Shopify Chief Executive Officer Tobi Lutke disclosed that artificial intelligence invented a new source of lazy work in the firm, which involves workers creating content at a higher rate than they can review and later passing the unchecked material to a colleague to sort out. Shopify refers to this practice as a “slop grenade.”
What Lutke Actually Said
Lutke made the comments on the podcast, The Knowledge Project, which was published on September 15. The host Shane Parrish asked him to what extent AI made a worse thing – lazy work. He did not mince words: “The failure case now of lazy work is not lack of output.”
It is a turnaround of the problem that has been discussed for the last several years. This issue was previously associated with the unwillingness to do the work, to deliver the results to the level of quality or just to perform certain functions half-heartedly. But according to Lutke, failure in laziness now is when a person does too much and does not stop to analyze whether any of this is correct and whether it actually needs to be done. Instead, he sends this to another person, expecting them to handle the trash that was given to them.
He gave two examples of such situations. In the first case, an employee asks an AI agent to make some changes to the code, after which he approvingly sends further the pull request without looking at it. An illustrative example was given by Lutke himself when a person used a language model to expand the short explanation email to a longer one, after which the recipient began to shorten it again with the help of another AI model.
So we call those slop grenades that people toss at each other,” Lutke said. His advice was to use the model to shorten instead of expanding emails. Earlier in the same interview, the entrepreneur warned that responsibility cannot be delegated to machines, but only to people.
Lutke is not the owner of the phrase slop grenade, as he wrote in his X (Twitter) message, crediting Shopify’s Harry Brundage with coining the phrase. He hopes that the term will become more widespread.
Shopify Set the Stage for This Years Ago
The slop grenade comment lands almost a year and a half after he told Shopify employees that having everyone on AI was not optional. In an internal memo in April 2025, he stated that the reflexive AI usage was a baseline expectation. The memo was not only motivational rather the utilization of AI was in peer and performance reviews and teams requesting more hires had to have a pitch as to why AI was not able to do the work.
By October 2025, a Shopify Engineer said the entire company was using AI code editors with thousands of licenses for Cursor, unlimited access to the best AI models on every team.
On the podcast, Lutke discussed River, an AI agent that runs internally within Shopify’s Slack messaging system and is capable of submitting pull requests independently. He said that he thinks that currently as much as half of the pull requests at Shopify code changes submitted by engineers for review come from chat conversations with River. He also described how he himself uses an “entire team of AIs” to debate both sides of a complex decision in his mind before making a decision.
Combine those two bits of data and the slop grenade remark looks less like raising alarm bells about some fringe concern and more like an explanation of some of the costs of the policy Lutke wanted the broad adoption of AI. Yes, in this version of the world, AI was driving us to higher production, but it also appears to have created a new category of unreviewed work that has to go somewhere, and that somewhere is often a coworker’s inbox.
The Same Pattern Is Showing Up Everywhere
Shopify is not an isolated case. The Guardian reported on September 2, that using internal data from three freelance websites, demand is rising quickly for people whose job is fixing up AI-generated work. On Freelancer. com, the number of listings with the term correct ai or AI hallucination rose 87% from August 2025 to June 2026, with over 10,700 listings in total across the globe. Upwork said there was a 70% year on year increase in AI remediation jobs over the same period and searches for AI cleanup services on Fiverr increased more than twenty times over from 2023 to 2026.
A short survey brings this number even higher. A September 2025 survey by workplace coach BetterUp, conducted with Stanford’s Social Media Lab, polled over a thousand U. S. -based full-time, desk workers. Roughly 4 in ten of them reported experiencing “workslop” in the previous month: work generated by AI that appears done but doesn’t actually advance the ball each time. The average time to fix each incident, self reported, was just shy of two hours.
Mapping those numbers to Lutke’s comments, and suddenly a pattern emerges. Be it a coworker turning back a weak AI draft in the trenches, or a contractor brought on to office-sweep AI output, the same issue arises. You can generate things faster and cheaper, but then you have to check that they’re accurate, and that process hasn’t gotten any faster.
Why This Matters Beyond Shopify
Based on Shopify’s own reports, the company has adopted AI code editors widely, and yet it is something that its CEO alone has called worse than before. Adoption figures don’t tell us whether that saving of time was real, or whether they’ve just outsourced the task to someone else to revisit and verify.
That distinction is important for any team that would be relying on AI to get the first draft of something, whether that is code, marketing copy, or a long email. The time required to develop a first draft has definitely declined. The time spent by a human to review the draft for quality, completeness, and applicability is a different cost that is not present in the same place and is easy to underestimate if no one is tracking it explicitly.
There’s no official figure in either the original 2025 memo or in the subsequent report on AI adoption for what proportion of time people spend checking the AI’s work. That’s the vital piece missing from most lengthy discussions of how much of an uplift the AI has provided: it’s straightforward to see how much output teams are cranking out, and report on it regularly, but it’s tricky to quantify verification time.
Is This a Temporary Problem
There is no consensus as to how long this cleanup phase should last. Within the group of freelance workers surveyed by the Guardian, one writer expected most AI cleanup demand to disappear within five to ten years as the underlying models keep improving, given the steady stream of improvements to the fundamental models. Then again, a freelance designer from the same independent economy had a gloomier outlook for her specific niche, expecting that AI would be capable of thinking at her level of logo design work within just a few years.
Lutke’s framing of it here suggests he doesn’t think this is a passing bug better models will just reset. His concern is about the burden, not the ability. Machines will never be responsible for their output, so someone even so has to validate it, regardless of how perfect it appears. The name Shopify has given the issue a grenade, thrown one desk to the next presents a version of a problem models aren’t going to address simply by becoming more powerful.
FAQs
Q 1: What is a slop grenade?
It’s Shopify’s internal name for AI generated output that an employee passes to a co-worker without first reviewing it to check, shifting the work of checking onto someone else.
Q 2: Who do they credit with coining the term slop grenade?
Shopify CEO Tobi Lutke credited his colleague Harry Brundage with it, and he hopes it catches on.
Q 3: What did Lutke have to say about real failure cases with AI at Shopify?
The failure case now is that there simply is too much unreviewed output getting passed on to other people, who may have no reason to expect it to be high quality.
Q 4: What did he say were some examples of slop grenades?
An employee approving an AI generated code change without reading it carefully, and someone expanding a short point of text into a long email using AI, which is then reduced back down with another AI by the recipient.
Q 5: When did they make its use mandatory?
Lutke announced to his employees in an April 2025 message that using AI was a baseline, and would factor into performance reviews and hiring decisions going forward.
Q 6: What is River at Shopify?
It’s an internal AI agent that works inside of Shopify’s Slack channels and can open pull requests, generating up to about half of the company’s pull requests, according to Lutke.
Q 7: But isn’t this a generic problem for everyone?
Yes, as evidenced by the fact that in the freelance marketplace, companies seem to be seeking out people to clean up after AI all the time, on sites like Freelancer, com Upwork, and Fiverr, and a separate survey found that many desk workers encounter low quality AI output regularly at work.
Q 8: What’s workslop?
It’s a term coined by Researchers at coaching app BetterUp and Stanford’s Social Media Lab that refers to AI generated works that are presented as polished and complete but ultimately don’t advance the task in any way, requiring additional editing time.
Q 9: How much time does a workslop normally take to fix?
Each such occurrence, according to the BetterUp and Stanford survey, took an average of nearly two hours to remedy.
Q 10: Does Lutke see an end in sight for this problem as techniques improve?
No, because he sees it as a responsibility problem, not a capability one, and there’s ultimately no way to hold a machine accountable for something, which means humans will always need to review its output, regardless of how advanced the model is.
