Shopify's Reflexive AI Memo: How Product Teams Apply It in 2027

In April 2025, Shopify CEO Tobi Lütke published an internal memo that made AI use a job requirement rather than a suggestion. Product leaders still bring it up in 2027, often while deciding whether to run their own version. This guide covers what the memo said, sourced to the press coverage that reproduced it. It also gives you templates for the memo, the headcount request, and review questions. We at BuildBetter work with B2B product teams on customer evidence, so the guide also covers where that evidence fits once AI makes prototypes cheap. Everything here is sized for a team that has a month and no Shopify-scale budget.

The Story: Shopify Made Reflexive AI Use a Baseline Expectation

Shopify CEO Tobi Lütke wrote an internal memo declaring that “Reflexive AI usage is now a baseline expectation at Shopify,” as reported by Digital Commerce 360.

• When: BetaKit reports the memo was dated March 20, 2025. According to CNBC via NBC San Diego, Lütke posted it publicly on X on Monday, April 7, 2025.
• Why public: BetaKit reports he shared it because the memo was in the process of being leaked.
• The headline rule: Before asking for more headcount and resources, “teams must demonstrate why they cannot get what they want done using AI” (BetaKit).
• Scale context: Shopify had about 8,100 employees at the end of December 2024, down from about 8,300 a year earlier (CNBC via NBC San Diego). Both figures predate the memo, so the memo did not cause that change.

The sections below cover what the memo requires, the questions Shopify answered, and how a smaller product team can apply each idea this month.

What Shopify Actually Did: The Mechanics of the Memo

The Shopify memo is a policy change: it rewired resourcing, evaluation, and the product process at the same time. Teams that copy its tone without changing those systems tend to end up with a memo nobody follows. Here is what it changed.

• Universal expectation. “Using AI effectively is now a fundamental expectation of everyone at Shopify” (BetaKit).
• Headcount approval. Teams must show why AI cannot do the work before they request headcount or resources (BetaKit).
• Agent-based planning. The memo asks: “What would this area look like if autonomous AI agents were already part of the team?” (CNBC via NBC San Diego).
• Evaluation. AI usage questions will be added to performance and peer review questionnaires (BetaKit).
• Product process. AI exploration should dominate prototyping for GSD projects. GSD is Shopify's internal “Get Shit Done” project tracking (BetaKit).

The memo also escalated an earlier message. Lütke said his previous call for employees to “tinker” with AI was “too much of a suggestion,” and the memo turns it into a requirement (BetaKit). He stated the rationale directly: “Stagnation is almost certain, and stagnation is slow-motion failure” (BetaKit).

The Questions Shopify Had to Answer

Any AI-first policy has to answer five questions: who it applies to, how it's measured, how it affects hiring, where it starts, and whether leaders are bound by it. Shopify's memo answered each one.

Is AI use optional or expected for every employee?

Expected. Lütke wrote, “Frankly, I don't think it's feasible to opt out of learning the skill of applying AI in your craft” (BetaKit).

How does AI use show up in performance reviews?

Through AI usage questions added to performance and peer review questionnaires (BetaKit). The exact questions have not been published.

When can a team ask for more headcount?

Only after it demonstrates why it cannot get the work done using AI (BetaKit). This is a justification step. It is not a hiring freeze.

Where in the product process should AI be used first?

Prototyping. BetaKit reports that the prototype phase of GSD projects should be dominated by AI exploration.

Does the expectation apply to leadership too?

Yes. Lütke wrote, “This goes for me as well as everyone else” (BetaKit).

Where Can You Read the Full Text of Tobi Lutke's AI Memo?

Tobi Lütke posted the full memo on X on April 7, 2025. BetaKit, CNBC (syndicated by NBC San Diego), and Digital Commerce 360 reproduced or quoted it extensively.

One caveat: we could not fetch the original X post directly. Every quote in this guide comes from press coverage that reproduced the memo. These are the core lines we verified:

“Reflexive AI usage is now a baseline expectation at Shopify.” (Digital Commerce 360; also BetaKit, CNBC)

“Using AI effectively is now a fundamental expectation of everyone at Shopify.” (BetaKit)

Teams “must demonstrate why they cannot get what they want done using AI” before asking for more headcount and resources. (BetaKit)

“What would this area look like if autonomous AI agents were already part of the team?” (CNBC via NBC San Diego)

“Frankly, I don't think it's feasible to opt out of learning the skill of applying AI in your craft.” (BetaKit)

“Stagnation is almost certain, and stagnation is slow-motion failure.” (BetaKit)

“This goes for me as well as everyone else.” (BetaKit)

This month: Save the original post and at least two press reproductions to your team wiki. When you reference the memo internally, cite the original.

When you write your own AI policy, quote primary sources rather than summaries. Policies built on paraphrases drift. Someone eventually says “Shopify banned hiring,” which the memo never said, and the policy debate ends up chasing a claim nobody made.

What Does 'Reflexive AI Usage as a Baseline Expectation' Mean in Practice?

Reflexive AI usage means AI is the default first move on any task, and choosing not to use it is the exception. Shopify treats it as a skill no one can opt out of learning.

Shopify applied the expectation to everyone, the CEO included. It also replaced the earlier, softer call to “tinker.” That second point is where many internal AI programs go wrong. Encouragement produces enthusiasts, and a baseline produces habits.

The difference comes down to two words:

• Baseline: expected of everyone, discussed in reviews, and visible in how work gets done.
• Aspiration: encouraged, celebrated when it happens, and optional in practice.

Define “reflexive” in terms your team can check. One working definition: before you start a draft, analysis, spec, or prototype, ask whether an AI pass could produce a first version. You can observe whether that happened, which you can't do with “be AI-curious.”

This month:

1. List your team's 10 most repeated tasks, such as release notes, competitive summaries, interview synthesis, or spec drafts.
2. Assign each task an “AI first pass” owner.
3. Run a two-week trial.
4. For each task, record whether the AI output was usable as-is, needed editing, or failed.

At the end of the trial you will have a short, evidence-based map of where AI earns its place on your team. You'll need that map for the headcount and review steps later in this guide.

How Do You Write an AI-First Memo for Your Company?

An effective AI-first memo does five things: it states the expectation in one sentence, explains why, defines how it will be measured, changes at least one resourcing rule, and names leadership explicitly.

As reported, Shopify's memo had six parts:

• A clear expectation statement
• A rationale (stagnation as “slow-motion failure”)
• A resourcing rule (prove AI can't do it)
• A measurement change (review questions)
• A process change (AI-dominated prototyping)
• Leadership accountability

AI-First Memo Skeleton (Fill in the Blanks)

1. Expectation: “Starting [date], using AI as the first pass on [drafts / analysis / specs / prototypes] is a baseline expectation for everyone on [team/company].”
2. Why now: “We are making this change because [specific competitive or customer reason]. If we keep working the way we do today, [concrete consequence].”
3. What changes in resourcing: “Headcount and tool requests will include a short record of what we tried with AI and where it fell short. Template: [link].”
4. What changes in reviews: “Our next review cycle adds [two] questions about how you applied AI and what improved. We will evaluate outcomes and judgment, not usage volume.”
5. Where to start: “Begin with prototyping and early exploration on [project type].”
6. Support and training: “You'll have [tools], [hours per week] of learning time, and [peer demo cadence].”
7. Leadership commitment: “This applies to me and every leader. Each of us will share one AI workflow we use by [date].”

This month: Draft the memo and test it with three managers for clarity. Add an FAQ that addresses job security directly, because if you don't answer that question, people will answer it for you. Then set a 30-day check-in.

Tone: Be direct and explain your reasoning. Lütke's memo carries urgency, but it also gives a rationale. A bare mandate gets compliance at best.

How Does a 'Prove AI Can't Do the Job Before Hiring' Policy Work?

A “prove AI can't do the job” policy requires a team to document what it tried with AI, and why that falls short, before it requests headcount or resources. Shopify's version requires teams to “demonstrate why they cannot get what they want done using AI” (BetaKit). It pairs that rule with a planning question: “What would this area look like if autonomous AI agents were already part of the team?”

The rule shifts the burden of proof. It works best as a documented experiment. Without guardrails, it turns into a veto that managers use to stall hiring.

One-Page AI-First Headcount Request Template

• The work to be done: outcomes, volume, and frequency. Describe the work, not the job title.
• AI approaches attempted: tools, prompts, agents, or automations tried, with dates.
• Results and failure points: where the output met the quality bar and where it broke down.
• Residual human work: what still needs a person, such as judgment, relationships, or accountability.
• Risk if unstaffed: customer, revenue, or compliance impact, backed by evidence where possible.
• Requested role: scope, level, and how this person will use AI in the role.

Fairness Guardrails

• Time-box the AI trial to one or two weeks, so it can't be extended indefinitely.
• Define the quality bar before the trial starts, so nobody moves the goalposts afterward.
• Recognize work that depends on judgment or relationships, such as enterprise account management, user research, and incident command.

This month: Apply the template to your next open request. Have a peer reproduce the AI attempt independently, and record the rationale for the decision either way. Keep in mind that self-reported speedups are unreliable. A July 2025 randomized controlled trial by METR found that experienced open-source developers using early-2025 AI tools took about 19% longer to finish tasks, while believing AI had sped them up by about 20%. Measure the output, not how the trial felt.

How Do You Add AI Usage to Performance Reviews?

Add a small number of questions about how someone applied AI to their work, and evaluate outcomes and judgment rather than raw usage. Shopify is adding AI usage questions to its performance and peer review questionnaires (BetaKit).

Shopify's exact questions have not been published in the sources we verified, so we won't guess at them. The prompts below are our suggestions:

• Self-review (suggested): “Describe one workflow you changed with AI and what improved.”
• Peer review (suggested): “Has this person shared AI techniques that helped the team?”
• Manager review (suggested): “Give an example of this person deciding not to use AI, and whether that call was sound.”

The third prompt is there on purpose. If you measure usage alone, people will perform it. Prompt counts, logins, and token spend are easy to game and correlate poorly with value. Someone who runs 400 prompts a week and ships nothing should not outscore someone who rebuilt the team's interview synthesis process with three well-designed prompts.

This month:

1. Add two questions to the next review cycle. Don't add five.
2. Hold a 30-minute manager calibration session with real examples of “strong,” “adequate,” and “weak” answers.
3. Write down explicitly that prompt counts and tool logins are not evaluation criteria.

Weight the evaluation toward quality of outcomes and good judgment about when AI fits. That way the review rewards the skill the memo was asking for.

How Do You Make AI Adoption Mandatory Without Losing Your Team?

Mandatory AI adoption works when you state the requirement explicitly, apply it to leaders first, give people time and tools to learn, and tie it to real work instead of abstract targets.

Shopify did this by saying out loud that “tinker” had been “too much of a suggestion,” then making AI use a requirement that covered the CEO: “This goes for me as well as everyone else.” For small teams, including leadership is the easiest part to copy. It costs nothing, and it shows the rule isn't a cost-cutting tool aimed only at individual contributors.

Four-Week Rollout

• Week 1: Leaders share their own AI workflows, including what failed.
• Week 2: Each team picks one real workflow to convert.
• Week 3: Peer demos of 15 minutes per team, showing the before and after.
• Week 4: Write the policy into review and headcount templates.

Common Failure Modes

• Mandates without training. People get told to use AI, then get no time to learn it.
• Tool sprawl. Every team buys its own assistant, and none of them share context.
• Unaddressed fear of replacement. If nobody talks about it, it shapes how people react to everything else in the rollout.

Workflow integration decides whether the mandate sticks. MIT NANDA's August 2025 report, The GenAI Divide, found that about 95% of the enterprise generative AI pilots it studied showed no measurable P&L impact. The methodology has been debated, so treat it as one data point. Still, it supports a pattern many teams have seen: AI adoption that isn't tied to specific workflows and success measures tends to stall.

None of the sources we verified report outcomes from Shopify's mandate. Set your own success measures before rollout, such as cycle time on specs, time from customer request to prototype, or review-cycle quality.

How Should Product Teams Use AI in Prototyping?

Product teams should make AI exploration the main mode of early prototyping, using it to generate and test many options before committing engineering time. Shopify's memo says GSD prototyping should be dominated by AI exploration (BetaKit). GSD is Shopify's internal “Get Shit Done” project tracking.

Once prototypes are cheap, building stops being the bottleneck. A PM can produce three clickable variations of a settings page before lunch. The harder question is which of those three, if any, solves a problem customers actually have. Choosing becomes the scarce skill.

This month:

1. Pull the top recurring customer problems from the last 60–90 days of calls and tickets.
2. Generate two or three AI prototypes per problem.
3. Show them to the customers who raised the problem, ideally on a call you record.
4. Kill or advance each prototype based on what customers do with it, not on how much the team likes it.

Anti-pattern to avoid: generating more prototypes faster from internal opinions alone. That gets you guesses at a higher speed. The memo's prototyping rule works best when every prototype traces back to customer evidence.

What to Copy From Shopify and What Not to Copy

Copy the structure of the Shopify memo, adapt the measurement details, and don't assume its scale or results transfer to your company.

Copy

• The explicit baseline expectation
• Leadership inclusion
• The headcount justification step
• AI-first prototyping

Adapt

• Review questions. Shopify's wording isn't public. Write your own and weight them toward outcomes.

Don't Copy Blindly

• Scale. Shopify had about 8,100 employees at the end of 2024 (CNBC via NBC San Diego). A 40-person team has fewer people to absorb experiments and less room for policy overhead. Shrink every step accordingly.
• Assumed results. The memo states intentions and policies. The sources we verified report no measured productivity results, so we make no claims about outcomes.
• Headcount causation. Don't attribute the drop from 8,300 to 8,100 to the memo. Both figures predate the March 2025 memo date.

Reminder: we could not access the original X post. The quotes in this guide come from press reproductions.

Where Customer Evidence Fits in an AI-First Product Process

AI makes prototypes cheap, so the scarce input in an AI-first product process is knowing which prototype customers need. If your AI prototyping isn't grounded in real call and ticket evidence, faster prototypes turn into faster guesses.

The evidence usually exists, but it's spread out. Sales has call recordings. Support has tickets. CS has Slack threads with customers. Research has interview notes. Nobody has them all in one place, and none of it is tagged by severity or business impact.

BuildBetter brings customer calls, Slack conversations, support tickets, and surveys together through 100+ integrations, including Zoom, Slack, Jira, Salesforce, Zendesk, HubSpot, and Intercom. That gives AI prototyping a shared evidence base. Each piece of feedback is analyzed individually for intent, severity, and business impact, using your own product taxonomy. BuildBetter also combines that external feedback with internal team activity, so you can see what customers asked for alongside what the team has already discussed or committed to.

BuildBetter then produces PRDs and Linear or Jira tickets from that evidence. A customer conversation becomes a prototype brief with the original quotes attached. The path runs from customer conversation to shipped decision.

The same evidence helps with the headcount rule. Compare two requests:

• Opinion-based: “We need another PM for onboarding.”
• Evidence-based: “Onboarding friction showed up in 41 calls and 120 tickets last quarter, across accounts worth $X in ARR. AI handled synthesis and first-draft specs. Discovery calls and stakeholder alignment still need a person.”

The second request is harder to dismiss and easier to approve.

Tools That Support an AI-First Product Workflow

An AI-first product workflow needs two layers: a customer evidence layer to decide what to build, and a prototyping layer to build it quickly.

1. BuildBetter — Best for Customer Evidence Behind AI-First Decisions

BuildBetter captures calls through a bot, through local recording without a bot, or on mobile. It combines internal and external feedback in one place and auto-generates PRDs, tickets, and summaries. It also includes rapid prototyping that captures any page in your app so you can iterate on it with voice and prompts. Pricing is usage-based with unlimited seats, so your whole org can use it without per-seat math. It is SOC 2 Type II certified and HIPAA-ready. Honest caveat: it isn't built for distributing enterprise surveys at scale or for dedicated research-repository tagging workflows.

2. v0 by Vercel

v0 generates UI from text prompts. It's useful for exploring interface directions before anyone opens a design file.

3. Lovable

Lovable produces functional app prototypes from natural-language descriptions. It works well for putting something clickable in front of a customer.

4. Bolt.new

Bolt.new is a browser-based full-stack prototyping environment, so non-engineers can build working demos without setting up a local environment.

5. Cursor

Cursor is an AI-assisted code editor that engineers use to turn a validated prototype into working code.

FAQ: Shopify's Reflexive AI Memo

When was the Shopify AI memo written and published?

The memo was dated March 20, 2025 (BetaKit). Tobi Lütke posted it publicly on X on April 7, 2025 (CNBC via NBC San Diego), reportedly because it was in the process of being leaked.

What does 'reflexive AI usage' mean?

It means using AI by default as the first move on a task, so that not using it is the exception. Shopify calls it a “baseline expectation” and says it isn't feasible to opt out of learning the skill.

Do Shopify teams have to prove AI can't do a job before hiring?

Yes. Teams must demonstrate why they cannot get what they want done using AI before asking for more headcount or resources (BetaKit). It is a justification step, not a hiring ban.

Does Shopify measure AI use in performance reviews?

Shopify said it would add AI usage questions to its performance and peer review questionnaires (BetaKit). The exact questions have not been published.

Does the AI expectation apply to Shopify's CEO?

Yes. Lütke wrote, “This goes for me as well as everyone else.”

How can a small product team apply the memo?

Start with AI-first prototyping grounded in customer evidence. A tool like BuildBetter can surface the top recurring problems from calls and tickets. Then add a one-page headcount justification template and two outcome-focused review questions. You can run all three within a month.

Sources and Further Reading

• BetaKit: Shopify CEO Tobi Lütke tells employees to prove AI can't do the job before asking for resources
• CNBC via NBC San Diego: Shopify CEO says staffers need to prove jobs can't be done by AI before asking for more headcount
• Digital Commerce 360: Internal memo: Shopify CEO declares AI non-optional
• Wikipedia: Shopify
• Wikipedia: Tobi Lütke
• METR, randomized controlled trial on AI and experienced open-source developer productivity (July 2025)
• MIT NANDA, The GenAI Divide: State of AI in Business 2025 (August 2025)
• BuildBetter: buildbetter.ai

Note: The original X post was not directly accessible to us. All memo quotes come from the press reproductions listed above.

Make Churn Optional

An AI-first mandate gets your team prototyping faster. Deciding which prototypes are worth building still depends on customer evidence. BuildBetter turns every call, ticket, Slack thread, and survey into PRDs, tickets, and customer follow-ups that keep your AI-first process tied to what customers actually asked for.

Make churn optional. Book a demo.