Shopify CEO Tobi Lütke has a blunt warning for any team using AI as a productivity shortcut: volume without verification is just expensive busywork. On a recent podcast, he described how employees are creating more output than ever and handing the review burden to whoever receives it.
What Lütke said
Appearing on The Knowledge Project, Lütke said AI has produced a new form of lazy work inside Shopify. The old problem was no output. The new problem is over-output that someone else must clean up.
As Lütke put it:
The failure case now of lazy work is not lack of output.
Shopify calls these unchecked handoffs slop grenades. Lütke gave two examples. A worker asks an internal AI agent for a code change, approves the pull request without reading it closely, and leaves colleagues to catch issues. Another employee uses a language model to expand a short point into a long email, which the recipient then shortens with another model.
Lütke credits Harry Brundage for the term. His advice? Use a model to make a point shorter, not longer. He also said machines cannot take responsibility for work — people can.
Why it matters for marketing teams
Shopify made reflexive AI use a baseline expectation in April 2025, and by October 2025 an engineer said the company had universal adoption of AI code editors. Still, Lütke says unreviewed output is the thing AI made worse. The metric that looks like adoption may actually be task-shifting.
Lütke estimated that up to half of Shopify's pull requests now begin as conversations in company chat, where an internal AI agent called River can open pull requests. That is impressive scale — but it does not prove the work got faster. Somebody still has to review the code.
The same economics show up in marketing. AI drafts ad copy, product descriptions, SEO briefs, campaign summaries and client emails faster than a human. But if a senior marketer spends an hour fixing hallucinations, brand-voice issues or fluff, the speed was an illusion.
Freelance marketplaces are already seeing demand for human cleanup:
- Freelancer.com listings tied to correct AI and AI hallucination grew 87%, reaching 10,760 from August 2025 to June 2026.
- Upwork reported a 70% year-over-year jump in AI remediation projects.
- Fiverr searches for AI cleanup rose more than 20-fold from 2023 to 2026.
A BetterUp survey with the Stanford Social Media Lab found that about four in ten U.S. desk workers had encountered workslop — polished-looking AI output that does not move the task forward — in the previous month. Each instance took nearly two hours to resolve, based on self-reported data.
Stop tossing slop
Treat AI output as a first draft, not a deliverable. If you generate it, you own the review. Before you hand anything to a teammate, read it, verify the key facts, and cut the fluff. For email, follow Lütke's rule: compress, do not expand.
One useful framework for marketing teams: add an owner review step before any AI-assisted asset enters a shared workflow. If the creator cannot explain the reasoning behind the copy, the audience likely will not feel it either. Speed belongs at the idea stage; judgment belongs before the send.
Shopify has not published a specific estimate of how much time employees now spend verifying AI output. But the trend should make every growth team question what AI adoption really measures. If your team ships more drafts and your reviewers are burned out, you have not removed work — you have moved it.
Source: Search Engine Journal
