AI-Wiki

Stanford AI Club: Alfred Lin on Building in the AI Era

11 min read

Lin (Stanford AI Club) — building in the AI era

▶ Watch on YouTube · Stanford AI Club · 56:49

This talk was recorded as the first event of the Stanford AI Club’s 2026–2027 Speaker Series, facilitated by Stanford AI Club Leadership Jason Zhang + Tanvir Bhathal.

For over 15 years at Sequoia Capital, Alfred Lin has partnered with founders from idea to IPO and beyond, including Airbnb, DoorDash, Citadel Securities, Kalshi, and Zipline. In this conversation, he covers why AI is the biggest megatrend of his career and why technology still diffuses slower than it looks, where value is accruing in the application layer, the “fire and ice” tension between passion and discipline inside great companies, how to tell a feature from an enduring business, whether circular money in AI is a real concern, and what ambitious students should optimize for at the start of their careers.

TL;DR

A 57-minute fireside conversation with Alfred Lin, partner at Sequoia Capital, opening the Stanford AI Club’s 2026–27 speaker series; published 15 October 2026. Two student moderators, then audience Q&A. Lin was chairman, COO and CFO of Zappos through its sale to Amazon, and earlier worked at LinkExchange and Tellme, both sold to Microsoft. Most of the talk is career and founder advice; five parts bear on AI and organisations:

  1. The diffusion gap. “What’s the slowest thing to move in technology? The humans that work with it.” He once predicted Amazon would kill Walmart; thirty years later Walmart is “30 times larger than it was in 1997.” The gap between what models can do and what people use them for “is going to always be a large gap.”
  2. Value in the application layer. “All this whole notion that software is dead is kind of BS,” because the application layer “is the gap between what the models are capable of and human beings using it.”
  3. Owning intelligence. “We’re not just going to let two or three models win. That’s not how economies work.” A pharmaceutical company will “take open weight models and program it on your own data.”
  4. Circular money is normal; leverage is the risk. NVIDIA investing in customers who buy its GPUs is like a bank lending deposits. “The bigger concern actually is just the amount of debt companies are taking on,” by firms that used to be cash-flow positive and now borrow to build data centres.
  5. Old priors. Companies were organised for “a lot of people to write software,” and AI organisations will be “much more decentralized.” He doubts the lean startup ever worked and warns against “simple narratives.”

Key claims

Students, risk and the problems worth solving

His parents, both commercial bankers, ranked careers from professor down to business person, and he “chose to do the exact opposite.” Students are “in a privileged situation to take risk,” especially “in an environment where we have scaling laws and the biggest megatrend that we’ve seen since the internet.” His thread through LinkExchange (sold to Microsoft for $265 million), Tellme (automating 800 numbers, building its own servers “because there wasn’t a concept of a cloud”) and Zappos: “Every problem has a solution. Every solution creates new problems.” Early in a career, “optimize for maximum learning” and for “working with really good people.” Great companies take ten years or more; Jensen Huang’s “first 20 years does not look like the last 10 years.”

Is being technical still worth it?

Sequoia’s habit is to hold “two opposite ideas in tension.” “Yes, math and coding is being commoditized, but the understanding of how it works has never been more valuable,” visible in what people with deep model knowledge are paid, and “the people who understand that really well are no longer coding.” He tells technical people “to also understand humans. Humanities is actually very important,” because “we build for 8 billion people… it all flows through people.”

Feature or company

With two people and an idea “there’s nothing defensible,” but large companies “are distracted… They’re trying to spend as much of their free cash flow to build data centers right now.” The test is whether a founder can tell the vision, not only the technical story. On Airbnb, he recalls how fragile it looked: a listing service, a stay with strangers, replies after 48 hours. His father, who travelled for his bank, always booked the same hotel room and knew where the hairdryer was: “he wanted zero variance.” For Brian Chesky “that’s a bug because that’s not living locally.” “Someone’s feature is another person’s bug,” and in a world where much is commoditised by AI, “the thing that doesn’t go away is human connection.”

The biggest megatrend, and why it still diffuses slowly

“Every megatrend has been the biggest megatrend of the time.” What sets AI apart: earlier revolutions moved muscle power into machines, and recent ones (internet, mobile) were communications megatrends; this is a “computational megatrend,” and it “is connected to 8 billion people,” which is why it moves faster.

Fast is still slow. He dropped out of a Stanford PhD because the internet was happening and predicted Amazon would kill Walmart. “Amazon last year was the first year that they had more revenue than Walmart.” The reason is “the gap between what is possible with the technology and the diffusion to regular human beings.” Technology diffuses by building ever higher abstractions away from the machine: his father programmed in machine code, students today will not. “One of the most underestimated things” is what a technologist must do to make a technology diffuse through the population.

Where the value is

“The models are clearly more powerful than we are using it for.” A partner (probably Pat Grady) “loves to say there is value in the application layer.” Sequoia’s examples: Harvey in law, OpenEvidence in medicine, Profound in marketing, Rogo in finance. Users on the other side “are not going to go all the way down and understand how the model works.”

Then owning intelligence: “How do you own your own intelligence for a pharmaceutical company? You’re going to take open weight models and program it on your own data.” Economies do not let a few suppliers win: “We don’t like that concentration of power… we don’t want to put all our data there.” Hence drug-discovery models (Chai Discovery, Isomorphic), developer tools (LangChain), open-weight model builders (Reflection). “The diffusion gap is only getting bigger,” so there is “just a lot of work to do.”

Asked whether application companies will become the service (Harvey becoming the lawyer), he notes Harvey started on frontier models, now trains its own “because they wanted to own their own intelligence,” and is moving from top law firms to corporates: “think much more ambitiously than before.”

Fire and ice

Fire is passion and vision; ice is “the cold hard discipline of management,” which “allows you to… have more fire.” At Zappos (shoes, then handbags, then clothes) Tony Hsieh was the fire and Lin the ice. Amazon did the same at larger scale: books made profitable, then music, then the next category, until it became “the everything store.” DoorDash’s Tony Xu holds both in one person. “You should know yourself very clearly on which side you land.”

Circular money in AI

“If money is only circular, it would be a problem. But don’t forget money has always been circular.” He uses bank lending as the analogy (deposits lent on, with a reserve held back) and would only worry “unless there’s no reserve requirement, meaning all the money that comes in goes exactly out.” NVIDIA invests because it wants customers on its GPUs and “to empower the ecosystem.” His concern lies elsewhere: “companies that used to be nothing but cash flow positive and now they’re cash flow negative and they’re taking on debt to build more data centers… leverage is very powerful. It can make you or break you.”

Audience Q&A: priors, check sizes, regulated markets

  • Conviction is not passion. It means having “actually investigated the problem.” Tony Xu knew restaurants from his mother’s work and saw that “kitchen capacity was always greater than the front of restaurant capacity.”
  • Investor backgrounds. “Early stage investing is about founder quality and growth stage investing is about business quality.” Banking and private equity teach the second.
  • Flawed priors. Earlier generations “built companies that required a lot of people to write software,” and organised around that; “management systems need to change for the new technologies,” and for AI “it’s going to be much more decentralized… this notion that one person… can build a product from beginning to end and launch it and generate revenue, we’ve never seen before.” On frameworks: seed rounds have only grown during his years at Sequoia, “so I don’t even understand whether lean startup ever worked” (its author, Eric Ries, is in the wiki at Lenny’s Podcast). “Beware of simple narratives and comps.”
  • Bigger outcomes. When he joined Sequoia the largest company was worth $400 billion; NVIDIA is now around $5–6 trillion, “15x in 15 years.” For founders, “the distance you have to travel is 15x,” in the same lifetime.
  • Regulated markets. A startup “shouldn’t try to break oligopolies” but work around them. OpenEvidence “didn’t break anything”: doctors want evidence-based answers, so it partnered for the published literature. DoorDash served small and mid-sized restaurants before large chains.
  • An AI-native hedge fund. Machine-learning pricing is already common, so price what incumbents ignore: “sometimes… the market is more valuable than the technology.”
  • The first 90 days. Find the people first. DoorDash’s four founders (business, marketing, product, engineering) and Airbnb’s three worked together before the problem “kind of just falls out of that.” His father on his Harvard friends: “they have too many options.”
  • On himself. “I’m not that good… look at the number of companies we invest in and how many times we’re wrong.” What he has is a tolerance for risk and for being an outsider: “there’s mistakes that you learn from and there’s mistakes you don’t learn from.”

Dynamic-capabilities reading

  • strategic-renewal/business-model — the move from selling tools to delivering the service (Harvey and law firms), and value captured in the application layer rather than the model layer.
  • digital-seizing/balancing-digital-portfolios — the choice between frontier models and open-weight models trained on a firm’s own data, framed as avoiding dependence on “two or three models.”

How it connects

What was actually ingested

The full talk and Q&A (56:49, length_seconds: 3409), from auto-generated captions only. Names are corrected in quotes (Kalshi, Zappos, Jensen Huang, Tony Xu, Tony Hsieh, Brian Chesky, Isomorphic, Claude); “Pat” and “Konstantine” are probable identifications of Sequoia partners named by first name only. Two short exchanges were unclear in the captions and are left out. No stills: a fireside conversation.

Linked entities and concepts

Dangling (single-source mention, deferred): Alfred Lin; Tony Xu; Harvey; OpenEvidence; Zappos.

Scope and reliability

A venture investor speaking to students, with his own portfolio as the examples; the application-layer and own-your-intelligence arguments are also Sequoia’s investment thesis. Figures are from memory and approximate: Walmart “30 times larger,” NVIDIA “5 trillion something, today might be 6 trillion,” and a bank reserve of 20% used as an illustration rather than as the actual US requirement. The Amazon–Walmart comparison and the diffusion argument are the most transferable parts; the circular-money answer is a view, not an analysis of AI financing.

Relationships

Supports

Graph View

Backlinks