# Shopify CEO Warns AI ‘Slop Grenades’ Shift Work To Coworkers > Source: > Published: 2026-09-21 15:14:46+00:00 Shopify CEO Tobi Lütke [says](https://fs.blog/knowledge-project-podcast/tobi-lutke-3/) AI has produced a new kind of lazy work at his company, where employees generate too much and leave coworkers to check it. Lütke made the comments on The Knowledge Project podcast, released Sept. 15. He said Shopify has an internal name for AI output that gets handed to a colleague without being read. Staff call it a slop grenade. The comments come roughly 17 months after Lütke [shared with employees](https://x.com/tobi/status/1909251946235437514) that using AI was a baseline expectation at Shopify. In the same interview, he mentioned that up to about half of the company’s pull requests, those are the code changes engineers submit for review, are generated from conversations in a company chat, where an internal AI agent helps out. On Sept. 3, I [shared data from freelance marketplaces](https://www.searchenginejournal.com/ai-slop-cleanup-job-listings-report/588235/) indicating increasing demand for individuals who correct AI output. Lütke explains this process as checking work within a company, where the cost is covered by a coworker’s time rather than an external bill. ## What Lütke Said Host Shane Parrish inquired if AI had made anything worse within Shopify. Lütke responded around the 16-minute mark of the episode. “The failure case now of lazy work is not lack of output,” Lütke [said](https://fs.blog/knowledge-project-podcast/tobi-lutke-3/). He explained the issue as over-output and shared two examples. The first is about an employee asking an AI agent for a code change, approving the pull request without reading it closely, and leaving colleagues to review it. The second example shows someone using a language model to turn a short point into a long email, which the recipient then shortens again with another model. “So we call those ‘slop grenades’ that people toss at each other,” he [said](https://fs.blog/knowledge-project-podcast/tobi-lutke-3/). Lütke’s advice on the email was to use the model to make a point shorter, not longer. Earlier in the interview, he [said](https://fs.blog/knowledge-project-podcast/tobi-lutke-3/) machines can’t take responsibility for work and people do. He doesn’t say the term is exclusively his. In a [post on X](https://x.com/tobi/status/2099862302467969232), Lütke credited Harry Brundage for coming up with it and mentioned he hopes it becomes popular. ## Shopify Made AI Use A Baseline Expectation In April 2025, Lütke posted an internal memo that said “reflexive AI usage is now a baseline expectation at Shopify.” Roger Montti [covered the memo](https://www.searchenginejournal.com/shopify-memo-marks-pivotal-moment-for-ai-workplace-usage/543840/) for Search Engine Journal at the time. The memo [said](https://x.com/tobi/status/1909251946235437514) AI use would be part of peer and performance reviews. Teams asking for more headcount would first have to show why AI couldn’t do the work. That October, a Shopify engineer [wrote](https://www.shopify.com/news/unserious-exploration) that the company had reached “universal adoption of AI code editors,” with thousands of Cursor licenses and unlimited access to top AI models for every team. On the podcast, Lütke described River, an AI agent that works in Shopify’s Slack channels and can open pull requests. He [estimated](https://fs.blog/knowledge-project-podcast/tobi-lutke-3/) that up to about half of Shopify’s pull requests are created from conversations in the company’s chat, where River operates. He dedicated much of the interview to explaining how he personally uses AI, including a team of agents he consults to debate both sides of a difficult decision. ## Review Work Is Showing Up Elsewhere The demand for freelancers who correct AI-generated work is rising across three platforms, as per internal platform data shared by [the Guardian](https://www.theguardian.com/technology/2026/sep/02/ai-jobs-freelance-cleanup) on September 2. On Freelancer.com, listings tagged with phrases like “correct AI” and “AI hallucination” grew by 87%, reaching 10,760 worldwide from August 2025 to June 2026. Upwork data indicated a 70% year-over-year surge in AI remediation gigs. Additionally, Fiverr searches for “AI cleanup” services increased over 20-fold from 2023 to 2026, [according to the report](https://www.theguardian.com/technology/2026/sep/02/ai-jobs-freelance-cleanup). BetterUp, which offers workplace coaching, ran a September 2025 survey with the Stanford Social Media Lab that explored in-house activities. The survey involved over 1,000 full-time U.S. desk workers. About four in 10 respondents [shared](https://www.betterup.com/workslop) that they experienced “workslop,” AI-generated work that appears polished but doesn’t really advance their tasks, in the past month. On average, each instance took nearly two hours to resolve. All this information is based on self-reported data. ## Why This Matters Shopify [reports](https://www.shopify.com/news/unserious-exploration) universal adoption of AI code editors, yet its CEO still calls unreviewed output the thing AI made worse. The adoption rate doesn’t indicate whether the work was saved or shifted to the recipient. For content and marketing teams using AI for initial drafts, the time editors spend reviewing these drafts is part of the overall cost. ## Looking Ahead The 2025 memo set expectations for how AI should be used, while Shopify’s October 2025 post [reported](https://www.shopify.com/news/unserious-exploration) that AI adoption had become widespread. However, neither source provides a specific estimate of how much time employees are spending reviewing AI output. The freelancers in [that Guardian report](https://www.theguardian.com/technology/2026/sep/02/ai-jobs-freelance-cleanup) disagreed on how long cleanup work will last. One writer expected much of it to dry up within five to 10 years as models improve. A designer predicted AI would match her logo work within a couple of years. *Featured Image: beast01/Shutterstock*