Mental Models That Change How You Think | Bill Gurley

KEYPOINTS

• 01

  Think in systems: complex systems are multivariable and nonlinear, so single-metric optimisation backfires later.

  Gurley sits on the Santa Fe Institute board and recommends Thinking in Systems. Weather and markets can behave one way for years until one variable flips. Changes have first-, second- and third-derivative effects, so don’t get deterministic about one metric—know what actually matters most.

• 02

  Know the bedrock of your field: deep history is rare and wildly differentiating.

  In a skim-the-summary world, few people study their field’s masters outside literature. Imagine 20 candidates interviewing at P&G or Pepsi and only you can discuss the great marketers. A friend advises college applicants to write about the founders of physics if they want to study physics—it signals passion. If learning the history feels tedious, ‘you’re not in the right lane.’

• 03

  Pair the oldest knowledge with the newest edge.

  Great founders share obsessive learning: the edge where disruption happens keeps moving—mobile once, AI now—and they read everything to be top 1%. Gurley keeps about five premium AI accounts so he doesn’t miss anything. The power combination for a young person: know the legends of marketing and really understand TikTok.

• 04

  Value investing and venture can coexist: ‘value’ means price below future worth.

  He started at First Boston, raised on Peter Lynch, Malkiel, Buffett, Graham and Howard Marks. Via friend Michael Mauboussin he met Bill Miller, who beat the S&P for 15 years while calling himself a value investor and holding Amazon for its network effects. Gurley sees Wall Street as the eventual buyer of what VCs create, so even at ‘two people in a PowerPoint’ you ask whether public investors will love it when it grows up.

• 05

  China’s open-source AI ecosystem may innovate faster than the West’s closed one.

  China has around ten strong open-source models that learn from and test one another, often publishing their techniques. His metaphor: two farming societies, one where farmers only sell at market, one where they must share best practices—which evolves faster? Quietly, many Silicon Valley startups use Chinese models by volume. Some Western players ‘are begging for regulation’ because it would raise the bar against them.

• 06

  Whether general models eat vertical AI apps is genuinely open.

  If models become near-sentient, verticals vanish. Gurley leans the other way: legal AI startups that ingest all the case law and embed in workflows build data moats that are hard to switch out—though Microsoft’s climb from OS to apps shows it can happen. Coding is the largest vertical, and Cursor already lets users swap models; price optimisation will make swapping more common.

• 07

  Model progress may hit an asymptote as data runs out.

  Labs are ‘painting in the corners,’ paying experts thousands of dollars an hour to fine-tune. Yann LeCun argues language-based systems have limits. AlphaGo’s surprising move is cited as proof of machine creativity, but it came in a constrained game; the real world has infinite paths. Tesla FSD is likewise relatively constrained—Gurley wouldn’t yet ride in the back seat because corner cases are unfathomable.

• 08

  The AI buildout is bigger than anyone imagined, and circular deals both raise and delay correction risk.

  Five years ago he wouldn’t have believed giants would take free cash flow of $50–100B a year toward zero for capex. Investors now believe in power laws, so they’re more risk-seeking: Amazon lost $2–3B before turning cash-flow positive, Uber about $15B, and AI leaders will lose far more. Cloud providers funding model makers who spend it back inflates growth. He views burn rate as a risk measure: a $1M monthly burn was risky a decade ago; now $100M+ a month.

• 09

  U.S. payments are a product of regulatory capture, and stablecoins may route around it.

  The UK built Faster Payments ~20 years ago; Brazil’s Pix hit 60–70% of transactions within six years; China’s WeChat Pay and Alipay let you pay at your table via QR code. The U.S. still has three-day ACH and $25 wires, and FedNow has stalled, protecting 2–2.5% card fees and Visa/Mastercard’s ~60% operating margins. Coinbase’s USDC pays around 4% and moves in seconds for pennies.

• 10

  IPOs are rigged for bankers; an anonymous auction is the obvious design.

  A freshman computer-science and finance student would match supply and demand anonymously, not let bankers pick the price and their favourite clients. Direct listings used an auction, but Wall Street retreated to its ‘controlled oligopoly.’ Tokenization could disrupt allocation, though unregulated tokens invite speculation and manipulation.

• 11

  Writing is a thinking tool and a magnet for deal flow.

  Gurley read little until business school, then went from business books to personal development to biographies to narrative non-fiction (Gladwell, Lewis, Krakauer). He codified marketplace investing in writing, like Bezos’s six-page memos, to catch loose ends—and founders who read it reached out. Storytelling and product instinct are founder superpowers; he estimates fewer than 5% of non-product people can be made good at product.