UpTrajectory Review

Tobi Lütke, Shopify's CEO, has drawn a bright line on what AI adoption should actually look like inside a company: if a tool doesn't reduce the total amount of work being done, it isn't working. The rule he circulated to staff is blunt — AI should not function as a way to shift the labor of verification, cleanup, and correction onto someone else. That framing cuts against how most AI deployments have actually unfolded over the past two years, where the pattern has been a junior employee generating a draft and a senior employee spending forty minutes fixing it, or a customer-service chatbot deflecting a query until a human has to untangle the mess. Lütke is naming something that many operators have felt but few executives have said out loud: the efficiency gains being promised are often just cost transfers dressed up as innovation.

For a small-business owner, this rule is more useful than any AI vendor pitch you will hear this year. Most small teams don't have the slack to absorb hidden rework — if your bookkeeper is using AI to categorize transactions and you end up auditing the output every month, you haven't saved time, you've created a second job. The practical test Lütke is proposing is one any operator can apply this week: pick one workflow where you've introduced an AI tool and ask whether the total hours spent by everyone touching that task have gone down, not just whether the person using the tool feels faster. If the answer is no, the tool is performing efficiency, not delivering it.

What makes this notable is that it comes from a CEO whose company sells tools to merchants, many of whom are small businesses themselves. Shopify has a direct commercial interest in its customers adopting AI-powered features, and Lütke is still telling his own staff to be skeptical of AI that creates downstream work. That gives the rule more credibility than similar statements from consultants or academics. Where we'd push slightly back is on feasibility: in a small operation, some verification burden is unavoidable when you adopt any new system, AI or otherwise. The question isn't whether anyone has to check the output — it's whether that checking is a temporary onboarding cost or a permanent structural feature of the workflow.

The second-order effect worth watching is how this reframes the build-versus-buy calculation for business software. If the real cost of an AI tool includes the senior hours spent reviewing its output, then the cheaper subscription that produces messier results may actually be the more expensive choice. This also has implications for hiring: teams that adopt AI carelessly may find they still need the same headcount, just redistributed toward quality control rather than production. Lütke's rule, applied honestly, could mean slower AI adoption in the short term and more durable gains in the long term — which is a harder sell to a board but a better outcome for the business.

The action item is straightforward: audit one AI-assisted workflow this month and measure total time spent across everyone involved, not just the primary user. If the tool is generating rework, either tighten the prompts and inputs, restrict it to lower-stakes tasks, or drop it. And when a vendor demos something new, ask them directly who catches the errors their system makes — if the answer is 'your team,' price that in.

“AI should reduce the total amount of work, not transfer the burden of checking and correcting it to someone else.” — Inc. Magazine

Takeaway: Audit one AI workflow this month: if total hours across your team haven't dropped, the tool is shifting work, not saving it.