Original episode:https://youtu.be/ZSM2uFnJ5bs?si=PmSUanLOTp2sgprn · Timestamps are clickable — they seek the player in place
This is David Senra (host of the Founders podcast) interviewing Shopify co-founder and 21-year CEO Tobi Lütke. Other founders mention Tobi so often that David calls him "singular." The conversation opens with a first-principles question — what is a company, actually — and follows Tobi through what he's actually done with that question: treating COVID as an opportunity to rebuild the company from zero; turning the entire company structure (titles, levels, compensation) into a machine-readable configuration file using Python and a SAT solver (Shopify OS); using neuroscience to explain why affirmations and public commitments can rewrite identity; and why "differentiation is an obligation, not a choice." It ends on how Tobi now manages a team of AI agents — exactly the way he used to play StarCraft — and the real mission he only discovered in 2014: building things was never the point; sharing the experience of the moment you become an entrepreneur was.
[00:02-03:41] Opening framing: David introduces Tobi as the founder other founders most often call "singular." Tobi's first claim: a company is a social technology. It legally lets you "pursue one thing at full intensity" without being cast out by society — itself an extremely rare institutional invention in human history. A company is, at its core, a counterfactual experiment run on the world you see: "you try to build what you think should exist, then let the market test it — if the market agrees, it sends energy back to you in the form of money, so you can keep doing it." The modern company is barely 500 years old, evolving from quasi-governmental entities (the East India Company) to its current form. If a company were proposed for the first time today, as an institution built from scratch, it would sound insane.
[03:41-09:50] "All companies are terrible, including mine": Tobi explains this is the most pleasant fact he knows — it means you're still improving. His saddest day: opening old code and finding it impressive — meaning he'd stagnated (no longer coding all day, but "running a company" instead). Competitor vs. rival: a competitor (competitor) makes a company reactive — the most active Slack channel becomes the competitive-analysis channel. A rival (rival) inspires you to become your best self — Agassi-Sampras; Michael Jordan admitting in The Last Dance that he may have invented slights to manufacture a rival. Shopify observed the same "high-agency" trait in the hip-hop/rap industry.
[09:50-26:40] COVID and rebuilding the company: after the IPO, Tobi started "cosplaying a serious public-company CEO" — a 60-year-old man in a suit, delegating trust, "okay, you look passionate, I trust you." The result: a team inside the company was quietly building functionality to let Shopify run supermarkets — someone in the Toronto office had unilaterally decided "the supermarket industry is big, we should take 1% of it." When COVID hit, Tobi realized every plan had to go back to zero and be re-derived from scratch. 16-hour-a-day project reviews for a week: he canceled 60% of the projects and replaced every executive within a year. The COVID discovery: in a crisis, the people you'd bet on beforehand often go to zero, and the people you wouldn't bet on go to a hundred — the single most reliable predictor: "have you ever started a company before?" The founders' channel saved Shopify: he asked for help in a Slack channel of founders of acquired companies, took them straight out of "founder daycare" (skunkworks/peripheral projects) and put them in core roles — and every one of them succeeded.
[26:40-40:00] Shopify OS — engineering the company itself: Tobi started with a fresh repository on GitHub and turned the company's structure into machine-readable configuration files — titles, levels, compensation data all encoded, modeled with Python plus a SAT solver. The discovery: an 8,000-person company had 5,500 distinct job titles, and "senior staff" ranked below director on some teams and above it on others. The Desired State System: borrowed from how React works — compute what the state should be, compare it to the current state, find the minimum steps to close the gap. The result: when a department head asks for 50 new sales hires, you can immediately see how many people that takes away from engineering. Politics disappears — it's no longer "the boss promised someone 50 sales reps on the golf course," it's "here's the decision, here are the consequences, do you still want it?" The individual-contributor (IC) track: Shopify added a path for every function outside engineering to be promoted without becoming a manager. The best engineers moved back from management into IC roles.
[40:00-50:30] The brain as a narrative-alignment mechanism: Tobi's takeaway from reading neuroscience: "the brain is a retrospective narrative-alignment mechanism — it tries to take the most salient version of your self-identity, then reconcile your history with it." It rewards behavior consistent with identity and blocks behavior that's dissonant with it. How affirmations work: Tobi wrote "I love public speaking" for ten minutes a day for a week — a week later, he actually did. Not a placebo — an active rewiring of the prefrontal cortex. Messages in a bottle: writing letters to his future self at key moments (with a built-in delay), using spaced repetition to internalize key insights. He requires executives to present once a year at an international conference on what Shopify does "differently and better" in their domain — itself a forced affirmation, since they then can't avoid living up to their own public claim (otherwise they'd be lying to Tobi, and they won't allow themselves to do that).
[50:30-01:01:00] Differentiation is an obligation, not an option: David reads a line from Dyson — "differentiation, and retention of total control. Make it different, even if it's worse." Tobi fully agrees: building your own version from scratch, even a 6/10, is worth more than copying someone else's 7/10 — because you have full control over the first version and can iterate it past 7. The internal podcast "Context": records how every major decision was made and why, averaging 23 minutes per episode (calculated from the average commute time of Ottawa/Toronto employees). The "Tobi said" problem: people inside the company would push their own opinions by saying "Tobi said we should do it this way" — his correction: "When? In what context? We never discussed this."
[01:01:00-01:14:58] Ownership vs. stewardship: once code is merged (a PR accepted), it enters the company's commons. You're not allowed to say "I own this code" — only "I am the steward of this code." The company runs more like an open-source project. This lets Tobi "publicly trash his own past work," even pieces he wrote himself, to free up the team's energy to improve it. David Ogilvy's "divine discontent": Dyson looks at a cup and asks "how could this product be better?", improves it, sets it down, picks it back up, improves it again. Tobi's internal monologue: "I found a problem = I found an obvious blueprint for improving the company" — which excites him rather than discourages him. The corporate-raider game: once a year he plays "I bought this bankrupt company at a fire-sale price, and now I'm going in to clean house" on his own company — listing every problem and generating an action list.
[01:14:58-01:28:40] The power of "for example": five words his co-founder Daniel taught him. When Tobi sees bad code architecture, his instinct is to walk to the whiteboard and tell the engineer "it should be done this way" — which triggers a defensive reaction. Daniel told him: "put 'for example' before every sentence. 'I can think of a few other ways to do this — for example, what about this?'" Now you're on the same side instead of opposing each other. The 2015 IPO decision: against Silicon Valley orthodoxy ("stay private longer"), Tobi chose to go public early, at 800 employees and a roughly $1.5 billion valuation. "We're not going public — we're creating a public version of Shopify." The Formula One rulebook analogy: other industries' rulebooks are read to comply with them; F1's rulebook is read to find loopholes in it. Shopify's IPO roadshow video: they found no rule requiring the video to look like every other IPO's — so they shot a documentary instead. Every IPO since has imitated that approach.
[01:28:40-01:57:00] Talent and office design: Stoicism and Adlerian separation — "your job is not to make people like you, it's to become someone worth liking." Likewise: "your job is not to recruit a better team, it's to become a company worth the best people working for." Shopify deliberately stayed in Ottawa/Toronto — it can be "the single best company" in an eastern tech community instead of competing with Silicon Valley's top salaries. Office design: co-founder Daniel Weinand (a generalist) invented Shopify's "pods" — workspaces comfortable for 5 people but cramped for 7, with no policy ever published, and people naturally organized into 5-person teams. Meeting rooms sit in the middle (away from windows), desks around the perimeter (good window light), the coffee area on the other side, and staircases cut through floors to create incidental encounters. Everything comes from Shopify's merchants — every item in the office either comes from a merchant on the platform, or the team helped a merchant create that category in the first place.
[01:57:00-02:08:00] StarCraft → AI agents → year 21: StarCraft taught Tobi about resource management (not just quantifiable resources, but attention), decisions under incomplete information, and "attacking your opponent's attention is more valuable than attacking their base" — because overloading their attention makes them make mistakes. He never played by the book in StarCraft, often using "moves that aren't in any book" (because they're objectively not optimal) to disrupt prepared opponents — fluid intelligence vs. crystallized intelligence. Managing AI agents now feels exactly the same as StarCraft: six agents running simultaneously, emailing each other ("I find this funny"), Tobi zooming in to micromanage one of them while a critic agent monitors everyone's task status. "If AI hadn't happened, I wouldn't still be running Shopify": his mission is to help more people start companies — if conditions were stable, other CEOs might be better suited than him — but turbulence plus interestingness is the state in which he's most productive. Year 21 is "probably the most interesting year in this company's history."
[02:08:00-02:23:01] The real mission (discovered in 2014): while preparing a talk for an internal summit, Tobi realized: "I loved engineering because as a kid, it was the only way I could build things. What I actually love is building things." The mission: building things to share an experience — the moment of starting a company, the feeling of the first sale. "I build products, then I build companies to build products, and that causes more of this kind of thing to happen." Shopify's 80% data point: a Shopify survey found that 80% of customers have someone in their life who would reply within 24 hours if they sent an entrepreneurial question. Tobi initially thought this showed how lucky his customers were — then realized the causality runs backward: they became Shopify customers because they had that person. David's mission and Tobi's mission are the same: "putting the possibility of entrepreneurship into more people's toolkit."
The company is the single most remarkable social technology ever invented: it legally lets you and thousands of other people pursue a counterfactual at full intensity, and lets the market test whether that counterfactual is correct. If proposed today as a new institution, it would sound insane.
"All companies are terrible" is the best possible news: it means the current state isn't the end point. The saddest moment is opening old code and finding it impressive — that means you stopped progressing.
Differentiation is an obligation, not a choice: "make it different, even if it's worse" (Dyson). Only by building your own version can you have control over the first version, and only then can you iterate a 6 past a 7. Copying someone else's 7 is a ceiling with no exit.
The Desired State System is the right metaphor for managing a company: like React's virtual DOM — "here's the state it should be in, here's the current state, what's the minimum path between them." Modeling the company as machine-readable configuration makes politics disappear and consequences visible.
The brain is a retrospective narrative-alignment mechanism: it rewards behavior consistent with your self-identity and blocks dissonant behavior. This can be actively exploited — through affirmations, letters to your future self, and requiring teams to publicly commit to a differentiated position — to rewrite your own (and your team's) identity.
A rival is worth more than a competitor: a competitor makes a company reactive — the most active Slack channel becomes the competitive-analysis channel, and the company gradually becomes a sophisticated follower. A rival makes you your best self — their purpose isn't to be copied, it's to be surpassed.
Ownership disappears once code enters the commons; stewardship is permanent: once code is merged, it belongs to the company, not to you. This design makes it psychologically safe to criticize old code — including your own — which makes continuous improvement possible instead of emotionally protecting past work.
Shopify's entrepreneurship data point: 80% of Shopify customers have someone who would reply within 24 hours to an entrepreneurial question. The key inversion: the causality runs backward — they became Shopify customers because they had that person. Entrepreneurial awareness itself is the decisive threshold.
Engineering everything vs. never writing it down: Tobi built Shopify OS — turning the company entirely into engineering, into configuration files, driven by a SAT solver. But later in the conversation he says flatly that "actively avoiding writing certain things down is sometimes the most important thing you can do" — once you write down hiring criteria, you'll only recruit people who can recite the criteria, not people who genuinely embody the traits. This tension is never fully resolved: what should be engineered (titles, levels, compensation) versus what should stay as tacit knowledge (taste, culture, who's actually worth hiring)?
No introspection vs. deep introspection: Tobi's brain-narrative framework, his messages in a bottle, his annual summit self-audits, his letters to himself, his use of affirmations to rewrite identity — this is a highly self-reflective system. But he also says "trust your gut," "intuition is the fast decision-making that comes from your entire life's accumulated story." His operating system demands enormous metacognition while also insisting on "letting intuition run instead of forcibly analyzing it." This directly contradicts Marc Andreessen's zero-introspection thesis — yet may also be two sides of the same coin.
A rule-bound system vs. an F1 mindset built to beat the rules: Tobi built an extremely precise rule system in Shopify OS, yet he admires the F1 engineer's instinct to "read the rulebook looking for loopholes." He even admits that when an employee found a loophole in the compensation system to earn more money, his reaction was "good for them" — "if the incentive system tells you to do something misaligned with the company's goals, that's a flaw in how I designed the system, not your problem." This directly contradicts the spirit of Shopify OS's design — a carefully engineered rule system invites carefully engineered evasion.
Tobi has been running Shopify for 21 years, and David says he's one of the names other founders mention most — the word they use is "singular." One claim runs through the entire episode: the right way to build a company is to treat the company itself as an engineering project.
There's a roundabout bit of logic at the start that's actually quite deep: Tobi says a company is a "social technology" — its only real function is to legally let thousands of people join you in pursuing a counterfactual. You say, "I think the world should be like this, but it currently isn't." If the market agrees, it rewards you with money so you can keep doing it. If it disagrees, the market is telling you your counterfactual was wrong. It's a strange invention in human history: if proposed for the first time today, derived from first principles, it would sound completely insane. But precisely because it exists, you get to spend fourteen hours a day chasing what you believe is right without being cast out by society.
After the IPO, Tobi made a mistake it took him fifteen years to fully recognize: he started "cosplaying a serious public-company CEO." A 60-year-old man in a suit, who delegates everything. He'd say "you look passionate, I trust you." The result: a team in the Toronto office quietly built functionality letting Shopify run supermarkets — nobody told him. It wasn't until COVID hit, and he spent sixteen hours a day personally reviewing every project, that he discovered all these "random projects." He canceled 60% of them and replaced every executive within a year. He says COVID saved Shopify.
His fix was, fundamentally, an engineer's fix: turn the company into a desired-state system. Starting from a new GitHub repository, he turned every rule the company ran on — titles, levels, compensation data — into machine-readable configuration files, modeled with Python and a SAT solver. He discovered that an 8,000-person company had 5,500 distinct job titles, with "senior staff" ranking below director on some teams and above it on others. Once these became things software could address, every crazy decision became immediately visible. He calls it Shopify OS. It eliminated politics: when a department head asks for 50 new sales hires, you can immediately see how many people that pulls from engineering, or which budget gets cut. No one can make a decision with only an upside on the golf course anymore.
"Differentiate, even if it's worse" is the Dyson line David read to Tobi, and Tobi fully agrees. The logic: if you copy someone else's 7/10, you can only ever be a 7 — that's your ceiling. If you build your own version from scratch, even if it starts at a 6, you have complete control over it, and you can iterate it to an 8. On speed: "the world belongs to the people who are fast, who iterate, who understand what's costly and what's unnecessary." The SpaceX Raptor engine's evolution chart is "today's Picasso" — an example of moving forward through subtraction, something almost nonexistent in industry, but precisely the force that puts SpaceX far ahead.
On the brain: Tobi believes the brain is a "retrospective narrative-alignment mechanism" — it continuously tries to keep your history consistent with your self-identity, rewarding identity-consistent behavior and blocking dissonant behavior. This mechanism can be exploited. He spent a week writing "I love public speaking" every day, and a week later, he actually did. His requirement that executives publicly explain once a year how Shopify does things "differently" in their domain isn't just an intellectual exercise — it's an identity-reinforcement tool: once they've stated it publicly, they can't not live up to it, because that would create dissonance with their own self-identity.
On talent: "you don't need to recruit better people, you need to become a company worth the best people working for." Shopify stayed in the east (Ottawa/Toronto) deliberately, rather than moving to Silicon Valley — because he knows he can't out-pay Silicon Valley, but he can become "the single best company" in the east, magnetizing every engineer who wants meaningful work.
Finally, on AI and year 21: managing AI agents feels exactly like playing StarCraft — six agents running at once, while he micromanages the most critical one and a critic agent watches all of them. Tobi says, "if AI hadn't happened, I wouldn't still be running Shopify." His mission is to cause more entrepreneurship — turbulence plus interestingness is the environment he's best suited for. Year 21 is the most interesting year in the company's history.
A faithful reconstruction and plain-language retelling of the episode, generated by PodLens.
This is one source-grounded reading, not a replacement for the original. Every point is anchored to its source, so you can check it yourself — and corrections are welcome.