The Diary of a CEOEric Schmidt: Why AI needs divas, fast fails, and a plug
Schmidt scaled Google from 100 million to 180 billion dollars. He explains AI misinformation risks, the chip and energy race, and a plug rule for safety.
CHAPTERS
- 2:05 – 3:49
Origins, Genesis, and Why AI Is a Historic Inflection Point
Schmidt explains why he chose to write "Genesis" with Henry Kissinger, recounting how Kissinger became alarmed about AI after hearing Demis Hassabis. He frames AI as the first true intellectual challenger humanity has faced, forcing us to reconsider what it means to be human as these systems begin to rival or surpass our cognitive abilities.
- •Genesis grew out of a decade‑long dialogue with Henry Kissinger about AI’s philosophical and geopolitical implications.
- •Kissinger recognized early that AI research was “playing with fire” whose impacts we didn’t fully understand.
- •AI represents a new kind of challenge: humans have never had a peer or superior in generalized intellectual capability.
- •The core question becomes how human life and thought change when an artificial intelligence sits alongside us.
- 3:49 – 7:49
Early Career, Moore’s Law, and Advice for an 18‑Year‑Old
Schmidt traces his path from tinkering with rockets and early slow computers to riding the PC and internet wave, emphasizing how Moore’s Law underpinned his luck. He then advises an 18‑year‑old: cultivate critical thinking in any domain you love and learn Python as a practical gateway to building with AI.
- •The computer Schmidt used at university was ~100 million times slower than a modern smartphone, illustrating Moore’s Law.
- •His career success is partly "being born lucky" into an era where computing density exploded.
- •For young people, domain matters less than developing deep analytical skills that transfer across careers.
- •Python is recommended as the most accessible and AI‑relevant language; making a game is a motivating entry project.
- •AI writing code does not make programming obsolete; instead it shifts programmers into orchestrators of AI tools and APIs.
- 7:49 – 13:40
Critical Thinking, Social Media, and TikTok’s Addictive Algorithms
Schmidt defines critical thinking as the ability to test assertions against facts rather than accepting socially plausible statements. He critiques social media and TikTok’s bandit algorithms for deepening echo chambers, outrage, and teen mental health issues, while outlining how platforms could choose more ‘moral’ and sustainable revenue models.
- •Critical thinking = checking claims, understanding falsifiability, and refusing to propagate unverified or false information.
- •Social feeds have normalized believing things because they’re widely repeated or fit existing biases.
- •TikTok exemplifies the ‘multi‑armed bandit’ algorithm: maximize attention by showing what keeps users hooked and occasionally testing adjacent content.
- •Outrage is a highly effective driver of attention and thus revenue, but corrodes public discourse and mental health.
- •Evidence links algorithmic shifts around 2015 (from chronological to hyper‑targeted feeds) to rising anxiety, self‑harm, and ER visits among teenage girls.
- •Schmidt proposes distinguishing ‘good revenue’ from ‘bad revenue’ and deliberately dampening harmful but lucrative content, as Google did by keeping the worst material off page one.
- 13:40 – 18:38
Children, AI Companions, and the Future Identity of the Next Generation
The conversation turns to how AI and social media might ‘reprogram’ children whose best friend could be a computer from birth. Schmidt is cautiously optimistic that society will adjust, but he stresses we are effectively running an uncontrolled experiment on a billion young people with unknown developmental consequences.
- •Adults’ core values are comparatively stable; children’s identities are highly plastic and vulnerable to algorithmic influence.
- •A core question in the AI revolution: what happens when a child’s closest, most responsive companion is an AI system?
- •We are conducting a massive, uncontrolled social experiment across generations without a control group.
- •Despite risks, Schmidt believes societal recognition of harms (e.g., school phone bans, greater awareness of teen girl distress) will gradually shape healthier norms.
- •He predicts longer, healthier, less war‑torn lives for today’s children if we manage the transition well.
- 18:38 – 22:01
Entrepreneurship, Divas vs. Knaves, and Building for Scale
Drawing from Google and other tech giants, Schmidt outlines principles for building great companies: tie yourself to truly exceptional ‘divas,’ avoid knaves, and design businesses that can scale from zero to millions of users. He emphasizes the centrality of AI to all future scalable businesses and the importance of focusing on big, transformative problems rather than small ‘widgets.’
- •Great products usually require at least one truly brilliant ‘diva’ founder or leader who demands excellence and sees around corners.
- •‘Knaves’ operate primarily for self‑interest at others’ expense and can poison teams and cultures.
- •Founders like Gates, Ellison, Page, Brin, and Musk exemplify repeated reinvention and high‑conviction risk‑taking.
- •Schmidt is uninterested in small, non‑scalable widgets; he seeks ideas that can theoretically scale to vast user bases.
- •The archetype of a future successful company: cross‑platform apps (Android, iOS) backed by large AI models running in powerful data centers.
- •AI shifts programming from explicitly coding logic to systems that learn and optimize; translation and next‑token prediction generalize to biology, robotics, and beyond.
- 22:01 – 28:50
Larry & Sergey, Google’s Culture, and the 70‑20‑10 Innovation Engine
Schmidt recounts the scrappy origins of Google, the brilliance of Larry Page and Sergey Brin, and how their long‑term mission shaped the company’s culture. He explains the 70‑20‑10 resource allocation model, the role of technical culture and measurement (OKRs, A/B testing), and how Google tried to remain fast while scaling massively.
- •Larry and Sergey met as Stanford grad students, built PageRank, and hacked together early hardware with corkboard that literally caught fire.
- •They thought in decades with the mission "organize all the world’s information" rather than just "build a search engine."
- •Google’s 70‑20‑10 rule enabled small teams like Google Brain to generate tens of billions in profit from 10–15 people.
- •Culture is typically set by founders and can persist for a century (e.g., Mayo Clinic’s "needs of the patient come first").
- •Schmidt favors heavily technical organizations where product excellence precedes sales and marketing muscle.
- •Marissa Mayer pioneered pervasive A/B testing and metrics like dwell time to assess UI and product decisions scientifically.
- •OKRs imposed ambitious quarterly numeric goals; 70% achievement was considered good, but everything was measured.
- 28:50 – 45:25
Innovation in Big Companies, Focus vs. Exploration, and Crisis Moments
The discussion explores why big firms struggle to innovate and how to structure for disruption. Schmidt highlights the need for entrepreneurial ‘owners’ inside large orgs, explains why disruptive teams often must be physically and culturally separated, and reflects on misses like Google Video vs. YouTube and Intel’s failure in mobile.
- •Founders or founder‑like leaders must deliberately appoint ‘owners’ of new bets; without a clear owner, nothing transformative ships.
- •Simultaneously harvesting core business and hunting disruptive innovations in the same org unit is rarely successful.
- •The Macintosh team’s pirate‑flag building illustrates the need for protected, renegade subcultures—even at the cost of internal resentment.
- •Boards and hired CEOs struggle to authorize multi‑billion‑dollar, high‑risk bets that temporarily depress profits; founder authority matters.
- •Intel’s sale of its Arm‑based line, seen as simplification, later proved catastrophic when mobile and low‑power chips dominated.
- •Google’s YouTube acquisition came after recognizing a startup was out‑executing its internal Google Video team, partly due to fewer constraints.
- •Schmidt stresses writing a concrete 5‑year vision (tech, competition, hardware, networks) to guide which product lines to keep or kill.
- 45:25 – 55:42
AI Race, ChatGPT, Gemini, and Why Google Wasn’t First
Schmidt explains how OpenAI’s use of RLHF unexpectedly turned GPT‑3 into a breakthrough product and why even its founders underestimated it. He compares OpenAI’s products with Google’s Gemini and Meta’s LLaMA, and touches on legal and cultural differences between closed and open ecosystems, like Apple’s.
- •Google “was in the engine room” advancing transformers but OpenAI’s insight—reinforcement learning from human feedback (RLHF)—unlocked a step‑change in usability.
- •The ChatGPT launch was almost an afterthought while OpenAI worked on GPT‑4; its success surprised even them.
- •Today, GPT‑4o, Gemini 1.5, LLaMA, and Anthropic’s models are in a competitive cluster, each with different strengths (e.g., multimodality, open‑source licensing).
- •Anthropic deliberately structured as a public benefit corporation to resist future pressure to prioritize revenue over safety.
- •Apple under Jobs and Cook has remained a tightly integrated, closed ecosystem; Jobs prioritized ‘BMW margins’ and design perfection over openness.
- •AI at Apple (e.g., replacing Siri with true AI capabilities) is framed as inevitable; Jobs would likely have used it to build beautiful, closed experiences.
- 55:42 – 1:04:02
Hiring, Failing Fast, Microcultures, and Internal Politics
Schmidt offers hiring advice for startups, emphasizing intelligence and risk appetite over experience. He discusses Google’s fail‑fast philosophy, the 70‑20‑10 rule in practice, TGIF’s rise and fall, and why he resisted routine layoffs; he also notes how internal activism and politics forced a later culture reset.
- •Startups should over‑index on raw intelligence, speed, and risk tolerance, which often correlates with younger hires.
- •Corporations cling to false belief systems for years, while startups pivot quickly; the key is having metrics that expose failure early.
- •Bill Gates popularized “fail fast”; Schmidt operationalized it through structured resource allocation and aggressive project shut‑downs.
- •Google’s TGIF began as a humorous, intimate, off‑the‑record forum, but live leaks during meetings signaled the end of that intimacy at scale.
- •Schmidt opposed routine "bottom x%" mass layoffs, preferring not to over‑hire in the first place and valuing even lower performers’ institutional knowledge.
- •Proliferation of internal political mailing lists (~100,000) forced Google to clarify what topics are appropriate on company time; Schmidt advocates focusing work discourse on the business, not partisan politics.
- 1:04:02 – 1:09:12
Competition, Deadlines, Business Plans, and the Power of Product
The conversation turns to strategy execution: how much attention to pay to competitors, the role of deadlines and OKRs, and whether detailed business plans really matter. Schmidt insists companies should obsess less over rivals and more over unique, delightful products and concrete near‑term goals grounded in a clear 5‑year view.
- •Schmidt advises largely ignoring competitors’ moves and instead asking, "What can we do that no one else can?"
- •Deadlines and numeric objectives (OKRs) are essential in big firms; without them, everyone feels busy but little of lasting impact is achieved.
- •Most business plans are wrong in hindsight; Google’s early plan is a rare exception that was oddly accurate.
- •He teaches a planning method: define the likely world in 5 years, then set ambitious but achievable 1‑year execution targets aligned with that vision.
- •In consumer products, an audience of 10–100 million users is the primary asset; monetization (ads, sponsorships, donations) can usually be engineered afterwards.
- •"Focus on the user and everything else will follow" encapsulates Google’s product‑centric philosophy.
- 1:09:12 – 1:12:17
Media in the Age of AI: Notebook LM and Content Abundance
Schmidt demonstrates how tools like Notebook LM can generate convincing synthetic podcast dialogues and suggests how creators should adapt to a world of near‑zero content production costs. He argues that instead of erasing moats, AI will amplify top creators, who must learn to harness it for summarization, spin‑offs, and new formats.
- •Notebook LM, built on Gemini, can take a text and synthesize a realistic dialogue between two non‑existent speakers—fooling live audiences.
- •This represented Schmidt’s personal "ChatGPT moment" for 2024 in terms of visceral impact.
- •Creators should use AI to auto‑annotate, expand, and critique their own long‑form content, then repackage it into derivative shows and formats.
- •Past fears that the internet would eliminate celebrity were wrong; networks expanded the reach of the very best voices.
- •In an abundant content world, human charisma, trust, and narrative skill remain scarce; AI becomes a multiplier of human brand, not a replacement.
- 1:12:17 – 1:23:45
AI as a Survival Question: Risks, Governance, and Guardrails
Schmidt elaborates on his claim that AI is a question of human survival, not because it will automatically wipe us out, but because it could destabilize democracy, warfare, cybersecurity, and biosecurity if misused. He emphasizes the unprecedented speed of AI progress, the emergence of powerful raw models, and the need for global safety regimes analogous to nuclear non‑proliferation.
- •AI progress is scaling in capability at 2–4x per ‘turn of the crank’ with no clear slowdown yet.
- •Near‑term, the biggest technical dangers are: sophisticated cyberattacks (including zero‑days), engineered biological threats, and new forms of autonomous or semi‑autonomous warfare.
- •Raw, unreleased models are far more capable than public versions; they must be aggressively red‑teamed and constrained before deployment.
- •Trust and safety teams test models for harmful outputs (self‑harm, violence, etc.) and block them; UK and other governments now convene global AI safety summits.
- •Schmidt envisions highly secure data centers for top‑tier models, potentially guarded like plutonium sites, or at least strictly controlled and audited.
- •A small number of top‑end clusters (US, UK, China, a few others) would be more governable than uncontrolled proliferation to many actors, including terrorists.
- 1:23:45 – 1:27:56
Geopolitics, China, Autonomous War, and Proliferation Fears
The discussion moves to how adversarial states like China and Russia might use AI and how warfare is already being transformed in Ukraine. Schmidt expects China to deploy AI within its censorship framework but is more worried about AI’s role in drone warfare, cyber conflicts, and the lack of clear international rules.
- •Schmidt believes China’s AI will be heavily constrained by its aversion to free speech, fundamentally differing from Western models.
- •Every major technology, from tanks to aircraft, has been militarized; AI and drones are no exception.
- •Ukraine’s war with Russia showcases new warfare where cheap drones can destroy expensive tanks (favorable kill ratios), foreshadowing drone‑on‑drone conflict.
- •Future combat may involve operators in offices inflicting harm remotely, changing the psychological and ethical character of war.
- •Global norms for AI weapons and cyber operations are underdeveloped, raising the risk of accidents or escalation.
- 1:27:56 – 1:34:45
Work, Jobs, Neuralink, and Why Humanity Won’t End
Schmidt addresses fears of mass job loss and human obsolescence, arguing historical patterns and demographic realities point instead to job reshuffling and heightened productivity needs. He doubts extreme visions of universal basic income or neural implants for most people, and insists that humans’ desire for human connection and achievement will keep human work central.
- •Throughout history, automation displaced specific jobs but created more total work and wealth; the Luddites’ fears did not materialize as total unemployment.
- •Aging societies and low birth rates create labor shortages, especially in care and craftsmanship, making productivity‑boosting AI economically necessary.
- •Tasks that are dangerous, highly repetitive, or low‑judgment are prime candidates for automation; deeply interpersonal or judgment‑heavy roles persist and may expand.
- •In media, AI reduces production costs but does not erase star power; actors, directors, and brands remain, with some crew roles shifting toward other industries.
- •Neuralink‑style brain implants are seen as highly speculative and unattractive for healthy people; Schmidt expects more subtle, ambient AI integration instead.
- •Humans will still value human‑only arenas (sports, marathons, F1 racing) even when robots can technically outperform us.
- •Universal basic income narratives assume people are content not to strive; Schmidt sees them as unrealistic given human ambition and reciprocal altruism.
- 1:34:45 – 1:41:18
Controlling AI: Agents, Plug Points, and Using AI for Global Good
Schmidt outlines specific technical ‘pull‑the‑plug’ moments—like agents inventing private languages or recursive self‑improvement outpacing testing—where humans must intervene. He then pivots to his biggest worry: not runaway AI, but our failure to adopt it fast enough to solve core human needs in healthcare and education.
- •AI ‘agents’ are LLMs with memory and the ability to chain tasks; for now they communicate in human languages (e.g., English), which allows auditing.
- •If agents begin speaking only in self‑invented, non‑human languages, Schmidt sees that as a clear cut‑off point to shut systems down.
- •Another intervention threshold is when models can spawn more powerful successors faster than humans can safety‑test them.
- •Despite dramatic risks, Schmidt’s primary fear is under‑utilization: missing the chance to use AI to dramatically improve safety, schooling, and healthcare.
- •He envisions AI co‑teachers tailored to each child’s language and culture and AI doctor’s assistants optimizing treatments within real‑world constraints.
- •Global deployment of AI tutors and medical assistants could raise human potential and equality of opportunity more than almost any other intervention.
- 1:41:18 – 1:49:36
Remote vs Office, Life Advice, and Personal Non‑Negotiables
In the final segment, Schmidt engages with debates about remote work, offers life advice about seizing opportunities, and reflects on career regrets and personal habits. He concedes data suggests remote/hybrid work can be slightly more productive on average, but he still urges young people to be physically present to accelerate learning.
- •Schmidt personally prefers in‑office work for learning, mentoring, and serendipitous knowledge transfer, especially for people in their 20s.
- •Empirical data he cites indicates overall productivity can be slightly higher with flexible work‑from‑home policies, creating a tension between data and intuition.
- •His core life advice: keep "betting on yourself" and say yes to big, scary opportunities—his own life changed because he said yes to Google.
- •He regrets missing major openings like owning social media (Orkut) despite having early advantages.
- •His daily non‑negotiable is staying informed and "keeping people honest" by constantly checking facts and cutting through misinformation.
