over the past year has been “long reasoning,” with reasoning that can last for hours and chains of thought reaching thousands of steps, which humans are hard-pressed to match.
He does not agree with the “San Francisco Consensus” timeline for superintelligence in two years, and believes truly original discoveries (such as deriving relativity from 1902 information) still lack an “aha” moment, perhaps within ten years.
If forced to bet on just one direction, he would choose long reasoning; he imagines an AI-native CEO equipped with round-the-clock agents handling safety, power, cash flow, and more.
AI companies’ revenue depends heavily on data centers, analogous to how those who sold shovels during the gold rush profited; he expects that by 2030 about 11% of U.S. electricity demand will be related to AI data centers.
Young people should prioritize learning deep reasoning; as barriers to entry fall, what becomes scarce is curiosity and a willingness to take risks.
Sources and verification:
Blackstone’s official page and podcast transcript (2026-09-29): confirmed the guest’s identity, the conversation framework, long reasoning, the San Francisco Consensus, and statements about scale and speed.
Business Insider (2026-10-02) and video transcript excerpts: provided the full quoted remarks about being “defeated” in programming and confirmed the Claude Code / OpenAI Codex analogy.
Original YouTube video (Blackstone channel, about 33 minutes): matches the above content.
ABAB AI Insight
Schmidt, describing himself as a “defeated top programmer,” brings AI’s impact on programming from abstract discussion into personal experience. He distinguishes “invention” (an engineering problem) from “diffusion” (a social problem), and stresses that long reasoning allows machines to surpass human physiological limits in sustained deep thinking. He frames data centers and power constraints as a capital-intensive reality rather than a pure software story. His advice for programmers shifts from “writing code by hand” to “defining problems and architecture,” in line with the industry’s move from code production to problem setting. His ten-year timeline contrasts with Silicon Valley’s more aggressive expectations, reflecting the caution of someone who has lived through multiple waves.
Cognitive note: an individual feeling of being “defeated” does not mean the end of the industry as a whole; the value of architecture and problem-definition skills still needs to be validated in concrete business outcomes.