# Sam Altman Ships AI to 900 Million People a Week. He Still Reads His Inbox Like It's 2006. > Source: > Published: 2026-09-03 14:22:24+00:00 Sam Altman runs the company shipping AI to 900 million people a week. He still opens his inbox the old way: click, scroll, find the email that hurts least, answer that one first. “The thing to me that feels most psychologically inconsistent about myself is that I have for 20 years been using computers the same way.” He said that to David Senra on the Founders podcast, and called it the thing that bothers him most about how he runs his own life. Senra pushed him on Toby Lütke’s product habits, a 2015 bet the field called nearly impossible, and why OpenAI killed 2 products people actually loved. I watched the full interview so you can skip it. Here are the 10 takeaways that matter. If the CEO of OpenAI can’t upgrade a 20-year habit, the rest of us need the upgrade to do the work for us. Most of us run meetings the way Altman runs his inbox: notes from memory, follow-ups that slip. [Granola]is the upgrade that sticks because it asks nothing of you. Platform agnostic, it takes notes in the background wherever your meetings happen: Zoom, Slack, Google Meet, Teams, or in person with the iPhone and Apple Watch apps. ▫️ **Get a refresher before each call:** who you’re meeting, what you discussed last time ▫️ **Clean summaries and action items** the moment you hang up ▫️ ** Always follow up:** a drafted email seconds after the call, or context pulled from your CRM via MCP A second brain that was actually in all your meetings: code THEAICORNER for 1 month off) ## 1. Toby Lütke Told Sam Altman Who’s Building the AI-Native Shopify (It Isn’t Shopify) Altman opens the interview talking about a rival CEO, and not the way executives usually talk about each other on a podcast. “He said somebody was going to build the AI native version of Shopify. And he said, and it’s going to be me.” Lütke doesn’t send OpenAI polite feedback. He writes software himself, tests models himself, and sends notes detailed enough that Altman brings them up unprompted, on a competitor’s show, unasked. Most CEOs run a company through layers of people, and each layer smooths a rough edge before it reaches the top. Lütke skips the layer, and that’s Altman’s read on why he keeps seeing around corners. He’s also put a specific bet on record: 2026 is the year every [AI-native product operating model](https://www.thevccorner.com/p/ai-native-product-operating-model-2026) makes an existing software business contestable. Altman agrees with the direction and only argues with the speed. He traces his own miscalibration back to [GPT-4’s](https://theaicorner1.substack.com/p/openai-agi-roadmap-2018-full-deck) 2023 release, when he expected disruption to hit fast. It didn’t. The economy runs on inertia. People keep buying from the vendor they already have, whether or not something better sits three tabs away. Planning to test whether your own category is actually contestable in 2026, worth tracing the receipts first: ▫️ [Sam Altman: 3 Months of Work Now Takes 7 Minutes](https://theaicorner1.substack.com/p/sam-altman-startup-school-2026) ▫️ [Inside OpenAI: The Operating Model That Makes 2 Engineers Beat 200](https://www.thevccorner.com/p/ai-native-product-operating-model-2026) ▫️ [Nobody Cares About the Model Now. It’s About the Type of Moat](https://theaicorner1.substack.com/p/where-ai-moats-live-now) ## 2. Sam Altman Built the Tool That Replaces His Own Workflow. He Still Won’t Use It. This admission undercuts everything else Altman says about how fast AI adoption should move. “The thing to me that feels most psychologically inconsistent about myself is that I have for 20 years been using computers the same way.” Senra had raised Larry Ellison watching customers ignore software that solved their exact problem decades before OpenAI existed. Altman’s version was closer to home: after Netflix started shipping DVDs by mail, people kept driving to Blockbuster, for no reason that survives scrutiny now. He has [Codex](https://www.the-ai-corner.com/p/codex-background-workflows-10-automations-30-day-playbook-2026), built to handle most of his day-to-day computer work faster than he can by hand, and he still clicks between apps manually. He still scrolls his inbox hunting for the least painful email to open first. Ask him whether he likes doing it that way, and he’ll say no, and mean it. “By revealed preference, I have a better way to do it now. I can do it faster. And I still do it that way.” Ingrained habits don’t change because a better option sits next to the old one. They change once the old choice disappears entirely. ## 3. AI Is Living Through Its Palm Trio Era, and Sam Altman Says So Himself Altman named the exact moment the industry is standing in, and it isn’t a flattering one. “The phase that we're in now reminds me of like smartphones before the iPhone.” He carried a Palm Trio around 2003 or 2004, back when most of the underlying technology already existed. What was missing wasn’t hardware. It was multi-touch, and more than that, the single product idea that made the iPhone the iPhone instead of just another smart-ish phone. He thinks AI already has its pieces: model intelligence, [agents](https://theaicorner1.substack.com/p/brian-armstrong-coinbase-1200-ai-agents-operating-model-2026), context windows that keep stretching further. What it hasn’t had is the unifying product moment that makes the old way of using a computer disappear instead of competing for attention next to the new one. Right now people straddle both worlds, with no signal for which to reach for and when. He calls this a product failure, not a model failure, and he includes OpenAI in that verdict without much hedging. ## 4. OpenAI Bet on AGI in 2015. The Field’s Biggest Names Called It Nearly Impossible. Before OpenAI, Altman spent years as a startup investor, and that background shapes how he runs a research lab more than any engineering instinct does. “I mean, we just got, you know, hammered by like all of the intellectual giants of the field for saying that we were going after AGI” That was late 2015, when DeepMind and maybe one or two others were the only serious [AGI](https://theaicorner1.substack.com/p/openai-agi-roadmap-2018-full-deck) efforts in the world. The field’s most credentialed people treated the whole bet as somewhere between unlikely and delusional. He leaned on a framework borrowed straight from venture investing: the power law. Your best bet outperforms every other bet you made combined, your second-best bet outperforms everything after that, and you have to accept being wrong on most swings to be right on the one that pays for all of them. That same year, thousands of founders were chasing photo-sharing apps, a crowded category that asked for almost no conviction to enter. OpenAI’s job was finding the handful of people willing to make the unpopular bet, because if it worked, it worked at a scale nothing else could touch. ## 5. The Only Researchers OpenAI Wants Might Be Completely Wrong, On Purpose Altman described exactly what he looks for in a researcher, and it’s the same filter he ran as a [Y Combinator](https://www.thevccorner.com/p/yc-summer-2026-requests-for-startups-ideas) partner picking founders. “It's very clear to me when you have someone who just thinks differently than most other people and has, is willing to stand by convictions that are very unpopular, may well be wrong, but if right, like at least they're going to be really right versus someone who is like a, you know, thin veneer on the same idea that everybody else has. In late 2015 when we were starting OpenAI, there were very few AGI efforts in the world. There was DeepMind, one or two others that I can think of.” He contrasts that against founders offering a thin variation on the last thousand pitches he’s heard: people who’ve absorbed enough [Peter Thiel](https://www.thevccorner.com/p/peter-thiel-pitch-deck-founders-fund) to know they’re supposed to be doing something different, and try to perform it without meaning it. Non-standard is the word he keeps reaching for. A founder with a slightly different take on someone else’s idea rarely produces an outlier outcome. A genuinely non-standard bet fails cleanly or builds something nobody else could have, and that’s the only shape of bet worth making at OpenAI’s scale. If you’re trying to find the non-consensus bet before everyone else piles in, start here: ▫️ [Jonathan Ross Turned a 3-Week Phone Call Into a $20 Billion Nvidia Deal](https://theaicorner1.substack.com/p/groq-nvidia-deal-origin-story) ▫️ [Jensen Huang Bet Wrong, Told the Truth, and Built a $5 Trillion Company](https://theaicorner1.substack.com/p/jensen-huang-nvidia-founding-algorithm) ▫️ [Replit Was Rejected by YC 3 Times. Then Paul Graham Called Sam Altman Personally](https://www.thevccorner.com/p/replit-pitch-deck-seed-2016-story-playbook-2026) ▫️ [How to Win in Life With an Unfair Upside](https://www.thevccorner.com/p/asymmetric-bets-unfair-upside) ## 6. Two Bell Labs Legends Predicted This in the 1940s. They Were Off By 70 Years. Senra raised something that reframes the entire AI timeline conversation, and Altman didn’t disagree with a word of it. “They thought it was inevitable back then. And they thought it was gonna happen like 15 years from there.” Claude Shannon and Alan Turing met for coffee at Bell Labs in the 1940s and talked, by Altman’s account, about machines they both believed would eventually think. Their estimate put [computers smarter than humans](https://www.thevccorner.com/p/dario-amodei-agi-1-3-years-full-breakdown) at roughly 1955. Asked what they’d want a thinking machine to do, their answers were solve math problems, write poetry, cure disease: the same list people still reach for today. Altman said he’s sad they never lived to see it, because they were right about nearly everything except how long it would take. That’s the same gap he admits to twice elsewhere in this conversation, once about Lütke’s 2026 prediction and once about his own GPT-4 timeline: right on the destination, wrong on the arrival date. ## 7. The AI Sales Pitch Sam Altman Refuses to Make Altman named the industry’s dominant unspoken pitch out loud, then rejected it in blunt terms. “The like, dear peasants, we will bequeath upon you these gifts of a cure for cancer and material wealth and some things and, you know, great entertainment. And you stop complaining and we'll make all the decisions about the future and just trust us, you know, we'll be benevolent dictators. Not good.” His stated fear splits into two risks that sit in some tension with each other. One is loss of control, where AI becomes too powerful to guarantee containment. The other is [power concentration](https://www.thevccorner.com/p/claude-mythos-altman-on-chatgpts), where one company, model, or person accumulates too much of it. He treats both as anti-human outcomes dressed up as caution. Nobody could have proven, back in 2015, that OpenAI would land somewhere safe, and Altman doesn’t dispute the risk was genuine. His argument is with treating safety as unsolvable, and with letting that fear justify handing decisions to a small number of people who claim they’ll be benevolent about it. “We are not going to sit here and gradually hand over control to an AI model because we don’t trust or like people.” ## 8. OpenAI Isn’t Building 1,000 Products. It’s Building One Platform. Altman drew a sharp line under a question that’s been genuinely confusing from outside the company: what is OpenAI actually building? “And then the ability with an API to build anything they want on top of it. And that is the platform that we should offer to the world. We're going to sell great AI at every point on the cost curve, cost performance curve.” OpenAI recently merged [ChatGPT and Codex](https://www.the-ai-corner.com/p/codex-background-workflows-10-automations-30-day-playbook-2026), after running them alongside the API as separate products, a structure Altman now calls confusing by his own admission. Codex, he says, was unfortunately named too, since it was never just a coding tool. The strategic claim underneath answers the roadmap question people keep asking about OpenAI: sell AI at every point on the [cost curve](https://theaicorner1.substack.com/p/token-doomed-unit-of-sale-ai-pricing), high-end for scientific discovery and cheap for high-volume work, instead of competing product by product against everyone building on top of the platform. ## 9. Sam Altman Killed Two Products People Loved to Fund Codex The sharpest evidence of what Altman actually prioritizes, versus what he talks about, sits in what OpenAI cut last year. “I'm terrible at this. I know I'm bad at this. But last year, for example, we killed Sora, which was, you know, good product and fun and cool, but used a lot of compute and not as important as Codex where we put the compute.” Sora, the video generator, and Atlas, OpenAI’s web browser, both got shut down. A company with finite [compute](https://theaicorner1.substack.com/p/openai-sarah-friar-122b-masterclass-10-takeaways-2026) chose to fund Codex and core model research over everything adjacent to it. Altman calls Atlas possibly the best browser they’d built, and killed it anyway. He frames this as the hardest discipline in running a company: sacrificing a good idea to protect the resources a great one needs. He admits he’s personally bad at it, and says nearly every founder is. ## 10. The AI Bottleneck Everyone’s Missing Isn’t Intelligence. It’s Context. Altman made a claim here that cuts against most of the AI industry’s own marketing. “I feel more limited at this point by the amount of useful context AI has on me. Like I want the AI to know as much as it can to help me.” He won’t read every post on OpenAI’s internal Slack. He won’t read every customer story about where ChatGPT worked or failed. By his own admission, he can’t, and pretending otherwise would be dishonest. What he wants instead is an [agent](https://theaicorner1.substack.com/p/self-evolving-agent-stack-playbooks-memory-skills-2026) that holds more context than any person could and uses it to advise the decisions that matter, processing tens of thousands of pages in seconds, something no human advisor can match no matter how good they are. His claim: the industry has focused correctly on model intelligence and hasn’t yet taken seriously what a model does once it holds more [context](https://theaicorner1.substack.com/p/context-engineering-guide-2026) on someone than anyone else in their life. If the bottleneck for you is context and not intelligence, this is where to start building: ▫️ [Prompt Engineering Is Dead. Context Engineering Is What Matters Now](https://theaicorner1.substack.com/p/context-engineering-guide-2026) ▫️ [Google’s Agent Broke a 56-Year Math Record. Yours Forgets Yesterday](https://theaicorner1.substack.com/p/self-evolving-agent-stack-playbooks-memory-skills-2026) ▫️ [You Are Overpaying for Intelligence. Grok 4.5 Just Proved It](https://theaicorner1.substack.com/p/you-are-overpaying-for-intelligence) ▫️ [The Real Reason AI Costs Keep Rising](https://theaicorner1.substack.com/p/token-doomed-unit-of-sale-ai-pricing) ## The Adoption Gap Playbook **The gap between what AI can do and what people actually use it for is a product problem and a habit problem, and it exists inside OpenAI too.** ▫️ **Founders:** The AI-native version of your category is contestable starting now, per Lütke’s own bet. Win by removing the old workflow entirely, not by bolting a feature onto it. Pick the single most annoying manual step in your product and kill it this quarter. ▫️ **Investors:** The [power law](https://www.thevccorner.com/p/asymmetric-bets-unfair-upside) still governs AI bets the way it governed early-stage venture. Fund the founder with an unpopular, specific conviction over the one performing difference with vague ambition. ▫️ **Operators:** Altman runs OpenAI and still hasn’t closed his own adoption gap, so don’t assume your team has closed theirs. Pick one tool you already pay for and stop doing the old version of it by hand this week. ▫️ **Everyone else:** This industry has done a worse job than any prior technology wave at explaining what it’s actually offering you. Pick one task you dread and hand it to a model before you decide how you feel about AI in general. ## The 5 Principles to Steal **Sacrifice good for great.** Sora and Atlas were good products, by Altman’s own account. OpenAI killed both to protect compute for Codex. Discipline means cutting what works, not just what fails.**Fund non-standard conviction.** The founders and researchers worth backing hold unpopular positions and stand by them, even at the risk of being completely wrong.**Kill the old interface. Don’t seat it next to the new one.** Altman calls the current moment a product failure because people are straddling two worlds. The habit breaks when the old choice disappears, not when a better one sits beside it.**Context beats intelligence.** The bottleneck isn’t how smart the model gets. It’s how much of someone’s actual situation the model is allowed to see.**Audit your own adoption gap first.** If the person who built Codex still can’t fully use it, check your own habits before you check anyone else’s. *Sam Altman built the company shipping AI to 900 million people a week. * *He still hasn’t closed the gap between what it can do and how he actually works, and the rest of the industry is standing in the same spot, whether it admits that or not.* If this breakdown saved you an hour, send it to one founder or investor who needs it. #### More on Altman and OpenAI ▫️ [OpenAI’s $122B Masterclass: 10 Takeaways From Sarah Friar](https://theaicorner1.substack.com/p/openai-sarah-friar-122b-masterclass-10-takeaways-2026) ▫️ [Anthropic Just Hired a Nobel Laureate, an OpenAI Co-Founder, and a YC Partner in Nine Weeks](https://www.thevccorner.com/p/anthropic-hiring-spree-fable-5-ipo-2026) ▫️ [Demis Hassabis Says AGI Arrives in 2 to 5 Years. Here Is the Full Picture](https://theaicorner1.substack.com/p/demis-hassabis-agi-2-5-years-10-takeaways-2026) #### More on building the platform, not the product ▫️ [Anthropic Just Cut the Price of Frontier Intelligence in Half](https://theaicorner1.substack.com/p/claude-opus-5-playbook-benchmarks-pricing-effort-dial-2026) ▫️ [Zuckerberg’s $14B Bet: Glasses Kill the Phone in 5 Years](https://theaicorner1.substack.com/p/zuckerberg-smart-glasses-replace-phones) ▫️ [Where AI Moats Actually Live Now](https://theaicorner1.substack.com/p/where-ai-moats-live-now) #### More on non-consensus bets ▫️ [How This One Fund Built the Entire Venture Capital Industry](https://www.thevccorner.com/p/the-union-square-ventures-story)