Tobi Lütke - Building Islands of Innovation - [Invest Like the Best, EP.394]
What this covers
Patrick talks with Tobi Lütke, Shopify's co-founder and CEO, for the third time on the show. The conversation opens with AI, then settles into a longer argument about how organizations are run: Founder Mode, keeping a company's temperature high, and building on an island apart from the mainstream. Lütke extends genetic isolation to memetics, arguing that ideas travel far faster than DNA, so the internet's connectivity erodes the separation islands need. His examples include Ruby, which Matz designed around programmer happiness, and open source as a marketplace run on legitimacy rather than money.
The AI sections are the most concrete. Lütke describes giving models personas and letting them argue, which beat single prompts, and says placing the agents in a 3D office where they bumped into each other improved results further. He forecasts conversational, agent-mediated interfaces, with shopping becoming a dialogue with an AI that knows your calendar. He says AI will replace the toilsome parts of white-collar work, not the jobs, and estimates that a frozen GPT-4 could still yield roughly $10 trillion of value over a decade. He also sees big models as a homogenizing force, with Anthropic and Meta's open-source releases creating new islands.
The rest is management and money. Lütke calls GAAP a dashboard for Wall Street that misses most of a company's real assets, and he says zero-interest-rate years turned Shopify's frugality into a liability. He explains why well-meaning colleagues politick against exothermic people, why he promotes on merit and skips rungs, and why change should land all at once, like saturating a sponge.
Lütke argues that durable companies and innovation depend on protected 'islands' led by founders who inject energy, legitimize action through first-principles truth rather than consensus plans, and keep distinct value systems instead of converging on the mainstream.
- Founder-led companies gain legitimacy from people and first-principles correctness, while managerial companies gain it from consensus plans that resist change
- Innovation comes from islands with membranes that filter mainstream ideas, and bridges to the mainstream erode differentiation
- AI is a powerful augmenting instrument that removes tedious work and rewards judgment and taste, and most of its value is still unrealized
This asset isn't compiled yet
You're seeing its claims, ranked. Compile it to build the argument threads, weight them, and check each claim against your library — the full view.
Ruby was designed by Matz to optimize for programmer happiness, with aesthetic choices such as 'poetry mode' that omit brackets and semicolons, and this single-minded vision is an example of an island that produces gifts for everyone.
“Ruby optimizes for programmer happiness. It implements, to the great pains of the implementers of the language, something called 'poetry mode'”
Open source functions as a marketplace that runs on legitimacy and merit rather than money, because anyone can fork a project and offer gifts that succeed only if they are better, and it is led by small groups with highly legitimate leaders.
“it's a marketplace that just doesn't run for money; it just runs purely for legitimacy and merit.”
Zero interest rate periods turned Shopify's frugal, capital-efficient characteristics into a liability, because companies cannot compete with competitors spending freely on free money, and this left Shopify too exposed.
“the zero interest rate periods really—they made all these characteristics of the business basically a liability because you cannot compete with everyone free-spending on free money.”
Lütke expects computer interaction to evolve from application windows toward conversational, agent-mediated workflows, with the old windowed and Unix paradigms persisting beneath, and expects shopping to become a conversation with an AI that knows the user's calendar and preferences.
“I think a lot of what we are doing on computers is going to take a similar form. It's just like programmers switch to a Unix command line.”
Most of Shopify's valuable assets, including customer trust, the ability to attract top engineers, and its system of record, cannot be captured on a balance sheet, yet they form a defensible investment thesis.
“almost every asset of this company can't be put onto a balance sheet or an asset statement.”
Much value that never gets added to global GDP does not exist because of friction, confusion, lack of access to information, and defensive naysayers, and Lütke hopes AI will remove those barriers to independent entrepreneurship.
“a lot of products, lots of things that don't exist—a lot of value that wasn't added to a global GDP—happens because friction, confusion, lack of access to information, and too much defense.”
GAAP accounting is basically a dashboard for Wall Street built on top of the facts of a business, and it fails to capture most of a company's real assets.
“I think GAAP accounting is basically a dashboard for Wall Street on top of facts of a business.”
Lütke extends genetic isolation to memetics, arguing that ideas are exchanged far faster than DNA, which means the internet's connectivity erodes the separation that islands need to produce distinct ideas.
“I'm just making the leap here from genetics to memetics—obviously not the first person to do this. Memetics looks exactly the same; the exchange of ideas just happens faster than the exchange of DNA.”
Tobi promotes people based on merit rather than seniority, moving individual-craft people into charge of large departments after they delivered results, skipping rungs that career ladders would normally require.
“I picked people from individual craft roles and put them in charge of large, hundreds-of-people departments after they ran a code red for me that just really worked.”
Computer interfaces will move toward conversational and agent-based interaction, where applications may exist only temporarily, similar to how programmers use a Unix command line beneath graphical windows.
“a lot of what we are doing on computers is going to take a similar form. It's just like programmers switch to a Unix command line.”
Large AI models have a homogenizing force that resembles mainstream fashion; OpenAI's fully formed products spread quickly to everyone, while Anthropic's different tone and Meta's open-source models help create new islands.
“Academia was an island here in this metaphor, and through the related peninsula, I suppose, of OpenAI, ended up creating fully formed gifts that then were exported to everyone.”
The technology industry is incredibly path-dependent, so being first matters because the first movers establish the norms that everyone else follows.
“Our industry is incredibly path-dependent, so this is where being first matters, because you establish the norms.”
Even if AI can custom-build all software, businesses will still need a trusted system of record and system of truth, which Lütke argues forms the basis for a durable investment thesis.
“Let's say tomorrow the all-knowing AI comes around that can write all software custom-build—well, you still need a system that you trust to be a store record of information, a system of truth.”
Giving AI models personas, having them argue positions, and placing them in chats with each other produces delightfully different and better outcomes than simply prompting a single model, because they can question each other and collaborate over multiple turns.
“giving AIs personas, and then having them argue a position, and then even putting them into a chat with each other, leads to just delightfully different outcomes than if you just prompt.”
AI will not replace white-collar jobs, but will replace the crappy, toilsome parts of work, letting people spend more net time on judgment and taste.
“I don't think they will replace the white-collar jobs. I think they will definitely replace the crappy parts, the toilsome tasks that we don't want to do anyway.”
Change should be loaded all at once, like saturating a sponge, because people resist only up to a saturation point, so any change beyond that point is effectively free, whereas long drawn-out change programs are incompatible with human psychology.
“change management is best done as just saturate the sponge—load up all change in one go—because people will mind, but they can't mind more than the saturation point.”
Even if model development froze at GPT-4 with unlimited cheap calls, Tobi estimates roughly 10 trillion dollars of industry value could be extracted over ten years, and society has captured only single-digit percentages of that potential.
“And I think I would put that at single digits.”
Placing the AI agents in a 3D office environment where they moved around and encountered each other unexpectedly produced vastly better results than the text-only setup, because the changing combinations of who met whom contributed to quality.
“once we ran it through that, actually, the results ended up being vastly better.”
Trust builds fastest when people disagree with leadership and are right, based on what they know, and then articulate it, because that shows they prioritize truth over consensus.
“people build more trust with me quicker when they disagree with something and are right—or at least right based on everything they know—and then articulate that.”
Converging on the consensus decision is deadly because it is the fitness function of the mainstream; an island must instead use actual true correctness as its fitness function to be better than the mainstream.
“You're trying to converge on the consensus decision. I think it's deadly. I think that's a fitness function of the mainstream.”
Tobi converts tasks that current models fail into evals, then reruns them on every new model to see which capabilities have changed, calibrating his expectations from there.
“I turn them usually into something called evals, just so every time I get a new model, I run it against the evals of stuff that I couldn't get them to do before”
Companies run by founders derive legitimacy from the founder's ability to rebuild the company from first principles, whereas managerial companies derive legitimacy from a consensus plan, which makes them slow to change.
“If legitimacy comes from a founder, rather than legitimacy comes from a plan, because this is the core of a difference.”
Once a plan has legitimacy, it is much easier to change the numbers within it than to change the plan itself, which is why companies see goalposts move and why they react slowly to change.
“It's much easier to just change numbers in a plan that already has legitimacy than it is to change the plan entirely. So it's much harder to react.”
Well-meaning people in organizations fight high-energy colleagues because they try to conserve energy, and the usual way out is to politic the energetic person away, after which the room-temperature norm returns.
“And the best way out of it is to just politic that person away.”
Founders protect exothermic colleagues by making politicking against them fail, and they can recruit like-minded people, such as other founders, into their teams.
“Not just are they injecting heat from the top into the whole thing, they also protect the others. When things are going well, they make the politicking against them not work.”
A high merit-to-ego ratio is necessary: brilliant jerks are toxic, but people who are genuinely good at what they do may be self-confident, because confidence without skill is arrogance and confidence with skill is just communication.
“So high merit-to-ego ratio is important. Some people are incredibly good at what they do, and they are allowed to be self-confident about that.”
Many CEOs suffer a split-brain problem in which data teams only forecast and finance teams only produce GAAP reporting, so business thinking becomes incongruent unless the two are tightly connected.
“many CEOs live in a split-brain problem world where data is just forecasting and finance is just GAAP.”
Innovation and difference come from islands, which are separated by barriers such as water and language, and as the mainland and the internet homogenize ideas, few islands remain, so progress depends on creating more of them.
“But innovation comes from islands. Difference comes from islands, literally.”
When a mainland bridge or highway is fully built to an island, the membrane separation breaks down, differentiation is lost, and the island can change only marginally, not with the richness people hoped for.
“If it's integrated fully—the bridges, the highways—the differentiation is broken.”
When critics say a product will never work in the real world, the right response is that Shopify is not building for the real world but for its own island, and it invites critics to use it if they find it better.
“I'm not building for the real world. I'm building for my own island. And I'll invite you over; if you think what I have is better, use it.”
Human emotions are the only net adders of energy into systems, so builders should identify which emotions, such as anger or a desire to prove people wrong, fuel their exothermic energy.
“Human emotions are the only net adders of energy into these systems.”
A committee would never approve building software for someone in their lunch break that also powers billion-dollar businesses, yet that is the right thing to do, because today's small merchants become the billion-dollar businesses.
“it would be very hard for a not-founder-mode company... to create a plan that, for consensus, will sign off on that says, 'Hey, let's build software for people in their lunch break that can also power billionaire businesses.' That's not a sensible thing to do. Yet it's exactly the right thing to do.”
Lütke argues that numbers are the facts that show up on trend lines, while forecasting and heuristics are models built on top of them, and he insists finance and data teams at Shopify work closely together to avoid a split-brain problem.
“many CEOs live in a split-brain problem world where data is just forecasting and finance is just GAAP.”
Taming complexity into an approachable interface is unglamorous work that can cause freezing if not done, yet it is required to serve both small entrepreneurs and multibillion-dollar businesses with the same software.
“So one thing that's unfun sometimes is this taming of complexity, which is building software that actually works for people that try to reach for independence and glory in their lunch break, and still the same software works for multibillion-dollar businesses.”
The best leaders are exothermic, acting as a wellspring of heat that speeds up everything around them and prevents stasis.
“The best leaders are literally exothermic. They are just like this complete wellspring of heat into the system.”
A founder can rederive strategy from new information by replaying the company's entire history with that information inserted at the beginning, asking where the company would be if the fact had been different.
“replay the entire history of the company, take this new piece of information, insert it at the beginning, run the counterfactual. Where would you end up if that's different?”
Lütke argues that building a company on an island requires separation: he built software that he wanted but could not find, and ignored critics who said it would never work in the real world, because he was building for his own island rather than the mainstream.
“I'm not building for the real world. I'm building for my own island. And I'll invite you over; if you think what I have is better, use it.”
Individual parts of Shopify could be argued to be suboptimal, but the holistic set of trade-offs is what makes Shopify work, and without maintaining these tenets through membrane separation the company would revert to the mean and become like all others.
“the holistic picture is like a set of trade-offs that work extremely well for Shopify to make Shopify and keep it Shopify.”
Difference does not automatically mean better; it only means different, and the value of islands is that participants can switch to another island when one is not working, as long as there is no coercion.
“But difference doesn't mean better. Difference just means different.”
A principle is not a principle unless it costs you something, so teams that do what is right even when easier options exist are the ones worth investing in during times of change.
“it's not a principle unless it costs you something.”
In the best companies people spend up to 20 percent of their time on politics, while in the worst companies it reaches up to 60 percent, and the margin for operating well is small.
“In the best companies in the world, people spend up to 20% of their time on politics, and in the worst companies, it's up to 60%.”
Tobi frames his agent 3D environment, Terrarium, as an instrument rather than a tool: like a flute, it produces wrong output if given a bad instruction, and results depend on the skill of the player.
“I've come to think of Terrarium as not a tool, but actually as an instrument. If you send the wrong instruction in, it goes totally off the rails, just like if you pick up a flute and hit the wrong note.”
Managerial CEOs are custodians who inherited their companies, and very few could rebuild the company tomorrow if everything went away, unlike founders.
“If you're a managerial-run company, you're a custodian of a company; you inherited it. Very few leaders of companies—there are some—but very few of them could rebuild the company tomorrow if everything goes away.”
In managerial companies, the plan rather than the CEO tells people what to do, so good outcomes from hiring good people and staying out of their way happen accidentally because everyone is simply following the plan.
“All these stories about hire good people and get out of their way actually work, but they work accidentally because everyone's just following the plan.”
Companies converge on a room-temperature equilibrium of low energy, which is their entropic equilibrium, and that is not a problem as long as everyone else operates the same way.
“I think companies run at room temperature because that's their entropic equilibrium of temperature.”
The Ruby programming language was designed to optimize for programmer happiness rather than other goals, reflecting its creator's aesthetic choices, which Shopify depends on.
“Ruby optimizes for programmer happiness.”
In a multi-agent setup, the most important insight was needing an agent acting as a project manager who knows what the human wants and holds the other agents accountable to progress toward the goal.
“Most important insight was I really needed someone representing me in this, like a project manager who knew what I wanted.”
Millions of niche businesses, such as a store specializing in pirate-themed poker sets, create net value and secondary employment, and much value that never reached GDP is lost to friction, confusion, lack of information, and naysayers.
“a lot of the products, lots of things that don't exist—a lot of value that wasn't added to a global GDP—happens because friction, confusion, lack of access to information, and too much defense.”
Attention, not minerals or resources, is the most limited resource in competitive contexts, so managing attention is one of the most important leadership skills.
“your most limited resource is not actually the minerals and the Vespene gas; it's actually attention.”
Products are good when the people who make them care deeply, because they fill the product with that caring, which the customers feel, and this is the simplifying principle beneath differentiation.
“products are good when the people who make them give a shit, because what they do is they fill up the products with giving a shit, and everyone feels it on the other side.”
When companies treat customers as a crop to be harvested rather than as partners, the approach works only for a moment, because it is not a sustainable long-term relationship.
“your customers, by a company, stop being treated as a customer and are now treated as a crop to be harvested. That only works for a moment”
Shopify's product philosophy is to make the important things super easy, like setting up a store in a minute, while keeping everything else possible, so advanced users can customize deeply.
“We follow a philosophy of make the important stuff super easy and make everything else possible.”
The most valuable actions of founder-led companies at scale are aggressive redirections, which managers would struggle to achieve because they require consensus that is rarely reached.
“It seems as though the most valuable potential actions of founders and founder-led companies, especially at scale, is aggressive redirection.”
Without island-like boundaries that protect a company's tenets, the company will revert to the mean and become like all others, which Lütke says is the outcome he would consider career-level failure.
“otherwise you're losing those tenets, and then reversion to the mean will happen, and then you have a company that's like all others.”
Participants on an island get what they want from it and do not need everyone else to adopt it; as long as no coercion is involved, a rich marketplace of ideas emerges where ideas can be judged, joined, adopted, or pulled into the mainstream.
“As long as there's no coercion, you end up in a much more rich marketplace of ideas where lots of people can just bring ideas into the manifold.”
Wall Street asks why Shopify does not monetize end consumers more, but treating customers as crops to harvest is unsustainable, and Lütke argues Shopify's sustainable approach is to build value for merchants who are partners.
“your customers, by a company, stop being treated as a customer and are now treated as a crop to be harvested. That only works for a moment, because this is not a sustainable long-term relationship”
The most magical moment in technology is when something goes from being impossible to being just very hard, because that is when all builders need to converge on it.
“The most magical moment in the world of technology is when something goes from being impossible to being just very hard.”
Growing quickly in the mainstream requires speaking mainstream language, often through marketing agencies, whereas island-based companies should tell their origin stories to people already on the island rather than to convert the mainstream.
“If you want to grow as fast as possible, you have to convince people to switch the mainstream narrative in your direction.”
As a leader, one of the best ways to develop high-performing younger engineers and leaders is to give them a reputation to live up to, set slightly above their current abilities, so they rise to the occasion.
“give them a reputation to live up to. And that's maybe slightly pushed from their current set of abilities, and then they rise to the occasion.”
A plan is not sentient and cannot decide to cut wasteful activity such as meetings; the authority to change it lies with a leader who has legitimacy, which managerial CEOs often lack.
“The plan is not sentient. The plan can't just say, 'Hey, this company is spending too much time in meetings.'”
Patrick argues that Shopify's clear principle for how things should be, combined with standards for bringing them to life, resembles Bret Victor's idea of inventing on principle, where violating the principle means you are out.
“You reminded me of Bret Victor's idea of inventing on principle.”
Open-source islands produce gifts for everyone without seeking adoption, as with Ruby on Rails built on Ruby, so that the island yields gifts that enrich the broader ecosystem.
“Like all good open source software, it's an island that creates gifts for everyone. It is not part of a mainstream.”
Shopify's pivot away from the Shopify Fulfillment Network toward AI-driven software was correct and unpopular, reflecting the first-principles judgment that Shopify should deepen its committed field rather than enter adjacencies.
“So, because of that, I'm like, cool, let's go deeper into the field that we already committed ourselves to and not go into the adjacency.”
Lütke admits Shopify's island was breached during the long zero-interest-rate period followed by a bust, when someone built a bridge to the mainland and the company absorbed ideas it did not want, and he rates his own defense as only a B minus or C plus.
“I got washed. Someone built a bridge to the mainland and I didn't notice. I got infected by some ideas I rather would not.”
Tobi admits that after more than a decade of zero-rate conditions following a ten-year boom and bust cycle, Shopify was infected by mainstream ideas, because its defenses were not strong enough to stay outside the mainstream.
“Someone built a bridge to the mainland and I didn't notice. I got infected by some ideas I rather would not.”
Shopify is the cheapest software in merchants' stacks, and merchants often pay ten times more for their CRM than for the software that runs their whole business.
“They often pay ten times more for their CRM than for the software that literally runs their entire business”
Once you assume something is now possible given a new development, you must test it, and whether or not it works, your world model needs to update to incorporate what is now possible.
“once you make this assumption that something would be possible now given this new development, and then you test it, it doesn't matter if it works. Your world model needs to update.”
After Tobi deleted recurring meetings, average meeting load fell from five and a half hours to three and a half hours per person for the next year, and the company looked better because engineers got more work done.
“Meeting load goes from people averaging five and a half hours to three and a half hours for the next year”
An idea is worth about one good bottle of Scotch, and implementation is what matters, so the many Wall Street ideas about monetization are less important than how they are executed.
“I just think people overestimate the value of an idea. An idea is worth one good bottle of Scotch. Everything else is implementation, and here's the thing: the implementation matters.”
Shopify uses founder shares, so Tobi controls about 40 percent of votes, and he argues good governance rather than one-vote-one-share should be the investor's test.
“I vote 40% of the thing. So it's not the traditional share company in that way”
Financial and operating numbers are the bold facts on the graph paper of trend lines, while everything else, such as forecasts, is only as good as the model and heuristics behind it.
“the numbers—the 'science,' as you call it—they are the facts. They are the most bold-faced points on the graph paper of trend lines, because those you know for sure; everything else is as good as the model and heuristics.”
Niche entrepreneurs, such as a Shopify store selling pirate-themed poker sets, create net new value and employment, and Lütke argues that the value of such products is load-bearing for economic growth and living standards.
“It's just net value creation. This is how value comes into the planet again. It never gets lessened.”
Hard but unglamorous work, such as taming complexity, is where exothermic energy must come in, because without it organizations freeze at zero and nothing happens.
“The stuff that's hard and not seen as significantly fun is much harder to do because that's where the exothermic energy needs to come in.”
Darwin's finches on islands illustrate how isolated populations diverge, and by analogy memetics works like genetics except that ideas spread faster than DNA, so isolated idea-islands are needed for memetic difference.
“Genetic islands are really important. I'm just making the leap here from genetics to memetics—obviously not the first person to do this. Memetics looks exactly the same; the exchange of ideas just happens faster than the exchange of DNA.”
Academia, OpenAI, Anthropic, and Meta have each acted as islands whose fully formed products or open-source releases were exported to the mainstream, and Tobi argues that this export of gifts is how islands contribute.
“Academia was an island here in this metaphor, and through the related peninsula, I suppose, of OpenAI, ended up creating fully formed gifts that then were exported to everyone.”
Lütke argues that times when the map is foggy, with uncertain paths but a clear destination, are more fun and productive for builders because everyone can pursue their own route to a shared goal.
“I think the map is foggy again, which I think is always much more fun to be a builder in times where the map is foggy and everyone has their own destination.”
Lütke values entrepreneurship because the pursuit of building something new for others is glorious and requires courage, and the energy that customers' creativity generates is renewable, since watching customers reach for difference reinvigorates him.
“If your customers are inspiring you, again, that's exothermic as well. And frankly, they are exothermic people.”
Lütke says Shopify's emphasis on efficiency, such as operating with capital efficiency and having kept cash from financing rounds in the bank at IPO time, reflects a 'fitness function' that treats waste as painful.
“even then, I think we basically had all the money that we raised from financing rounds still in the bank at that time.”
A team made entirely of exothermic people is not possible at scale because such people are uncommon, though the greatest agents of change, like the US Founding Fathers, were all exothermic in different ways.
“I don't think it's possible to try that at any kind of scale, just because it's just not that common.”
Shopify replaced its website with a video telling the founding story of why Shopify exists, using the same explanation it recommends to merchants, because the company realized it had not followed its own advice.
“We replaced our website recently, just put a video on top there, basically.”
Lütke argues that the greatest agents of change, such as the US Founding Fathers, were all exothermic in different ways, and that they agreed on the broader picture and came together when it mattered, which is why they are fascinating.
“I'm sure the Founding Fathers of the United States had no passive players among them. ... They're all exothermic in different ways.”
Founders are recruited from exothermic people to begin with, because starting a company is an act of injecting heat into the world and fighting entropy.
“By definition, the main act of starting a company is you're trying to inject heat into planet Earth.”
Mainstream thinking is driven by fashion, shifting from one trend to another, and this aesthetic has shaped how companies are built, so Shopify could not make significantly different decisions from prevailing mainstream ideas.
“Mainstream thinking is incredibly fashion-oriented.”
Tobi believes the future is already here but unevenly distributed, so tool builders must live in a relative future, testing the newest capabilities in their own work and vacationing in them.
“I'm a tool builder, so therefore I need to live in everyone else's relative future, the very latest version of it.”
Shopify is not a place for everyone; its people are there because they want to be, and the fact that anyone can walk away at any time helps keep the company healthy.
“You can walk away from Shopify at any moment, and I think that's actually an important part of keeping things very, very healthy.”
During COVID, when Shopify had heavy executive turnover, Tobi recruited many former talent-acquisition founders from a Slack channel into leadership roles, and most of his current direct reports are other founders.
“I literally collected them during COVID. I had a lot of executive turnover.”
Whether AGI is already here is unimportant for most people, because the relevant question is how much utility can be extracted, and everyone else gets to build basically magic.
“So is AGI here? No idea. Does that matter? I think this is an extremely important question for a small group of people, and everyone else gets to build basically magic.”
Investors should buy into teams that are alive, somewhat exothermic, and willing to correct themselves, because the business is large and growing with long-term value.
“a highly alive, somewhat exothermic, sometimes wrong but always willing to do something about it”
Lütke sees AI as the latest in a series of platform shifts (internet, Web 2.0, mobile) that force the technology industry to reinvent everything, and he believes much of what happens in the world will be augmented by computers that now have highly valuable capabilities.
“And I think this was like the internet 2000s, web 2.0, mobile over the next ten years, and now we get to reinvent everything on top of AI.”