Tobias Lütke

The pattern they actually represent

In 2004 Lütke was selling snowboards online. He was unhappy with the shopping software available to him, so he wrote his own. Two years later he made the decision the entire company now rests on: the snowboards were not the business. The cart was. He threw away the thing that already worked and relaunched what had been infrastructure as the product.

That is not a story about spotting an opportunity. Plenty of people notice that their internal tool is better than their business. What is rare is being willing to delete the business, and the habit that makes it possible is the one he names directly.

Subtraction is genuinely harder than addition, and not for sentimental reasons. Adding is legible — you can point at it, ship it, put it in an update. Removing looks from the outside like having done nothing, or worse, like admitting the removed thing was a mistake. So organisations accumulate. The snowboard business, the feature nobody uses, the process that made sense two years ago.

The second half of the pattern is what makes the first half survivable: he refuses to treat a failed bet as waste. In his framing a mistake is only wasted if you fail to extract what it taught, which turns a failure into instrumentation rather than a verdict.

Put those together and you get his actual operating move, repeated at every scale of his career: a public, written, whole-company bet. Ruby on Rails in 2004 when almost nobody had heard of it. The pandemic expansion. Reflexive AI in 2025, announced as a baseline expectation rather than an initiative. Each time the thesis goes in writing first, and the memo itself becomes the commitment device — much harder to quietly abandon a bet you have already published.

The blockage it speaks to

Attachment to Code is the obvious one, and it is not really about code.

It is about the specific paralysis that sets in once you have built something. The work is done, it functions, it has your fingerprints on it — and that makes it almost impossible to evaluate on its merits. You end up defending it instead of assessing it. The question silently changes from "is this the right thing to have" into "was I wrong to have made it", and nobody answers the second question honestly about their own work.

So the codebase keeps the abstraction nobody needs. The company keeps the product line. You keep the service you stopped wanting to sell eighteen months ago, because stopping would mean the eighteen months were wrong.

Lütke's answer is to make removal the default rather than the exception. If subtraction is the normal move, deleting something stops being a confession. It is just Tuesday. That single reframe is what lets him look at a working snowboard business and ask what it is actually for, without the question feeling like an attack on the person who built it.

His anti-consensus streak does the same job from the other direction. Treating industry best practice as a way of avoiding risk rather than as accumulated wisdom means the existing thing never gets to defend itself just by already existing.

Three moves you can steal

Do a removal pass before an addition pass. Before the next feature, service line or process, spend the same amount of time asking what could be deleted. Most people only ever run the addition review, which is why every system they own grows monotonically and nothing is ever assessed twice.

Write the bet down before you make it. Not the plan — the thesis, in a paragraph, with what you expect to be true if it works. Publishing it internally does two things: it makes the bet real enough to act on, and it gives you something specific to be wrong about later, which is the only condition under which being wrong teaches anything.

Separate "was this a mistake" from "what did it teach". Answer the second question in writing before you let yourself answer the first. Lütke's own 2022 memo does exactly this: it states the bet, states plainly that he got it wrong, and describes what the company now believes instead — in that order, in public.

Where the pattern breaks

Fast, high-conviction, publicly-committed bets carry no hedge, and the record shows what that costs when one resolves against you.

Shopify bet that pandemic buying behaviour was permanent and expanded against it. On 26 July 2022 it cut around ten per cent of staff, with Lütke's memo saying plainly that the call was his and that he got it wrong. On 4 May 2023 a further twenty per cent went, along with the entire logistics business, sold to Flexport.

The pattern's speed is real, and so is the correction: two rounds in ten months, landing on people who did not place the bet. If you are running it, the hedge has to be deliberate, because the method will not supply one.

What to do this week

Pick the thing you have built that you would defend hardest if someone questioned it. That reaction is the signal — you are not defending its usefulness, you are defending your having made it.

Now write two lines about it. What it was supposed to produce, and the last time it actually did. Do not decide anything yet; just get both lines onto the page, because most of this work is simply making a thing evaluable again after your name has been attached to it for a year.

Then, before you add anything at all next week, delete one thing. Something small and genuinely unnecessary. The point is not the saving. The point is proving to yourself that removal is available as a normal move, so it is there the day the question is a much larger one.