Tobi Lütke: Why Shopify Welcomes the AI Agents Amazon Blocks

Shopify CEO Tobi Lütke explains why Shopify embraced Meta’s Muse shopping agent while Amazon blocked it, and how AI agents will change who gets paid in commerce. In a long sit-down with Alex Heath on the Sources Podcast, recorded at Shopify’s Toronto office, Lütke covers agentic shopping, the AI backlash, his reflexive AI memo, take rates for agent startups, the future of advertising, and the GPU rack he installed in his old wine cellar to run a fully local personal agent.

TLDW

Lütke frames the Amazon and Shopify split over AI agents as the same twenty-year philosophical divide between an everything store and independent merchants, and says Shopify will support a purchase whether a person or an agent makes it. He argues that optimism about AI is rational because entrepreneurship is the job that benefits most from it, and that the pessimism around the technology is a status signal more than an analysis. On money, he tells agent startups to stop counting on a 2% cut of every transaction and to become too valuable to ignore, says he cares about gross profit dollars rather than percentages, and warns that advertising inside an assistant that is supposed to work for you is a breach of trust waiting to happen. He closes with his home server rack, a Dell GB300 running a personal agent that knows his calendar, his location and his heart rate.

Thoughts

The sharpest idea in the conversation is Lütke’s answer to the question every agent startup is quietly asking: where does the money come from? Heath points out that nearly every personal agent company is pitching a take rate, a slice of each transaction that flows through it. Lütke’s reply is blunt. Shopify takes roughly 3%, most of that goes to the card networks, and the agents are “not going to get 2%… welcome to the world of basis points.” Then he says the more revealing thing: he would rather Shopify charge a lower percentage on far more volume, because the number he manages to is gross profit dollar growth, not margin. That is a coherent strategy for a platform whose fortunes depend on how many new businesses get started, and it quietly reframes the agent economy. If the incumbent payment and commerce layers are already fighting over basis points, agent companies will have to earn their keep the way software always has, by being so useful that users pay for them directly or would pay a lot never to lose them. His “reverse money test” (what would you have to be paid to give up your agent forever?) is a better way to think about agent value than any take rate spreadsheet.

The advertising section deserves more attention than it will get. Lütke is not anti-ads. He calls Meta’s ad engine the greatest boon to small businesses of all time and makes the interesting argument that advertising works as a kind of proof of work, because a business that can afford to put money behind a product has probably already sold some of it. But he also names the structural problem plainly: an agent is supposed to act on your behalf, and if it turns out your assistant is being paid to choose products, that is a breach of trust. His worry is concentration. Department stores rewarded whoever bought the best shelf space, and the internet broke that by letting small brands find their thousand true fans. If agents collapse discovery into a single recommendation, and that recommendation can be bought, the shelf space problem comes back with a much better disguise. Whoever designs a discovery market that agents can use without betraying their users will own one of the most valuable businesses of the next decade, and Lütke clearly knows it.

Lütke’s description of how he avoids becoming an incumbent is unusually concrete. At the start of every year he works out how he would build a better Shopify from scratch, then spends the rest of the year making Shopify more like that. Later he describes the same loop as imagining the perfect version of the product with a magic wand, pouring a stiff drink, and running the diff against what actually exists. Plenty of CEOs say they are willing to cannibalize themselves. Fewer admit, as he does, that they are now the incumbent and that the real competition is “two people in a garage plus a data center of agents.” That framing explains the reflexive AI memo better than the memo did. A company that expects tiny AI-native teams to attack it has to make AI use the default internally, and it helps that, as he notes, a founder can send that memo without a committee reviewing it first.

The wine cellar story is played for laughs, but it points at something real about how personal agents will spread. Lütke says that as a public company CEO he cannot send much of his life to an outside lab, so he built a local rack and runs his agent entirely in his house. It reads his calendar, his location and his Fitbit data, writes his briefing notes, and once told him which colleagues give him the most stressful meetings based on his heart rate. The value comes from access, and access is exactly what people with something to protect will refuse to give a cloud service. That suggests the most capable personal agents may first be adopted by people who can afford to run them privately, and that local inference is a feature, not a hobbyist detail. It also shows how much a well-connected agent can infer, which is both why he calls it the most compelling product he has ever used and why the stress-meeting anecdote lands as slightly alarming.

The opening news hook, Amazon blocking Muse while Shopify welcomed it, makes more sense in light of all this. Lütke says Shopify deliberately never built a marketplace because a marketplace has three sides, and when buyers and sellers conflict the platform always ends up favoring buyers, whose behavior is easier to measure. Staying on the merchants’ side of the table leaves Shopify with nothing to lose when an agent shows up at checkout instead of a person. Amazon, with an ad business to protect and a storefront it wants shoppers to see, has much more to lose. The difference in reaction is not about how either company feels about AI. It is about which business model an agent actually threatens.

Key Takeaways

Shopify supports purchases made by AI agents on a customer’s behalf and is building APIs to make agent checkout work well, while Amazon chose to block Meta’s Muse agent.
Lütke sees the split as the long-running contrast between Amazon as a digitized department store and Shopify as the home of independent, storytelling brands.
Shopify never built a traditional marketplace because three-sided platforms end up favoring buyers over sellers, and Shopify wants to stay on the merchants’ side.
Early agentic shopping skews toward considered, project-based, high order value purchases, such as outfitting a home gym or buying everything needed for a weekend 3D printing project.
Lütke believes pessimism dominates because it sounds sophisticated, and that decades of dystopian science fiction contributed to a gloomy mood inside the tech industry.
He calls entrepreneurship the job that benefits most from AI, since founders must understand every role and previously relied on having someone who would answer their questions quickly.
The most common time for a new Shopify store to be created is the local lunch break, which Lütke reads as people trying the plan B they have been thinking about.
His reflexive AI memo still holds inside Shopify, and he says it was more controversial in the press than inside the company.
He argues companies should make their culture explicit, including being clear that AI skeptics are unlikely to become top performers in a company built around agents.
Agent startups hoping for a 2% cut of transactions will be disappointed, he says, because commerce already runs on basis points, so value to the user has to come first.
Lütke manages Shopify to gross profit dollar growth rather than percentage margins and would rather charge less to more merchants.
He credits Meta’s open-sourcing of the Llama models with setting a norm for open AI models and praises how quickly Mark Zuckerberg rebuilt after the Llama 4 setback.
Advertising will survive the agent era in some form, but paying an assistant to pick products would break the trust that makes agents useful.
Lütke replaced his wine cellar with a server rack holding a Dell GB300 and runs a fully local personal agent that uses his calendar, location and health data.

Chapters

0:00 Amazon Blocks Muse, Shopify Embraces It

Heath opens on Amazon blocking Meta’s Muse agent, which Lütke learned about after Shopify had already announced it was embracing Muse. He describes Amazon and Shopify as two long-standing philosophies of commerce rather than right and wrong answers. Shopify has no ad business to protect and no marketplace, so it is happy to support checkout whether a human or an agent is buying, and it wants merchants’ products everywhere they can be sold.

7:00 How AI Agents Are Changing Shopping

Agentic purchasing is growing quickly from a low base. Lütke sees considered, project-style orders, such as setting up a home gym, and is most excited by agents that buy everything needed for a weekend project and place orders with local stores. He expects agents to move from helping people do existing tasks better to enabling things that were not possible before, like an agent connected to a 3D printer that reorders filament.

11:05 Science Fiction and Tech Optimism

Lütke recently reread Cryptonomicon and was struck by how much of it came true. He argues science fiction says more about the era it was written in than the future it describes, and that a long stretch of dystopian bestsellers fed pessimism in tech. He sees optimism returning now that we have software we train rather than write.

17:57 Zero-Sum Thinking Versus Abundance

Zero-sum thinking, Lütke says, is a choice made by people who misunderstand how value is created. The AI industry pulled trillions of dollars of value out of floating point math and matrix multiplication that had existed for decades. He points to Shopify merchants who quit accounting jobs on a lunch break and built businesses hundreds of thousands of people now buy from.

21:20 The AI Backlash and the Founder Advantage

Lütke blames part of the backlash on the Prometheus myth, the instinct that every powerful gift carries an equal punishment, and says his p(doom) is zero. He compares AI attitudes in the US, a more cynical Europe and Canada in between, credits Toronto’s role in deep learning, and explains why he keeps Shopify outside Silicon Valley. He also praises Meta’s open-source Llama release and Zuckerberg’s recovery after Llama 4, and argues that a founder in the CEO seat can cut through bureaucracy in ways a hired manager cannot.

45:28 Building an AI-First Shopify

The 2025 memo requiring reflexive AI use was controversial outside Shopify but not inside it. Lütke says he has never had an annual plan, sees something astonishing in internal reviews roughly once a week, and believes companies should be explicit about who they are for. Including AI use in a peer review question, he says, gave the company a reputation to live up to.

52:18 Who Profits From Agentic Commerce?

Heath asks whether agent startups can really live on a cut of transactions. Lütke says they will not get 2%, that commerce runs on basis points, and that agents providing real value will find ways to charge, whether through subscriptions, metering or transactions. He prefers lower percentages on more volume, manages to gross profit dollar growth, and wants AI agents to be able to sign merchants up for Shopify with as little friction as possible.

1:03:22 How Shopify Stays Ahead

The real risk, Lütke says, is someone building a better Shopify, so he starts each year by designing one himself. He admits Shopify is now an incumbent, describes the sunk cost fallacy as the thing that kills companies, and ties Shopify’s mission to the oldest human habit of trade, from Ötzi the iceman to clay tablets recording early ventures. His advice to startups is to be too good to ignore.

1:10:39 Advertising and Discovery in the Agent Era

Companies throwing up walls against agents should return to problem-solving mode, he says, since ads and subscriptions are only solutions they adopted along the way. Advertising will change shape but discovery still needs a market. He calls Meta’s ad engine the greatest boon to small businesses ever, describes ads as a form of proof of work, and warns that a paid assistant choosing products would be a breach of trust.

1:16:02 The Rack Cellar and a Fully Local Agent

Lütke turned an underused glass wine cellar into a server room with a Dell GB300 so his personal agent can run entirely at home. It guessed who he was during setup, writes briefing notes, and draws on his calendar, location and Fitbit data, including a query that revealed his most stressful meetings. He says losing it would feel like one of the most significant regressions he can imagine.

Notable Quotes

“Pessimism is always what is the default of everyone because it sounds sophisticated. You sound smart being pessimist. Optimism sounds naive.”
Tobi Lütke, on why gloom dominates conversations about technology

“I think to me dystopia is always just 40 years ago.”
Tobi Lütke, on fears about where AI is taking society

“You’re not going to get 2%. Like I mean welcome to the world of basis points.”
Tobi Lütke, on agent startups planning to live on a take rate

“I don’t care about percentages. I don’t care about gross profit. I care about gross profit dollar growth.”
Tobi Lütke, on whether Shopify should charge more as agents do more work

“I spent beginning of every year trying to figure out how I would build a better Shopify and then spend the rest of the year trying to make it more like that.”
Tobi Lütke, on the biggest risk to Shopify

“A business that can’t put money behind their product hasn’t made money.”
Tobi Lütke, on advertising as a kind of proof of work

“There would be a lot of things I would sacrifice on the altar of thought experiments long before my claw.”
Tobi Lütke, on how much he relies on his personal agent

Watch the full conversation here.