The Diary of a CEO

Karen Hao: Why 'AGI' is a slogan, not a destination

Through race-or-die narratives, AI leaders extract resources and legitimacy; hidden labor and worsened conditions sit beneath automation gains.

KHKaren HaoguestSBSteven Bartletthost

Mar 26, 2026·2h 9m·Watch on YouTube ↗

📖 CHAPTERS

1. 1

   AI companies are “gaslighting” the public: profits built on fear-based myths

   Karen Hao opens by arguing that much of today’s AI industry is “inhumane” and that major AI leaders profit from promoting a myth: race-or-die urgency (“if we don’t build it first, others will”). She frames the core claim of her reporting: companies intentionally shape public emotion and policy narratives to extract resources, labor, and legitimacy.

2. 2

   Karen Hao’s reporting journey and the scope of her OpenAI investigation

   Hao explains her path from MIT mechanical engineering to tech startup life, then to journalism after seeing profit override mission. She details the scale of her investigation for Empire of AI, emphasizing that understanding AI’s impact requires reporting far beyond Silicon Valley.

3. 3

   Where AI’s story starts: Dartmouth, ‘AGI,’ and the problem of defining intelligence

   Hao traces AI’s origins to the 1956 Dartmouth conference and argues the field began with marketing-laden framing: recreating human intelligence without agreeing on what intelligence is. She contends today’s term “AGI” remains malleable—redefined to fit whichever audience OpenAI is addressing.

4. 4

   Altman, Musk, and the founding narrative: persuasion, alignment, and a power split

   The conversation turns to early OpenAI dynamics: Sam Altman’s 2015 existential-risk rhetoric and how it aligned with Elon Musk’s public warnings. Hao argues this language served strategic recruitment and coalition-building, later contributing to a bitter split when control of the for-profit entity was decided.

5. 5

   Why Sam Altman polarizes insiders: ‘vision of the future’ determines trust

   Hao describes unusual polarization around Altman: insiders either see a generational leader or a manipulative figure. She connects this to competing visions for AI’s trajectory and highlights Dario Amodei’s arc—from OpenAI executive to founding Anthropic after a values and strategy break.

6. 6

   Ilya Sutskever, ‘brains as statistical engines,’ and why mechanism matters

   Hao explains that some AI leaders (e.g., Ilya, Hinton) believe intelligence can emerge from scaling statistical models—an unproven hypothesis contested by other fields. She argues mechanism matters because these beliefs drive massive investments, labor practices, and environmental footprint in pursuit of AGI.

7. 7

   ‘Empires of AI’: data grabs, labor exploitation, and monopolizing knowledge

   Hao introduces “empire” as her central metaphor for the AI industry, arguing it best captures the scale, extraction, and control dynamics. She outlines empire-like behaviors: claiming others’ data/IP, exploiting global labor, and shaping what research gets funded or suppressed.

8. 8

   Controlling the narrative: journalist access, intimidation claims, and OpenAI’s non-cooperation

   Hao details how AI companies use access as leverage over journalists and platforms, and describes alleged intimidation tactics (subpoenas) aimed at critics. She recounts OpenAI’s refusal to participate in her book—despite extensive requests for comment—while Altman publicly endorsed other forthcoming books.

9. 9

   Inside the boardroom: the lead-up to Altman’s firing and the ‘instability’ charge

   Hao reconstructs the events that led to Altman’s removal, citing multiple sources close to the process. The core concern: chaotic, distrustful internal dynamics at a company believed to be building world-transforming technology—plus governance red flags like the OpenAI Startup Fund structure.

10. 10

    Why the firing backfired: stakeholders blindsided and employees revolt to reinstate Altman

    The conversation explains how the board’s rapid, secretive move created backlash among employees and partners. Microsoft was informed just before the decision, and the lack of stakeholder alignment catalyzed a campaign to bring Altman back within days.

11. 11

    The ‘race with China’ argument challenged: intelligence, ‘jagged frontier,’ and profit-driven capability choices

    Steven plays devil’s advocate on national-security acceleration, while Hao disputes key premises: that scaling equals general intelligence and that it inherently yields military supremacy. She argues capability improvements are selective and guided by revenue opportunities, not some automatic march toward omnipotence.

12. 12

    Robots, surgeons, and self-driving cars: what’s plausible vs what’s hype

    They debate high-profile predictions (AI surgeons, full autonomy) and what current systems can and can’t do. Hao emphasizes probabilistic failure, retraining needs across environments, and social/legal barriers—arguing timelines are consistently overstated even when progress is real in constrained contexts.

13. 13

    Jobs and the hollowed-out ladder: Klarna, entry-level collapse, and the rise of data annotation work

    Hao agrees job impacts are real but stresses they’re driven by both model capabilities and executive decisions (including layoffs based on “good enough” automation). They discuss Klarna’s headcount reduction and the broader pattern: entry-level jobs shrink, new roles appear at the top and bottom, and the career ladder erodes.

14. 14

    Hidden human costs: data annotation precarity, community harms, and AI’s environmental footprint

    Hao contrasts ‘AI makes work more human’ for leadership with the reality for many workers doing precarious annotation labor under high anxiety and surveillance-like dynamics. She also details data-center impacts—power draw, water competition, air pollution—highlighting disproportionate burdens on vulnerable communities.

15. 15

    What to do now: slow the ‘flawless adoption’ and build ‘bicycles of AI’ instead of rockets

    Hao argues the goal isn’t banning AI but dismantling empire dynamics—forcing fair value exchange, democratic oversight, and responsible deployment. She advocates contestation (local action, lawsuits, procurement and policy choices) and a shift to efficient, targeted AI systems that deliver benefits without runaway extraction.