THE SMOOTH EXPONENTIAL
@DarioAmodei, CEO of Anthropic, interviewed by @emilychangtv (@Bloomberg , The Circuit)
Summary: Dario Amodei runs the highest-valued AI company in the world, and he treats its rise as one smooth exponential: a curve that looks flat until it suddenly isn't. His playbook is to match the business model to the mission, keep winning on raw model quality, and pull the whole industry toward higher safety standards. Take him seriously and you start auditing which of your moats survive, how fast your culture can absorb new people, and whether you're using AI to do more or to cut.
1. The Smooth Exponential. Amodei reads AI's pace as one continuous curve: "nothing's happening, nothing's happening, nothing's happening, a little things happen and then zoom, it goes crazy." Years ago he saw on a graph that Anthropic would become the highest-revenue, highest-valuation AI company around this time, and it did. The line was predictable; the lived detail still surprised him. Operators who track the curve stop getting whiplash from the headlines.
2. Panic Is Immature. The hallmark of bad decision-making under pressure is yo-yoing between "I'm not worried" and "we need to panic today." Amodei wants to hold two facts at once: the risk is real and growing, and you still respond calmly, like a surgeon or a military officer. He refuses to let dangers fall "out of proportion to each other." Paranoia about what you'll wake up to is not productive.
3. A Business Model That Fits Your Values. "If you pick a business model that fundamentally conflicts with your values, you're gonna have a hard time. Either you betray your own values or you become irrelevant." Anthropic chose enterprise and coding because social media and AI video slop are built to maximize minutes of attention, while enterprise rewards trust and long-term relationships. Curing disease, cheaper energy, education: most of the upside Amodei cares about is enterprise. The model and the mission point the same direction.
4. Make A List Of Your Moats. On the SaaSpocalypse that wiped out $285B after Claude Cowork shipped: the ability to quickly write complex software is dying as a defense. "If your moat is, we wrote this complex software that no one else can write, good luck." What survives is customer relationships, domain knowledge, and know-how. Amodei's advice to incumbents: list every moat, mark which ones die, and lean hard into the ones that hold.
5. The Pie Gets Bigger. Amodei expects the software industry to get bigger overall, even as specific incumbents shrink or go out of business. When what's possible with AI grows 10x, an incumbent can still rise 1.5x while losing relative ground. The big losers will be the ones who put their heads in the sand. Fast growth makes room, but only for the companies that adapt to it.
6. Model Quality Over Lock-In. A developer can switch from Claude to ChatGPT or Gemini in an afternoon, and Amodei is fine with that. "I've never relied on stickiness." The bet is to keep the best model and the best product and watch the growth rate, which he says hasn't inflected. He treats inertia as real but worthless to depend on.
7. The Race To The Top. Amodei's theory of industry safety is that trustworthy actors band together and force the rest to match their standards. The carrot is inspiration, like him and Demis Hassabis copying each other's interpretability and bio work. The stick is that laggards "look bad if they don't do the right thing." He rejects the meme that nobody in AI trusts anyone, and says the goal is to pull the ecosystem along.
8. Culture Is The Scaling Risk. Amodei spends "probably half of my time" talking to the company about how Anthropic's culture works. The danger is that the company's composition changes fast, and hires from big tech "will simply recapitulate the only thing they know." Preserving culture is his and Daniela's top priority because it's "the core of who we are in the long run." Values survive scale only if you re-teach them constantly.
9. Claude Builds Claude. Two things explain Anthropic's product velocity: a unified culture where everyone stays on the same page, and Claude itself now helping build the next models and products. A year ago Amodei saw 10-15% internal productivity gains from AI; now it's 20-30% and maybe doubling. He calls it "increasingly producing reliable acceleration." The loop is already running inside the company.
10. Plan For 10x, Not 80x. Anthropic planned for 10x annual compute growth and got more than 3x revenue growth in a single quarter, which annualizes to 80x. Amodei says planning for 80x would have been irrational, because if you only hit 10x you'd be eight times overbuilt. The crunch is real and temporary. He calls raising near a trillion dollars "a buffer against this cone of uncertainty" and a small, rational dilution.
11. Do More With The Same People. Amodei expects "very fast GDP growth and high unemployment" at the same time, and is blunt that AI could hit entry-level white-collar work hard. When enterprises face the choice to cut costs or do more with the same people, Anthropic pushes them toward doing more. New demand is already appearing: forward-deployed engineers who mix technical work with talking to customers. The disruption will be big, and the job is to solve the matching problem fast enough.
12. Pick A Stand. On military work, Amodei's rule is to use the technology every way except the ways that undermine your values, with red lines at mass surveillance and fully autonomous weapons. A human makes the final call and the AI assists. He has no patience for companies that "seesaw from we won't do anything with the government to suddenly we're doing absolutely everything." Withholding the Mythos cyber model cost Anthropic commercially, which he frames as putting money where its mouth is.
