Tobi Lütke – Building Shopify and the Future of AI | Ep. 50

Key points

• **The founder slot as infrastructure + credibility-token bank.** Tobi's most repeatable operating insight: 'People are massively underestimating what you can do when the founder is still present. It's not about the individual — it's about the slot of having the founder slot filled. As a founder you get so much social credit for having started the company that it's a bank. Every time someone onboards, they hear how the company was created — that deposits credibility tokens into a virtual bank account. You spend it on big important changes.' **Direct architectural complement to [Eric Ries's Long-Term Benefit Trust thesis from Issue 05](/issues/2026-05-17)** — Ries codifies the structural side (LTBT, PBC); Tobi describes the *operating* side (credibility-token cash-in). Both are the moat under AI founder mode.
• **Net Impact reviews disclosure.** Shopify formally evaluates two engineers — equally good 15 minutes ago, one fully on the AI train, one not — and writes the impact differential in the performance doc. *'It feels incredibly unkind not to tell people one of them has an exoskeleton on and not point it out. So we did. And I included a bunch of other things that are true.'* The internal Vault wiki shows each employee's token usage and percentile in their department. **They had a leaderboard at one point — 'led to immediate really bad effects' — so they removed it.** Companion to [Krishna Rao's '90% of Anthropic code by Claude' disclosure from Issue 05](/issues/2026-05-17) and [Andrew Feldman's '$25-30K/engineer/month' from this week's no-priors](/issues/2026-05-24) — three operator confirmations in three weeks.
• **10x tokens/year + 3x GPUs = 'lines not converging anywhere good for price savings.'** Tobi on the upward sloping curve: 'We are 10xing the amount of tokens we want every year right now. We are 3xing GPUs we are putting into the world. Those lines are not converging anywhere good at price savings.' His framing on customers spending 'tens of percentage points of revenue on tokens': 'I'm grateful Shopify is at the stage where it's a no-brainer — we like the tokens we're buying.' Implicit risk-flag: 'We can't spend 70% of revenue on AI tokens forever and have valuable companies.' **Cross-reference to [Goldman 24x / Lemkin 250x agentic-token forecasts from Issue 05](/issues/2026-05-17)** — Tobi is the operator data-point for both.
• **Small team (3-4-5 people) as the new dominant molecule.** *'The small team is my bet. It's now possible because the agentic harness around our teams is routing really good summarisations of what customers say automatically — back to the team. So that's now available to everyone and everyone can do more skill. Everyone is a 7/10 on every skill now.'* **Direct continuation of [Chesky's Project Hawaii 10-12 person Navy SEAL teams from Issue 04](/issues/2026-05-10)** — Tobi's bet is smaller (3-4-5) and is built around the agentic context-routing layer rather than just headcount discipline. The team-size pendulum is still swinging down.
• **6-week review cycle is now too slow.** *'I run the company by the 6-week review cycle to set a pace ceiling. Sigba 6-6 cycle was faster than what we instituted. Because if you don't do it, you're run by a quarter — and the moment you see H1/H2 in a PowerPoint, you're fucked.'* But: *'I actually think 6-week-review is way too limiting now. We can do so much more. We're trying to figure out what is replacing it.'* **The pace ceiling is being broken by AI productivity.** This is the operating-side echo of [Cat Wu's Anthropic '6-month → 1-month → 1-day' velocity disclosure from Issue 02](/issues/2026-04-26). The 6-week cycle was state-of-the-art operating cadence 24 months ago. It's now legacy.
• **'Markets are extremely good — they'll figure out the correct clearing price for tokens.'** Tobi's deep-felt economic optimism on the AI bubble question: 'There's few providers and maybe there'll be more in the future. There's all sorts of interesting moves around distillation. Companies will know how to wield these tools within their budgets. We are doing a good deal of this ourselves. But we're still charging ahead because frankly we like the tokens we're buying.' **The exact same Jevons-paradox + market-clearing frame as Krishna Rao's Opus 4.5 pricing disclosure from Issue 05** — but expressed by the customer side, not the supplier side.
• **The customer side's anti-doom data: 'you techies fix computers.'** Tobi's strongest counter to the AI-permanent-underclass narrative: *'Our customers cannot reproduce the doomer conversation anywhere. What we hear is: hey, you guys fix computers. You techies talked about computers being incredible things that can do anything. And now we can talk to it and it just does the thing. It just works with me. I've expanded my business and hired all these people.'* **Every 36 seconds someone makes their first sale on Shopify.** Companion to [Helberg's 'America as global underdog' frame from Issue 05](/issues/2026-05-17) — both are anti-decline, anti-doomer, but the data sets are different (Tobi: small-business creation; Helberg: AI as 1/3 of US GDP growth).
• **'Build me a million-dollar business' prompt path.** Tobi: 'I think we're actually getting there. You can use Shopify without products — we help you find manufacturers via Collective. People should have a product the world wants. AI should then do absolutely everything else. We are looking at humanoid robotics. Making the products will become much more tractable. Build-me-a-house is 10 years out.' **The 'prompt builds business' framing is the consumer parallel to [Andrew Feldman's 'Netflix-becomes-a-studio' frame from this week's no-priors](/issues/2026-05-24)** — fast AI doesn't make existing models better; it unlocks new ones.
• **Underhyped: AI deployment in companies. Overhyped: forcing everything into the programming domain.** Tobi's calibrated framing: *'Where it's underhyped is just deployment in companies. No one's using it enough. Where it's overhyped is having to work so hard on bringing things into the programming domain — all of this is just going to get much more natural and easier as models appear.'* Most useful Lab vs Customer reframe: *'I don't even think being in a lab is actually the best position. It's being like using everything and paying attention to how everyone else uses what the labs release — usually on X.'* **Practical implication: the highest-ROI org investment in 2026-27 is internal deployment maturity, not lab access.**
• **Why public + the IPO defence.** *'Going public induces a diligence, a data-drivenness, a set of responsibilities I think are worth having because you're responsible for thousands of people's jobs.'* Shopify IPO'd at $1.5B; today ~$150B = ~100x for retail. *'It's remarkable — I can meet people anywhere and they tell me they bought shares and it was really important to them.'* Counter-frame to the [Andrew Feldman 'corporate adolescence to corporate adulthood' framing this week](/issues/2026-05-24) — same destination, different language for why public matters.

Notable quotes

Themes

• Founder slot as infrastructure — credibility-token bank cash-in for change-management
• Net Impact reviews — Shopify formally writes the AI-with vs AI-without engineer differential
• 10x tokens/year + 3x GPUs = no Jevons-paradox-savings convergence in sight
• Small team (3-4-5 people) replaces EPD triad as dominant org molecule
• Customer-side anti-doom data: every 36 seconds someone makes first Shopify sale