--- title: "What can creators learn from Empire of AI?" description: "Karen Hao's Empire of AI in 7 frameworks with steps, and the three Tom Sharman lessons it backs." canonical: https://tomsharman.com/bookshelf/empire-of-ai/ updated: 2026-11-03 --- # Empire of AI: Dreams and Nightmares in Sam Altman's OpenAI > Karen Hao's Empire of AI in 7 frameworks with steps, and the three Tom Sharman lessons it backs. The frameworks here are Karen Hao's, from Empire of AI. Claude read the book in Tom's Brain and linked it to his lessons. ## How Tom uses this - [A brand that compounds has a mission and an enemy](https://tomsharman.com/lessons/a-brand-that-compounds-has-a-mission-and-an-enemy/index.md): OpenAI shows the mechanism at industrial scale, with a grand mission and a named rival (Google, DeepMind, later China) pulling in talent, capital and public goodwill. (pp. 34, 62 and 413) - [Move the people you have before you hire](https://tomsharman.com/lessons/move-the-people-you-have-before-you-hire/index.md): Altman warned in his own memo that fast headcount growth would dilute talent density and mission fit, and the post-ChatGPT hiring surge produced exactly that culture damage. (pp. 272 and 273) - [The brand is the argument](https://tomsharman.com/lessons/the-brand-is-the-argument/index.md): OpenAI's name, nonprofit status and 'open' claims were contradicted by how it behaved, so the claim did all the work until the gap became the story. (pp. 23, 24 and 97) - Author: Karen Hao - Pages: 621 - ISBN: 9780593657508 - Chapters: 15 - Frameworks: 7 - Categories: AI, Money and finance, Content and packaging - Book 13 of 46 on the bookshelf: https://tomsharman.com/bookshelf/index.md ## What it argues Karen Hao argues that OpenAI began as a sincere nonprofit but became an empire, because its mission to build AGI for the benefit of humanity is vague enough to justify centralising talent, capital, data, labour, land and water. Scale was a choice made by a few people in Silicon Valley and was never inevitable, and Sutskever's faith in deep learning, Altman's gift for storytelling and deal-making and Microsoft's money turned it into the industry's doctrine. The costs fall on data labellers in Kenya and Venezuela, on communities facing data centres in Chile and Uruguay, and on artists and writers whose work became training data, while the benefits accrue upward. The 2023 board crisis and the 2024 Omnicrisis show how a handful of fallible people, with no real checks, steer a technology that affects everyone. Hao closes with alternatives, small consent-based models such as Te Hiku's, and with a call to redistribute power over knowledge, resources and influence. ## Worth keeping - Mission as a formula for power: talent, capital and resources get centralised behind a vague goal that can be reinterpreted at will (p. 413). - Altman's pattern, as reported by Murati, Sutskever and Amodei, of telling each person what they want to hear and avoiding putting things in writing, which ended in a board crisis (pp. 358, 362, 375). - Storytelling and networks as the real assets: Altman's fundraising skill, one-word intros and small bets at volume built an empire (pp. 43, 50). - Compute use in AI breakthroughs doubled every 3.4 months, and Hao calls scaling laws a self-fulfilling prophecy that no law of nature requires (pp. 71, 127). - Generative AI depends on hidden human labour, with moderators paid $1.46 to $3.74 an hour and workers blacklisted when a platform decides they are no longer useful (pp. 203, 233). - Survey of 2,500 workers: 96 percent of C-suite leaders expected generative AI to boost productivity, while 77 percent of employee users said it added to their workload (p. 28). - Altman's own warning that growing OpenAI too fast would dilute talent density and mission alignment, which then happened after ChatGPT (pp. 272, 273). - The employee letter showed founder dependence: over 700 of roughly 770 staff threatened to follow Altman, and Murati's equivocation helped tip the outcome (pp. 20, 379, 380). - The GPT-2 'too dangerous to release' episode doubled as publicity and policy positioning, a pattern to watch in any AI product claim (pp. 137, 138, 139). - Te Hiku Media's alternative: consent, reciprocity and sovereignty, a small task-specific model, 310 hours of data and two GPUs (pp. 423, 424, 425). ## The questions it answers ### What does AI actually change about the shape of a content team? The book documents the hidden annotation and moderation labour behind chatbots and a survey in which 77 percent of employee users said AI tools added to their workload. ### How do you build a personal brand that generates inbound rather than requiring outbound? Altman's rise shows a story people want to be part of plus a dense network of small favours and one-word introductions generating inbound deals, along with the cost of drifting from the truth. ### Which of my stories travel, and which only feel like they should? Ralston describes Altman telling 'a tale that you want to be part of, that is compelling, and that seems real', which is a reference for what makes a story travel and for how it can mislead. ### What is the smallest team that can produce at the volume I am used to? Altman's 2020 memo argued OpenAI could fulfil its mission with a surprisingly small number of great people, then headcount grew past the cap and culture suffered. ## The frameworks 1. [Be number one in four categories](https://tomsharman.com/bookshelf/empire-of-ai/be-number-one-in-four-categories/index.md): Altman's late 2019 vision memo setting what OpenAI must lead on by the end of 2020 2. [The data-work crisis playbook](https://tomsharman.com/bookshelf/empire-of-ai/the-data-work-crisis-playbook/index.md): How Scale AI and similar firms built a cheap, disposable global labour pool 3. [Big in scale, tractable, unfairly neglected](https://tomsharman.com/bookshelf/empire-of-ai/big-in-scale-tractable-unfairly-neglected/index.md): Effective altruism's test for picking the highest-priority problems 4. [Altman's three policy recommendations to Congress](https://tomsharman.com/bookshelf/empire-of-ai/altmans-three-policy-recommendations-to-congress/index.md): What Altman proposed in May 2023 to shift attention from present harms to future systems 5. [The formula for empire](https://tomsharman.com/bookshelf/empire-of-ai/the-formula-for-empire/index.md): Hao's three-ingredient account of how OpenAI's mission consolidates power 6. [The formula for dissolving empire](https://tomsharman.com/bookshelf/empire-of-ai/the-formula-for-dissolving-empire/index.md): Hao's recommendations to redistribute power along three axes 7. [Consent, reciprocity and sovereignty](https://tomsharman.com/bookshelf/empire-of-ai/consent-reciprocity-and-sovereignty/index.md): Te Hiku Media's three guarantees for community-led AI development ## Working with Tom on this Tom works with creators and founders one to one, from a single session to a fractional role. [Work with Tom](https://tomsharman.com/consulting/)