History of Shopify: From a Snowboard Shop to Agentic Commerce, and How Tobi Lütke Runs It (2026)

In 2004 a German programmer living in Ottawa wanted to sell snowboards online. He tried every e-commerce product on the market and hated all of them, so he wrote his own. The snowboard store was modest. The software under it became Shopify, a platform that processed $378 billion of merchandise in 2025, holds more than 14% of US e-commerce, and reported $11.6 billion in revenue.

That programmer, Tobi Lütke, still runs the company 22 years later. He never took the standard path. He dropped out of school at 16, did not take a salary for about four years, built early Shopify on an IKEA desk in his wife's childhood bedroom, and went public with only one executive who had ever worked at a public company.

Most histories of Shopify stop at the IPO or the pandemic boom. This one goes further: the near-death of the post-IPO years that Lütke describes as "cosplay," the COVID reset in which he cancelled most of the roadmap, the org chart he wrote as code, and River, the AI agent in Shopify's Slack that now starts a large share of the company's code changes. It ends with agentic commerce, where AI assistants do the shopping, and with what Lütke says the next decade will reward.

TL;DR: Shopify began in 2004 as Snowdevil, a snowboard store whose homemade software became a platform in 2006. It went public in 2015, surged in the pandemic, cut back in 2022-2023, and reported $11.6B revenue and $378B GMV for 2025. Its AI agent River, the Universal Commerce Protocol and agent-ready storefronts built with Vercel define its 2026. Build your own AI-powered workspace.

Shopify at a Glance

What Is Shopify?

Shopify is a commerce platform that lets any business sell online, in person and now inside AI assistants from one system. A merchant gets a hosted storefront, checkout, payments, point of sale, inventory, a marketing stack and an app ecosystem. Shopify earns subscription fees plus a larger share of revenue from merchant solutions such as payments, which is why its revenue grows with the sales of the businesses it serves.

The company's stated mission is to make commerce better for everyone. Lütke's personal version is narrower and more telling: to cause more entrepreneurship. He wrote that down in 2014 while preparing a talk for Shopify's internal Summit and has used it to judge his career since.

That mission explains decisions that look odd from outside. Shopify bought from Shopify stores to furnish its offices. It turned down the Silicon Valley playbook of staying private. It gave its own employees a slider to choose how they are paid. Each decision is a small bet that entrepreneurs, given better tools, will do more than anyone expects.

The Complete Shopify Timeline (2004-2026)

Before Shopify: Tobi Lütke's Apprenticeship

Tobi Lütke learned to program as an apprentice, not in a classroom, and that shaped how Shopify teaches people today. He grew up in Koblenz, Germany, left school at 16, and joined an apprenticeship at a Siemens subsidiary. In a May 2026 essay he wrote that the most interesting people there worked in the basement with Delphi, and he learned to be a programmer by watching them and making them coffee.

Germany has a word for that environment: Lehrwerkstatt, a teaching workshop where the whole shop floor is the classroom. The idea runs through Shopify's history. Early offices seated junior and senior engineers together in small pods on purpose. In 2026 it became the design rule for River, which only works in public so everyone can watch.

Two other threads from those years matter. Lütke became an early contributor to Ruby on Rails, the web framework David Heinemeier Hansson released in 2004, and he played StarCraft competitively as a teenager. He still credits StarCraft with lessons he later used at Shopify: information is everything, there is no right decision, only context in which decisions turn out to be correct, and attention is a resource you can spend or attack.

He moved to Canada in the early 2000s after meeting Fiona McKean, who became his wife. Ottawa, not Silicon Valley, is where Shopify began, and Lütke later argued that the distance was an advantage.

Snowdevil: The Store That Became a Platform (2004-2006)

Shopify exists because a snowboard shop could not find software good enough to sell its snowboards. In 2004 Lütke, Scott Lake and Daniel Weinand opened Snowdevil, an online store for snowboard gear. Lütke tried the e-commerce products of the day and found them clumsy, so he built the store himself on Rails, which was new and let one programmer move fast.

The store worked. The software was the part people noticed. Other merchants wanted it, and the founders slowly realized that the market cared more about the tool than the snowboards. Lütke's phrase for this, in a January 2026 conversation with David Senra, is that "the market pulled Shopify out of the project I started."

The decision was not obvious at the time. In a September 2026 interview on The Knowledge Project, Lütke pointed out that the snowboard store was already profitable, so his local incentives said to keep selling snowboards. Choosing the platform meant choosing the option with no quick payoff. That pattern, picking the good option that lacks a fast feedback loop, comes back again and again in how he makes decisions.

Shopify launched in June 2006. It shipped with Liquid, a template language Lütke designed so merchants could change how their stores look without being able to break the server. Liquid is still the basis of Shopify themes and was later open-sourced and used well beyond Shopify.

The Bootstrap Years (2006-2010)

Shopify ran for six years on almost no money before it took any venture capital. Lütke did not draw a salary for about four years. He and Fiona lived with her parents, and he built much of the product at an IKEA desk in her childhood bedroom. When payroll came due and the account was short, his father-in-law covered it.

Lütke told Senra that the company came within a week of running out of money "so many times" and that if you reran the first six years 10,000 times, Shopify would not survive most of them. He does not tell this as a hardship story. He tells it as a warning about money: "The main thing money does is it gets you a whole lot more of what you got before." A team that is loose with money early becomes much looser at scale.

Scott Lake stepped down as CEO in 2008 and Lütke took the role. It was not a job he wanted. He wanted to write code, and years later that tension produced one of the most important mistakes in Shopify's history.

In June 2009, on its third anniversary, Shopify launched its API and App Store with more than 5,000 merchants on the platform. That decision turned Shopify from a product into an ecosystem. Outside developers could now build features Shopify did not have time to build, and every good app made the core product more valuable.

The first outside funding came late. A $7 million Series A led by Bessemer Venture Partners closed in December 2010, six years after Snowdevil. A $15 million Series B followed in October 2011.

Payments, Plus and the Road to IPO (2011-2015)

Shopify stopped being a subscription business in 2013, when it started earning a share of what its merchants sell. On August 12, 2013, it launched Shopify Payments, built with Stripe, and Shopify POS for in-person sales on the same day. From then on Shopify's revenue grew with its merchants' revenue, which aligned the company's interests with its customers' for good.

A $100 million Series C led by OMERS Ventures and Insight Venture Partners followed in December 2013, valuing Shopify near $1 billion. In February 2014 the company launched Shopify Plus, a white-glove tier for large brands starting at $995 a month, with early customers that included the LA Lakers, Budweiser and Tesla Motors.

The IPO nobody in the room had done before

Shopify went public on May 21, 2015, at $17 a share on both the NYSE and the TSX. The company was valued at about $1.27 billion, had more than 162,000 merchants in 150 countries, and had reported $105 million in revenue for 2014. The stock closed its first day at $25.86, up 51%.

Silicon Valley friends told Lütke to stay private. He did the opposite and later explained why with a list: a public listing is free marketing, a currency for deals, and a discipline, and the downside was "a couple of annoying phone calls I have to do every couple of quarters." He also argues that keeping companies private reserves growth for accredited investors. Anyone who bought Shopify at its IPO got returns usually available only to venture funds.

The team was remarkably inexperienced for the job. Only one member of the executive team had ever worked at a public company. Their motto was: "We are not going public. We are creating a public version of Shopify." When they read the SEC rules for the investor road-show video, they found no rule that said it must be a CEO in front of slides, so they made a short documentary about why the company exists. Investors kept bringing it up on the road.

Shopify's Toronto office in August 2015, three months after the IPO. Photo: Wikimedia Commons, CC0.

The Cosplay Years: When Shopify Almost Lost Its Way (2015-2019)

After the IPO, Tobi Lütke tried to become a conventional public-company CEO, and he now says it almost killed Shopify. In his words to Senra, he decided to "cosplay a public company CEO, like a 60-year-old guy in a suit," delegating everything into business lines and trusting each leader to run their area. The financials looked fine. The product did not.

The problem was invisible from the top. Teams learned which projects the CEO cared about and brought him only those. Across a company of four to five thousand people in several cities, large projects ran without anyone connecting them. One office built features to run supermarkets on Shopify, on the theory that capturing 1% of groceries would be a big business. The team delivered on its brief and probably earned its bonus, and the result was "an island upon itself" that did not work with anything else in the product.

Shopify also had little competition in these years, which Lütke counts as a danger, not a comfort. He draws a line between a competitor, whom you copy and react to, and a rival, who pushes you to be your best. Andre Agassi needed Pete Sampras, even an imagined version of him. Without a rival, "even if something seems really good, you don't know, because there's no one else to keep you honest."

The same years were busy on the product side. Shopify Capital (2016), Shopify Pay (2017) and a package-tracking app called Arrive (2017, later the Shop app) all shipped. In 2019 Shopify announced the Shopify Fulfillment Network and bought the warehouse-robotics company 6 River Systems for about $450 million. Physical logistics was about to become Shopify's most expensive lesson.

COVID: The Crisis That Saved Shopify (2020-2021)

The pandemic doubled Shopify's business in a year, and Lütke says it also saved the company, because it forced him to stop managing like someone else. GMV rose from $61.1 billion in 2019 to $119.6 billion in 2020. On May 6, 2020, Shopify passed the Royal Bank of Canada to become the most valuable public company in Canada. On May 21, 2020, Lütke declared Shopify "digital by default": "Office centricity is over."

Inside, he treated the pandemic as a changed axiom. His reasoning, borrowed from programming, goes like this. Every plan rests on a tree of decisions that starts from assumptions. If one assumption fails, such as people being able to leave their homes, every decision above it is suspect. The only correct response is to prune the tree back to the broken assumption and rederive everything from there.

He reviewed every project himself, in 16-hour days. He cancelled roughly 60% of them and, over the following year, replaced every one of his executives. He is careful to say this was not about bad people. Trust had broken, and a crisis tests people in ways nothing else does: "If everyone is a one before, some people go to zero in a crisis, some people go to a hundred." He says he could not have predicted who would do which.

He found one reliable signal: people who had started a company adapted fastest. Shopify had acquired many startups over the years and kept a Slack channel and an annual offsite for their founders. Lütke went to that channel, told them he needed help, and asked many of them to become executives. He also promoted individual engineers into very large roles. "Every one of those things worked."

His explanation is that founders are "irritants." They do not settle. If something is bad, they say so, even after everyone else has agreed to move on. Most large companies protect themselves from such people by moving them into innovation labs, which Lütke calls "daycare for people who otherwise tell you that your s*** doesn't smell." He did the reverse and put them "right in front of you, or, in fact, on top of you."

The Correction: Deliverr, Layoffs and the Meeting Purge (2022-2023)

Shopify bet that the pandemic had permanently moved commerce online, and in 2022 that bet failed. The stock fell roughly 80% from its late-2021 peak. Growth slowed from 57% in 2021 to 21% in 2022. In July 2022 Shopify laid off about 10% of its staff, and Lütke took the blame in writing: "We bet that the channel mix would permanently leap ahead by five or even 10 years… It's now clear that bet didn't pay off. I got this wrong."

The biggest part of the bet was logistics. In May 2022 Shopify bought Deliverr for about $2.1 billion, its largest acquisition ever, to build a fulfillment network. One year later, in May 2023, it sold Shopify Logistics to Flexport for a 13% equity stake and cut about 20% of staff. It later sold 6 River Systems for a small fraction of its purchase price.

Lütke's own account of the drawdown is unusual. He told Senra he felt relieved when the stock fell. The business was healthy, he had no plan to raise money, and at the peak the stock traded at more than 50 times revenue, "not exactly value investing." What he took seriously was the effect on employees whose stock options were now underwater through no choice of their own.

Flex Comp: letting employees choose

Shopify rebuilt its pay system in September 2022 so every employee chooses their own mix of cash and equity. Each person gets one total number and a set of sliders for cash, restricted stock units and options, and can change the mix every quarter. There is a bonus for choosing more equity and no one-year vesting cliff. Because grants rebalance each quarter at the current price, a falling stock buys more units next time. Shopify reported that 91% of eligible employees enrolled at launch.

Lütke's point is agency. Employees had received options at the peak without any say in the matter, so they reasonably expected the company to make them whole. Flex Comp gives them the choice. If they want the traditional options package, they can build it themselves with the tool. As he put it, "sometimes orthodoxy can come back on the table if you get there from good principles."

The meeting purge

In January 2023 Shopify deleted every recurring meeting with more than two people. COO Kaz Nejatian announced it with the line "meetings are a bug." No-meeting Wednesdays came back, and meetings with more than 50 people were confined to one Thursday window. Press coverage called it the Chaos Monkey approach, after the Netflix tool that breaks systems on purpose to find their weak points. Reports put the removed calendar events in the tens of thousands.

When meetings crept back, Shopify released a calendar extension in July 2023 that shows the estimated cost of a meeting as you schedule it. Its example: a 30-minute meeting with three people costs roughly $700 to $1,600.

The philosophy behind both moves is one Lütke repeats constantly: the best thing a founder can do is subtraction. In his September 2026 interview he said it more bluntly: "You cannot make things better and better by adding stuff. You can't. You must prune. You must rebuild. You must create an end for things."

Shopify by the Numbers

Shopify grew revenue about 56-fold between 2015 and 2025, from $205 million to $11.6 billion. Gross merchandise volume, the value of everything sold through Shopify, grew from $7.7 billion to $378.4 billion over the same decade and passed $1 trillion cumulatively during 2024.

Source: Shopify annual results releases and SEC filings.

The shape of the curve tells the story of this article. Growth slowed after 2021 as the pandemic boom faded and Shopify unwound logistics. Then it reaccelerated to 26%, 26% and 30% while headcount fell. The company that exited 2025 was larger than the one that entered 2022 and had about a third fewer people.

Employee figures from Shopify annual filings; third-party trackers report slightly different numbers.

How Tobi Lütke Runs Shopify: The Company as Code

Tobi Lütke treats a company as a technology that can be engineered, and he rebuilt Shopify's operating model from first principles after COVID. In his words: companies are "social technology," the one socially acceptable reason to spend 14 hours a day on a single pursuit, and a way to "run the counterfactual to the world you see around you" and let the market test it. He also believes nobody knows how to build companies yet: "All companies are terrible, including mine."

The rest of this section describes the specific mechanisms. Most of them come from Lütke's own accounts in the two 2026 interviews, and several have never appeared in a written history of the company.

Shopify OS: the org chart as a desired-state system

After the COVID reset, Lütke started a GitHub repository to describe what Shopify should look like, and then computed it. He and a few colleagues turned titles, levels, span-of-control rules, compensation bands and market data from PDFs into machine-readable configuration. A Python program with a SAT solver takes those agreed constraints and computes the departments, levels and headcount the company should have.

The first version was wrong, which was the point. The exercise exposed how much was undecided. In Lütke's recollection, Shopify had about 8,000 people and about 5,500 distinct job titles, and "senior staff" sat above director in one group and below it in another.

He borrowed the design from software. React and similar tools are desired-state systems: you declare what the page should be, the system compares that with what is, and it takes the minimum steps to close the gap. Shopify OS does the same for the org, and HR becomes the reconciler.

The practical benefit is less politics. When a sales leader asks for 50 more salespeople, the model recomputes and shows what that costs elsewhere. Nobody can agree to a hire "while playing golf" and leave another team to find the budget. The same model gave Shopify a mastery track, so individual contributors in every discipline can earn as much as vice presidents. Many of Shopify's managers were great engineers who wanted to go back to engineering.

Boxes, phases and the review you can preview

Shopify gives teams a problem space with autonomy inside it instead of a list of rules. Lütke says he does not do "corporate baby-proofing." Policies protect against the worst people and cap the best ones. His alternative is to change the environment so the right thing is the intuitive thing, and to define a "box": a problem whose best solution sits somewhere in the dark.

Projects move through phases. In the prototype phase a team explores the box freely. To move to build, they present what they learned: engineering and design leads give the first approval, and Lütke gives the second. At that point the risk transfers from the team to the company, "a trade of accountability for autonomy," and the team regains freedom until the release review.

One detail shows how far the company-as-code idea goes. Teams can run their proposal past an AI trained on all of Lütke's previous reviews before the real meeting, to see what he is likely to say. That saves him time and saves the team a wasted review.

The corporate-raider game and hit pieces on the past

Once a year Lütke pretends Shopify went bankrupt and he bought it at a fire sale. Previous management was "crazy," and he walks in on day one with a to-do list. He asks teams to play the same game. The goal is to beat the sunk-cost fallacy, which he calls "firmware" every human ships with.

He goes further and writes what he calls hit pieces on the past, cataloguing everything wrong with systems he is proud of. He says nostalgia was once written on death certificates in the 1800s and wants to move "nostalgia" into the bad words and "bias" into the good ones, because without bias you have no convictions.

He is careful about how this lands. Once work is merged, it belongs to the company's commons. Shopify does not let people talk about "ownership" of parts of the codebase, only stewardship, "like an open-source project." Criticizing a system is not a verdict on the person who built it.

Five words that change a review

Co-founder Daniel Weinand gave Lütke one piece of advice that changed how he reviews work: start with "for example." Lütke used to walk to a whiteboard and redesign a team's architecture on the spot. The team had just shown him their best work, so they got defensive. Weinand's fix was to put a phrase like "I could think of doing this a couple of other ways. For example…" in front of the same idea. "Now you're on the same side."

Hiring, pay and the life-story interview

Shopify hires for high agency, not credentials. Lütke is a high-school dropout, so demanding degrees "would be rich." The core of the process is a life-story interview that zooms in, minute by minute, on a moment when something went wrong. Many employees were Shopify merchants before they applied.

He refuses to write down the hiring checklist. His reasoning: the people who study a written list most carefully are the people who perform the traits without having them. "If it can't be put in numbers or can't be written as a list, companies hate it. Yet almost everything you have to do to build a great company is those things."

Shopify is also deliberately never the highest offer. People who optimize only for pay do poorly there. Lütke's pitch is compounding skill instead: you will learn faster at Shopify than anywhere else, so your career will be worth more.

Five-person pods and Norman doors

Shopify designs its physical and digital spaces so that good behavior needs no policy. Weinand's rule was that "no one can be more creative than the space around them." Offices were built around pods for about five people, because every team split costs a large drop in productivity and the military arrives at similar numbers. A seventh person in a pod is physically uncomfortable, so no rule is needed. Meeting rooms sit in the dark middle and desks sit at the windows.

Lütke is "fundamentally allergic" to Norman doors, the doors whose handles tell you to pull when you must push. He calls them "the most fundamental design error": an affordance that makes people predictably do the wrong thing. "Product is an abstraction. A product is millions of little details." Even the furniture comes from Shopify stores. If Shopify needs something that no Shopify merchant sells, it asks someone to start that store.

Shopify Toronto, 2017. Photo: Wikimedia Commons, CC0.

The operating system in one table

Shopify in the AI Era (2023-2026)

Shopify went from launching an AI assistant for merchants in 2023 to running much of its own engineering through an AI agent by 2026. The public milestones are Sidekick (2023), the April 2025 AI memo, the ChatGPT checkout partnership (2025), the Universal Commerce Protocol (2026), River (2026) and the Hydrogen and Vercel Shop launches with Vercel (2026). Each one pushed AI deeper, first into the product, then into how employees work, then into how customers buy.

The memo: "Reflexive AI usage is now a baseline expectation"

On April 7, 2025, Lütke published the internal memo that made AI use mandatory at Shopify. He posted it on X himself after hearing it was about to leak. Its most quoted line changed how companies across the industry talk about headcount: "Before asking for more headcount and resources, teams must demonstrate why they cannot get what they want done using AI."

The memo also put AI use into performance reviews and was blunt about opting out: "Stagnation is almost certain, and stagnation is slow-motion failure." It drew criticism at the time. Nine months later Lütke told Senra that in two years people will read it and think he was "saying the sky is blue."

Two months after the memo, Lütke posted that he preferred the term "context engineering" to prompt engineering, because it names the real skill: "the art of providing all the context for the task to be plausibly solvable by the LLM." Andrej Karpathy amplified the idea a week later, and the phrase spread across the industry. Our context engineering guide and the history of prompt engineering cover where it went from there.

River: the AI colleague that only works in public

River is Shopify's internal AI agent, and its defining rule is that it never works in private. It lives in Shopify's Slack. You mention @river in a public channel and it reads code, runs tests, opens pull requests, queries the data warehouse and looks at production traces. If you send River a direct message, it politely declines and offers to open a public channel for the two of you.

Lütke's reason goes back to his apprenticeship. Remote work cost Shopify the office "osmosis learning" that once happened when juniors sat next to seniors. In his May 2026 essay Learning on the Shop floor, he wrote that Shopify wants to be a Lehrwerkstatt at scale: "It's osmosis learning, because it does not require a curriculum, a training plan, or a manager." A new hire can scroll back through River threads to see how senior engineers scope a request before sending their first one. His own #tobi_river channel has more than 100 people watching, reviewing and, in his words, reminding him "how rusty I am."

River also has a personality, which was a deliberate bet. Lütke considers Microsoft's 2023 Bing chatbot, known internally as Sydney, the first truly remarkable chatbot. After Sydney's long conversations went wrong in public, the industry made every assistant polite and patronizing. Shopify chose the opposite: River has a name and a profile picture, is allowed to be sarcastic, and is allowed to tell you when a request is stupid. "People take great glee," Lütke says, when River makes fun of him.

The numbers climbed fast. When Lütke first wrote about River in May 2026, 5,938 employees had used it across 4,450 channels in 30 days, and it opened 1,870 pull requests in one week, about one in eight merged. Shopify Engineering's Under the River post later reported 59,918 sessions in 5,170 channels in 30 days, touching more than 7,000 people, with 3,536 River co-authored pull requests merged. The median session ran 19 minutes with 50 tool calls. By September 2026 Lütke estimated that "up to about 50%" of pull requests start in conversations with River.

Hand-written code has not disappeared. Lütke says it survives at the limits of complexity, in reviews, and above all in state management, "the thing that's really the hardest to get right," which engineers still do by hand before they "vibe the rest around it." Engineers who do write code often run 10 to 50 agent instances at once.

How River remembers: skills and nightly review

River improves without retraining a model, by rewriting its own instructions from the day's work. Memory is kept per person. During off-hours River reviews the conversations it had, what it struggled with, which skills it used and what mistakes it made, and updates its skill files. Lütke calls the process "dreaming." When Shane Parrish compared it to post-training on yourself, Lütke added: "but the result is text files."

That is the key design choice. Improvement lives in readable, reviewable files: skills, conventions and AGENTS.md files in the monorepo. Humans can inspect and correct them, and the next session starts from them. Our history of agent memory explains why durable, inspectable memory is the hard problem for every agent product.

The platform under River

River only worked because Shopify rebuilt its engineering foundations in 2024, before agents were good. That spring Shopify made two unpopular decisions: move everything into one monorepo, called World, and build all environments with Nix so development, CI and production are reproducible. The bet, in the engineering team's own words, was: "Code is going to be increasingly written with AI, and our infrastructure needs to be the substrate for that."

The engineers' main lesson was that agent-friendly is human-friendly. A fragmented repository hides things from agents and new hires alike. Undocumented knowledge cannot be learned by either. Their second lesson was that local agents have a ceiling: if every AI session happens in a private window, only the person at the keyboard learns anything.

        THE AQUIFER PATTERN (Shopify's agent platform, as published)

  ┌───────────────────┐    ┌───────────────────┐    ┌───────────────────┐
  │ SESSION           │    │ HARNESS           │    │ SANDBOX           │
  │ durable identity  │◄──►│ the agent loop:   │───►│ where code runs:  │
  │ append-only log   │    │ read history,     │    │ filesystem, shell │
  │ lives in Postgres │    │ call the model,   │    │ repo, build, test │
  │                   │    │ emit tool intents │    │                   │
  │ MUST SURVIVE      │    │ cheap, disposable │    │ disposable        │
  └───────────────────┘    └───────────────────┘    └───────────────────┘

  "Decouple the brain from the hands."
   Safety          the loop sits outside the rm -rf blast radius
   Replaceability  swap models without touching the sandbox
   Observability   every decision flows through one place

  River = one profile: prompt + skills + sandbox policy + model defaults.
  PR review, research and migration agents = more profiles, not platforms.

Shopify calls the platform Aquifer. River is one "profile" on it: a system prompt, a set of skills, a sandbox policy and model defaults, shipped as a bundle. A PR-review agent is another profile, woken by an external system with no human in the loop. The published principle is that the next agent should be a new profile, not a new platform. Our deep dive on the agent harness covers the same brain-and-hands split across the industry.

Tobi's AI council and the chief of staff

For important decisions, Lütke runs a council of AI experts across several models and keeps the final call for himself. He has a personal AI chief of staff with access to his notes and many company systems, which he messages like a person. For a hard question it creates five or six sub-agents with different roles, such as a data analyst, a paper researcher, a business view and an engineering view.

Each role runs against several frontier models, and the model that synthesizes the results is chosen at random so no single model's bias dominates. The syntheses are then read by the strongest model available. The result arrives as an audio message he listens to in the morning. A council costs him $15 to $20 in tokens and about half an hour, for work that could otherwise take a month.

He is clear about the limit: "It's not as good at judgment… I use it for creating the right environment for judgment." The reason is responsibility. "Machines can't take responsibility," he says, and he calls that "probably the most overlooked thing in the entire stack." The ideal is a human-in-the-loop decision surface, like a trader's Bloomberg terminal: the machine informs, the human decides and answers for it.

Slop grenades: what AI made worse

Lütke says the failure mode of lazy work has flipped from too little output to too much. Inside Shopify people call it lobbing slop grenades: asking an agent for a change, approving a pull request without reading it, and handing it to colleagues to review. The same happens with email. Someone expands three points into a long AI-written message, and the reader has to feed it back into an AI to shrink it. "Why did we invent decompression and recompression?"

He also notices AI phrasing creeping into how people write and talk, which he calls "Claudisms," and admits he uses one of the favorite ones more than he used to. Our explainer on AI slop covers the wider pattern.

Agentic Commerce: When AI Does the Shopping

Agentic commerce means AI assistants find, compare and buy products for shoppers, and Shopify has spent 2025 and 2026 making its merchants the default supply for those assistants. Lütke describes it as a "box" his teams are exploring: how to expose merchant catalogs to AI, how to complete transactions safely, and what information models need to be good personal shoppers.

Three standards now matter, and they are easy to confuse:

The first version of ChatGPT checkout did not last in its original form. By March 2026 OpenAI said Instant Checkout "did not offer the level of flexibility that we aspire to provide" and shifted toward discovery, letting merchants keep their own checkout. The protocol continued.

UCP had a stronger start. Google announced it during Sundar Pichai's keynote at the NRF retail conference with endorsements from Etsy, Wayfair, Target, Walmart, Stripe, Visa, Mastercard and others. Google added cart, catalog and identity-linking capabilities in March 2026 and expanded it across AI Mode, Gemini and YouTube Shopping. With its Winter 2026 Edition, Shopify turned on agent access by default for its merchants, so their stores can appear in AI assistants without setup. Shopify President Harley Finkelstein has been quoted as saying AI-driven order volume grew about fifteenfold after January 2025.

For a merchant, the practical takeaway is that product data now has two audiences: people and the AI shopping assistants that act for them. Clean titles, complete attributes, accurate stock and clear policies are what an agent reads before it recommends a product. You can see the shopper side of this in a ready-made AI shopping assistant agent and a deals-finder agent, and write better catalog copy with e-commerce prompts.

For the protocol background, see our history of the Model Context Protocol and the guide to the best MCP servers. For the browser side of agents acting on the web, see the history of Browserbase.

Shopify and Vercel: The 2026 Partnership for Agent-Ready Stores

In 2026 Shopify and Vercel joined forces on the storefront: v0 can prompt a Shopify store into existence, the two companies rebuilt Shopify's Hydrogen SDK to run on any JavaScript framework, and Vercel Shop gives merchants an open-source Next.js store that people and AI agents can both shop. Shopify keeps the parts that are hardest to get right: the catalog, the cart, payments and checkout. Vercel supplies the framework, the hosting, the AI tooling and a reference store. "Checkouts are ridiculously hard to get right," Lütke said when Vercel Shop launched.

The two companies already shared customers. Vercel says Shopify, Next.js and Vercel power the storefronts of brands such as AG1, Bombas and PAIGE. What was missing was a portable toolkit. Until 2026, Hydrogen was its own framework, built on Remix and later React Router, and Vercel's verdict in June was that it "made headless storefronts easy to ship, but not portable." A headless store is a custom front end on top of Shopify's commerce back end. The 2026 work made that front end something any framework, and any coding agent, can build.

How an AI agent shops a Vercel Shop store

An agent does not have to click through the store: Vercel Shop publishes the store's Universal Commerce Protocol tools over MCP at the merchant's own domain, so the agent can search the catalog, check availability, build a cart and get a checkout link. Shopify supplies the prices and confirms each cart change, and the buyer stays in control of payment in Shopify checkout.

One store serves five kinds of visitors from the same code:

The reference build is vercel.store, Vercel's merch store. It launched on Vercel Shop with limited-edition drops, and its shopping agent helps shoppers find products, add them to a cart and continue to checkout. You can start your own copy with pnpm create next-app@latest my-store --example https://github.com/vercel/shop.

The Third Customer: Lütke and Rauch on agents as buyers

"The Third Customer" was the closing fireside of Vercel Ship 2026 in San Francisco, where Lütke and Rauch talked about building for agents that research, compare and buy products for people. Lütke's case was that agents widen commerce rather than just speed it up: "Agents can help people follow through on things they otherwise wouldn't have had time to do," such as project-based shopping. He did not predict the end of the storefront. "There's still room for exceptional shopping experiences," he said. "The businesses that will struggle are the ones offering an undifferentiated experience."

The same talk showed how far Shopify's engineering has moved toward verification. Per Vercel's recap, Shopify engineers now think of themselves as "verification engineers" who build harnesses and acceptance tests, the company is rewriting its storefront renderer from Ruby to Rust and comparing the output "byte for byte before switching over," and River, the internal Slack agent, "topped the code-deletion leaderboard by a wide margin" at an internal hackathon. Lütke's advice to developers was to "explore the catalog APIs," which reach merchants' products without first recruiting both buyers and sellers. Earlier, at Vercel Ship in New York, Shopify VP of Product Vanessa Lee put the same idea in one line: "Commerce stops being a destination and becomes part of what you build."

Rauch carried the frame into an interview published October 29, 2026, on Delphi's Library of Minds. He said he now asks "are the agents happy" next to questions about customers and internal developer experience, and that Shopify serves entrepreneurs through e-commerce while Vercel serves them through more than e-commerce. His hope for agent shoppers is plain: "Agents should just choose the best products." For a merchant, that puts the weight back on the basics from the section above: accurate titles, attributes, stock and policies.

The idea reaches past commerce. A site that "speaks agent" gives machines a clean way in, not just people. Taskade does this for its own content: every published blog post has a Markdown copy, including this one at /blog/history-of-shopify.md, the site publishes an /llms.txt index, and every paid plan includes access to the Taskade MCP server. For the Vercel side of the story, read our history of Vercel, Next.js and v0, the v0 review and the Taskade Genesis vs. v0 comparison.

Shopify vs. Other Commerce Platforms

Shopify competes with open-source store software, other hosted builders and marketplaces, and it wins most often on ecosystem and checkout. Each alternative has real strengths, and the right choice depends on who runs the store.

Shopify's structural advantage is that its revenue grows with merchant sales, which pushes it to keep the merchant in charge of the customer. That same principle shows up in its agentic-commerce design, where the merchant keeps the order and the customer relationship even when an AI assistant made the sale. For the WordPress side of this story, see our history of WordPress.

The Future: What Tobi Lütke Says Comes Next

Lütke predicts the future by living in other people's relative future and looking around. His September 2026 interview is the clearest statement yet of where he thinks software, work and Shopify are going.

Software that molds itself around you

Lütke's strongest prediction is that software will reshape itself around how each business works. His own computer runs Omarchy, a Linux setup created by his friend David Heinemeier Hansson, which he changes by telling an agent what he wants instead of editing configuration files. During a meeting he needed a screenshot annotation tool, described it by voice with three follow-up requests, open-sourced it that night, and found six outside pull requests on it the next morning. "My computer fulfills wishes."

He then connected it to Shopify directly: "This is directionally where Shopify is going… You are describing how your business runs, and Shopify will mold itself around this." He added that "collaborative multiplayer software is the future." His home AI agent already fixed a power-outage problem by itself, finding the server it ran on, waking it over the network and repairing the Wi-Fi. He learned about it from a voice message after he woke up.

Superintelligence is a city

Lütke argues that we already live inside a superintelligence and that the synthetic kind will arrive quietly. He keeps the illusion that he could fix his own plumbing only because he lives in Toronto, a "super intelligence" of specialists he can call. Society, the city and the community are intelligence far beyond any individual, governed by systems we built over centuries. AI adds more intelligence to that system. When it passes new thresholds, he expects the same reaction the Turing test got: no ticker-tape parades, just a normal day.

That view also explains why he would let a better AI run Shopify in principle, except for one thing: accountability. An AI cannot go to jail. He compares agents that game their goals to Goodhart's law and to companies that overfit to the quarterly stock price. Humans must stay in the loop for choices that matter.

The skills AI makes more valuable: taste and judgment

Lütke says the skills that will matter most in ten years are taste and judgment, which have always been valuable and will now reach their limit. Taste comes from repetitions: the person who sketches a new logo on a napkin has usually spent 30 years designing logos. He tells young people to study the greats, which AI now makes easier because you can ask for a curriculum, and to study systems that lasted.

His examples are specific. The Catholic Church has run for more than a thousand years with only about four layers of management. Double-entry accounting sounds dull until you learn how Venetian traders invented it to solve real problems. His family saying is that "everything is interesting when you understand how it was invented." He also warns that people prefer complex answers because simple ones make bad stories, "this is why Frodo doesn't take the eagles to Mount Doom."

Judgment, in his definition, is finding the best path when no path is obviously best. Intuition is judgment at an instant: you have practiced taste and judgment so often that you can apply them immediately, and it may take you a long time to explain afterward why you were right.

Here he disagrees with the standard account from Daniel Kahneman's research, which says reliable intuition needs a regular environment, practice and quick feedback. Lütke argues that intuition is most valuable exactly where there is no quick feedback, and that he sees "a very high correlation between the right path and the ones that don't have feedback loops attached." Companies drift short-term, he says, because their executives are rational actors in incentive systems built on quarterly results. As Charlie Munger put it: show me the incentive and I'll show you the outcome.

The hard part of strategy, he adds, is not finding a right answer. Even weak management teams find right answers. The hard part is choosing among several good answers, one of which pays off this quarter and four of which do not. The snowboard store was a good answer. Shopify was a better one with no quick payoff.

He is not a reflexive contrarian. When he sees an orthodox solution he is "incredibly suspicious," but in regulated areas such as payments, "the orthodox way of solving a problem is actually the correct way," because the rules often require it.

Beauty, refounding events and the word "yet"

Lütke believes beauty is how intuition talks to you. His theory is that much of the brain's effort goes into visual processing, and after enough practice that system sends judgments back as a feeling that something looks right, faster than reasoning. Early Shopify was written in Ruby's readable "poetry mode," code that reads almost like English to coworkers and still runs on machines.

His favorite picture of progress is the evolution of SpaceX's Raptor engine, where each generation looks simpler than the last. The first version was covered in pipes that later turned out to be unnecessary once 3D printing made a cleaner design possible. "A lot of teams can't move forward by subtraction," he says. His prescription for companies is the refounding event: "A department sometimes needs a refounding event." Start a new version from the top, keep what works, and drop the rest.

Goodhart's law comes up again here. Shopify repeatedly had to correct the instinct to treat every closed store as bad churn. Because Shopify serves entrepreneurs so early, a closed store is often a finished experiment by someone who will try again on Shopify. A metric turned into a goal leads you to overfit.

At home the rule is simpler. His children may not say "I'm not good at this" without adding "yet," and three people in the room will say it if they forget. "You are malleable, the company is malleable, our product is malleable."

Tobi Lütke's reading list

Lütke still calls books the closest thing to cheat codes for real life, and AI has not changed his mind. The only change is that he now trusts recent nonfiction less, because it is "the product of its time," while books that have lasted keep their value. The books he named in 2026:

Asked for the third time what success means to him, his answer was consistent: cultivate skills, and use them to make "products or toys" that make other people's day a little better, or give them more power, motivation or ambition than they would otherwise have.

What Builders Can Learn From Shopify

Shopify's history is a set of reusable decisions, and most of them apply to a five-person team as much as to a company with 7,600 people.

The throughline is that Shopify tries to design environments, not rules. River is the clearest example: nobody is told how to use AI, but everyone can see how the best people use it.

How to Build a Shopify-Style AI Workspace With Taskade

You do not need Shopify's monorepo and agent platform to give a small team a shared AI colleague with memory. Most of what makes River work is structural: agents live where the work happens, their work is visible to the team, they remember context, and a human owns every output. Taskade Genesis packages that structure into a workspace any team can start on the Free plan.

Here is how the River pattern maps onto a Taskade workspace:

1. Put the agent where the team works. AI agents in Taskade live inside shared workspaces next to your projects, so teammates see the same agent, the same conversation and the same results. That is the open-channel rule without building a platform.
2. Give it memory. Agents keep persistent memory and can use your projects as knowledge, so the next request starts from what the team already learned. Our guide to AI agent memory explains the types.
3. Use more than one model. Taskade runs frontier models from top AI labs, and you can set up multi-agent teams so different agents take different roles, like Lütke's council.
4. Let automations execute. Automations connect agents to your tools across 100+ bidirectional integrations, so a decision turns into action and the result writes back to the workspace. Premium integrations such as Gmail, Slack and Google Sheets need a paid plan.
5. Keep a human accountable. Role-based access from Owner to Viewer controls who can change what, and every agent result lands in a project a person reviews.

If you run an online store, start with a working example. The AI e-commerce automation guide shows agent workflows for orders, support and catalog updates, the guide to AI inventory software covers stock tracking, and the e-commerce agent templates give you a starting team. Taskade's direct Shopify connector is built but on hold while its Shopify App Store listing is paused, so today you connect store data through the other integrations and your own projects. The AI apps gallery and Taskade Community have apps you can clone and adapt, and our roundup of free AI app builders compares where to start. Pro costs $10 per month billed annually, or $20 billed monthly. On Free, you earn free AI credits with referrals and invites, and you can upgrade any time. Facts checked on 2026-11-07.

Quo Vadis, Shopify?

Shopify enters its third decade as a software company again, after a detour through warehouses, with agents writing much of its code and assistants doing more of its customers' shopping. The 2025 numbers show a company that grew 30% with fewer people than it had in 2022. The 2026 moves show a company trying to be the default supply behind every AI assistant that sells things.

The open questions are real. Agentic checkout is new, and OpenAI already reworked its first version. Protocols compete. River's success depends on a culture of reading what the agent produces, and "slop grenades" show that the culture can slip. And Lütke himself said that without the AI shift he might not still be CEO, because "there are much better leaders for stable times."

He does not expect stable times. He called 2026, Shopify's 21st year, "the most interesting year" in the company's history, and possibly in "literally everyone's career," because everyone will be measured by how fast they can re-derive what they do. That is the same move he made in 2004 with a snowboard shop, in 2020 with a broken axiom, and in 2026 with an AI colleague in every channel: prune back to what is true, and rebuild from there.

▲ Memory ■ Intelligence ● Execution

Frequently Asked Questions About Shopify

Further Reading

Company Histories

• History of Apple - The last defender of skeuomorphism, according to Lütke
• History of NVIDIA and Jensen Huang - Another founder-CEO who never left
• History of WordPress - The open-source path WooCommerce took
• History of GitHub - Where Shopify's code, and its agents' pull requests, live
• History of Y Combinator - The Silicon Valley playbook Shopify chose not to follow
• History of OpenAI and ChatGPT - The partner behind Instant Checkout
• History of Vercel, Next.js and v0 - Shopify's 2026 storefront partner
• History of Anthropic and Claude - The lab whose phrasing Lütke calls "Claudisms"
• The Complete History of Taskade - Another company built on a workspace idea

AI Agents and How They Work

• What Are AI Agents? - The execution layer behind River
• History of AI Agents - From SHRDLU to the agent loop
• History of the Agent Harness - The software around the model
• History of Agent Memory - Why agents forget, and how skills files help
• Multi-Agent Systems - Councils, teams and roles
• Context Engineering Guide - The term Lütke popularized
• History of the Model Context Protocol - The standard behind Shopify's MCP servers
• AI Slop Explained - The over-output problem, named

Build With AI

• Free AI App Builders - Tools to build your first AI app
• Best AI Website Builders - Store and site builders compared
• AI E-commerce Automation - Agent workflows for online stores
• One-Person Companies - Lütke's "cause more entrepreneurship," in the AI era

Explore Taskade

• Taskade Genesis - One prompt, one living app with data, agents and automations
• AI Agents - Agents with persistent memory in shared workspaces
• Automations - Reliable workflows across 100+ bidirectional integrations
• AI Apps Gallery - Working apps you can open and study
• Taskade Community - Clone real apps other people have built

Sources: Shopify annual results releases and SEC filings (2015-2025); Shopify news releases on funding (2010, 2011, 2013), POS and Payments (2013), Plus (2014), the IPO (2015), Capital (2016), the Deliverr acquisition (2022) and the Flexport sale (2023); Shopify Engineering, "Under the River" (May 2026), "Building the Universal Commerce Protocol" (January 2026) and "The Engineering Story Behind Flex Comp"; Tobi Lütke, "Reflexive AI usage is now a baseline expectation at Shopify" (April 7, 2025) and "Learning on the Shop floor" (May 2026) on X; OpenAI, "Buy it in ChatGPT" (September 2025); Google Developers Blog on UCP (2026); Vercel, "Build and deploy Shopify storefronts on Vercel" (June 4, 2026), "Vercel and Shopify are rebuilding Hydrogen" (June 30, 2026), "Vercel Ship 2026 recap" (October 15, 2026) and "Vercel and Shopify make agents first-class shoppers" (October 16, 2026); Shopify developer changelog, "Hydrogen now deploys to Vercel" (June 30, 2026) and "Hydrogen 2026-10 is now available" (October 14, 2026); Hydrogen developer preview release notes (October 8, 2026); the Shopify DotDev 2026 lineup; Practical Ecommerce, "Shopify, Vercel Arm Stores for AI Agents" (October 20, 2026); Guillermo Rauch on X (June 5, 2026) and on Delphi's Library of Minds (October 29, 2026); Tobi Lütke interviews with David Senra (January 2026) and Shane Parrish, The Knowledge Project (September 2026). Interview quotes are transcribed from the recordings.

Images: Tobi Lütke, 2015, photo by Union Eleven, Wikimedia Commons, CC BY-SA 3.0. Shopify Toronto office photos, Wikimedia Commons, CC0.