Shopify's CEO made AI use a condition for getting more people
Tobi Lütke published the memo himself because it was leaking. The line everyone quoted was about reflexive AI usage. The line that actually changes how a company runs was about headcount.
Tobi Lütke
CEO, Shopify
An Authority Report is where we read what people with real operating responsibility said in public, in their own words, and work out what it means for a company that isn’t them.
In April 2025 Tobi Lütke published an internal memo himself, because it was in the process of leaking. The headline everyone repeated was about AI use becoming an expectation. The sentence that actually changes how an organisation behaves was three paragraphs down, and it was about hiring.
What the memo says
The title was the thesis:
Reflexive AI usage is now a baseline expectation at Shopify.
Reflexive is the operative word. Not “encouraged”, not “available”: the expectation is that reaching for AI is the first move rather than a considered choice, in the way that reaching for a search engine is.
Two mechanisms carry it. The first is that AI use became part of performance and peer review, which turns a stated value into something with consequences. The second is the one worth reading twice, and it is the headcount rule: before a team can ask for more people and resources, it has to demonstrate why it cannot get what it wants done using AI.
That is not an efficiency programme. It is a change to the internal budget process, and budget processes are where companies actually decide things.
Why the headcount gate is the interesting half
Most AI adoption programmes fail in the same way: the tools are provided, a training session happens, usage peaks in week two and decays. Nothing in the organisation depends on the tool, so nothing changes.
The headcount rule attaches AI to something a manager already wants intensely. Nobody has to be persuaded to care about a hiring requisition. By making the AI question part of that conversation, Lütke put the burden of proof where it produces work: on the person asking for resources, at the moment they most want them.
Read plainly, the rule has teeth for managers rather than for individual contributors. The person who has to answer “why can’t AI do this” is the one with a budget, not the one doing the task. That is what separates it from the usual staff-facing adoption push.
The part that is easy to get wrong
The rule is a gate, not a substitute. “Demonstrate why you cannot get this done using AI” is answerable, and in most cases the honest answer is a good one: the work needs judgement, or a relationship, or accountability that a system cannot hold.
Where this design fails is when the gate becomes a formality in the other direction: when saying “we tried AI” is enough, or when it becomes politically impossible to hire and teams quietly ship worse work instead. Nothing in the public memo prevents that, and it is the thing to watch in any company that copies it.
The version of this that a smaller company can actually run is narrower and better: not a hiring gate, but a default question in one specific recurring decision. Before adding a person to support, before outsourcing a repetitive process, before buying the next seat-based tool, one honest look at whether the work is the sort a system does well.
What this means at 40 people instead of 8,000
Three things carry over, and one does not.
The one that does not: the performance-review mechanism. At forty people, adding AI usage to reviews reads as surveillance and produces theatre. The behaviour you get is people demonstrating tool use, not people getting work done differently.
Tie the question to a decision that already exists. The memo works because it lives inside a process managers care about. Find the equivalent in your company (the hiring conversation, the budget round, the annual tool renewal) and put the question there rather than in a separate initiative.
Make “no” a legitimate answer. A gate that only opens one way is not a gate, it is a mandate, and mandates produce compliance rather than results. The value comes from the cases where the answer is genuinely “a person has to do this”, because that answer is now written down.
Expect the honest answer to be “our process isn’t written down”. In most mid-sized companies the first attempt at this question fails not because AI can’t do the work but because nobody can describe the work precisely enough to tell. That discovery is worth more than the automation it eventually produces, and it is the same finding behind why most AI pilots produce nothing measurable.
The read
The memo is usually filed under “CEO says use AI”. That reading loses what is actually novel about it.
What Lütke did was refuse to run AI adoption as a programme. There is no rollout plan in the memo, no centre of excellence, no phased pilot. There is a change to how one existing decision gets made, and then the organisation figures out the rest under that constraint.
That is a defensible bet, and it is roughly the opposite of what most companies do, which is to run a pilot, declare it a success, and change no decision at all. Whether Shopify’s version produces better outcomes or just fewer hires is a question the numbers will answer over years. The design principle survives the answer either way.
An expectation that no existing decision depends on is not an expectation.
Frequently asked questions
What exactly did the Shopify memo require?
That AI use is a baseline expectation for everyone, that it becomes part of performance and peer review, and that teams demonstrate why AI cannot do the work before requesting additional headcount and resources.
Why did Lütke publish an internal memo?
He said it was in the process of being leaked and would likely be shown in bad faith, so he posted it in full himself.
Is a headcount gate like this legal in Germany?
The hiring decision itself is a management prerogative. Evaluating individual employees on their AI usage is a different matter and, where it involves systematic monitoring, generally engages works council co-determination under § 87 BetrVG.
Does this mean Shopify stopped hiring?
The memo does not say that. It changes the burden of proof for a request, not the possibility of one.
What is the smallest useful version of this?
One question, attached to one decision that already happens. Not a programme, not a review criterion, and not a mandate, a question with a legitimate “no” available.
Photo: Tobi Lütke, by Union Eleven, CC BY-SA 3.0, via Wikimedia Commons. The memo was published by Lütke on his own account; the headcount provision was reported by CNBC.
Strategy Call
