What are 'slop grenades' in the context of AI?

'Slop grenades' refer to unchecked AI-generated content, such as emails or code, that are passed to colleagues without sufficient human review, potentially increasing workload.

How does AI affect productivity according to Tobias Lütke?

AI can make output appear larger without reducing actual work, as it may require colleagues to spend time checking and correcting AI-generated content.

'Workslop' is content that appears legitimate but lacks substance, often requiring additional work to correct, thus reducing productivity.

What is the estimated cost of dealing with 'workslop'?

Research estimates the cost of dealing with 'workslop' to be about $186 per employee each month, leading to significant productivity losses for large companies.

What approach has Shopify taken towards AI use?

Shopify has made AI use a 'baseline expectation' and requires employees to justify why tasks cannot be completed with AI when seeking additional resources.

Shopify Went All-In on AI, But Its CEO Says Productivity Gains May Just Mean Employees Are Passing Work Along

Tobias Lütke says employees must review, fact-check and condense AI-generated emails, memos and code before passing them on

Portrait of Shopify CEO Tobias Lütke — Shopify's CEO warns AI can make staff look more productive while quietly pushing the fact-checking, fixing and rewriting work onto colleagues downstream instead Union Eleven/Wikimedia Commons

Shopify has pushed artificial intelligence deep into its workplace, but chief executive Tobias Lütke says apparent time savings can be misleading when employees use AI to produce work that colleagues then have to check and fix.

Speaking on The Knowledge Project podcast, Lütke warned about what he called 'slop grenades': unchecked AI-generated emails, memos and code passed to other employees without enough human review.

He said some staff were allowing AI to 'go nuts' before handing the results to colleagues. That can make the original worker appear more productive while shifting the real workload downstream.

AI Can Make Output Look Bigger Without Reducing Work

Traditional productivity measures often focus on what an individual produces, such as reports completed, emails sent or code submitted.

Lütke argued those figures can miss the work created for other employees. A report produced in minutes with AI may still require a colleague to spend an hour checking its facts, rebuilding its argument or correcting its code.

He also described cases involving long AI-generated emails. A recipient may use another AI tool to summarise a message that could have been written clearly in the first place. Lütke questioned why organisations were creating what he called 'decompression and recompression' instead of simply communicating clearly.

Lütke has made AI use a 'baseline expectation' at Shopify. Employees seeking additional resources are expected to show why a task cannot be completed with AI. The company also uses an internal AI agent called River, which handles a substantial share of production code pull requests.

Lütke's comments suggest that counting AI-assisted tasks does not necessarily show whether the technology is reducing work across the business.

Research Shows the Cost of AI-Generated Clutter

Research from BetterUp Labs and Stanford's Social Media Lab has examined similar problems under the term 'workslop'.

The researchers define workslop as content that appears legitimate but lacks substance, including polished emails containing broken links or other basic problems. Their 2026 research found that recipients of workslop reported spending an average of 3.4 hours a month dealing with such material from colleagues.

The researchers' 2025 study estimated the cost at about $186 per affected employee each month. Accounting for workslop's prevalence, they estimated roughly $ 9 million in annual lost productivity for a company with 10,000 employees.

The research also found reputational costs. Workers who produce workslop were viewed as less competent and less friendly. Some 36% of recipients said they wanted to avoid working with those colleagues in future.

Duolingo Reported Similar AI Quality Problems

Duolingo has taken a similarly aggressive approach to AI, with chief executive Luis von Ahn introducing an 'AI-first' mandate and replacing some contractors with automated tools. But the company also encountered quality problems. Von Ahn said that generating 1,000 learning stories with AI could leave around 20% as 'pure slop'.

Lütke said accountability must remain with the person using AI. He said employees should review and condense AI-generated material before sending it on.

For companies measuring AI-driven productivity, that means looking beyond output volume and asking whether the technology actually reduces total labour, errors and unnecessary work.