A manager spends the morning correcting a teammate’s 12-page market research report, finding a competitor’s name wrong, statistics without sources and repeated conclusions. Even if AI takes 20 minutes to produce a report, checking and revising it can take more than 2 hours. As AI shortens writing time, review work shifts to colleagues and managers.

Shopify made AI use a basic expectation for all employees in April 2025. CEO Tobi Lütke said employees seeking more headcount or budget must first show why AI cannot solve the problem. In a September 2026 podcast, Lütke introduced River, a chat-based AI agent (software that carries out tasks on a user’s behalf). River modifies code and checks systems at employees’ request, and handles half of all code changes across Shopify. Alongside this adoption, the piece describes a problem: employees pass unverified AI output to colleagues. It calls this a “slop grenade”—a burden that can pass unnoticed by the writer but leave the recipient with substantial review work.

BetterUp Labs and Stanford’s Social Media Lab surveyed 1,150 full-time U.S. office workers in September 2025. They defined “workslop” as AI output that looks plausible but lacks substance. Forty percent said they had received workslop from a colleague in the previous month, and checking and correcting one item took an average of 1 hour 56 minutes. Among recipients, 53% felt fatigue and irritation, while 42% questioned the colleague’s competence and reliability. The piece says a company with 10,000 employees loses about $9 million in productivity each year to review delays and friction. Because reports move up to managers, the burden can concentrate on team leads and executives; one example describes five employees each submitting an unverified report.

The piece attributes the problem partly to organizational performance measures, not only individual carelessness. When employees are judged by output volume, such as the number of reports or lines of code, they may be pushed to produce more quickly instead of checking carefully. Writers may not see the cost when a manager spends 2 hours fact-checking a report they wrote in 20 minutes. Lütke criticized the practice of expanding a one-line idea into 10 pages with AI, then having the recipient summarize it again with AI. If asked for a source, a writer cannot simply say “AI wrote it”; the person responsible for the work must put their name to it and take responsibility, the piece says.

The proposed rules are: writers should not send a document unless they have read and understood it from beginning to end; reports should open with a “four-line briefing” covering the purpose, which parts AI drafted and which the writer analyzed, facts and sources checked firsthand, and unverified claims or risks; and AI performance should be assessed not just by the number of items produced but also by supplementary measures such as rejections and rework due to factual errors or poor quality. Marking unverified areas can help a manager focus the review and cut the time from 2 hours to 15 minutes. If too much remains unverified, the report is not ready for approval.

The piece says Lütke is not opposing AI adoption, but calling for the people who produce work to be accountable for it. AI has made writing and coding dozens of times faster, while people’s speed at verifying truth and understanding context has not changed. The piece concludes that unless workplace rules assign responsibility for bridging that speed gap, review work may fall heavily on people making important decisions.

A manager spends the morning correcting a teammate’s 12-page market research report, finding a competitor’s name wrong, statistics without sources and repeated conclusions. Even if AI takes 20 minutes to produce a report, checking and revising it can take more than 2 hours. As AI shortens writing time, review work shifts to colleagues and managers.

Shopify made AI use a basic expectation for all employees in April 2025. CEO Tobi Lütke said employees seeking more headcount or budget must first show why AI cannot solve the problem. In a September 2026 podcast, Lütke introduced River, a chat-based AI agent (software that carries out tasks on a user’s behalf). River modifies code and checks systems at employees’ request, and handles half of all code changes across Shopify. Alongside this adoption, the piece describes a problem: employees pass unverified AI output to colleagues. It calls this a “slop grenade”—a burden that can pass unnoticed by the writer but leave the recipient with substantial review work.

BetterUp Labs and Stanford’s Social Media Lab surveyed 1,150 full-time U.S. office workers in September 2025. They defined “workslop” as AI output that looks plausible but lacks substance. Forty percent said they had received workslop from a colleague in the previous month, and checking and correcting one item took an average of 1 hour 56 minutes. Among recipients, 53% felt fatigue and irritation, while 42% questioned the colleague’s competence and reliability. The piece says a company with 10,000 employees loses about $9 million in productivity each year to review delays and friction. Because reports move up to managers, the burden can concentrate on team leads and executives; one example describes five employees each submitting an unverified report.

The piece attributes the problem partly to organizational performance measures, not only individual carelessness. When employees are judged by output volume, such as the number of reports or lines of code, they may be pushed to produce more quickly instead of checking carefully. Writers may not see the cost when a manager spends 2 hours fact-checking a report they wrote in 20 minutes. Lütke criticized the practice of expanding a one-line idea into 10 pages with AI, then having the recipient summarize it again with AI. If asked for a source, a writer cannot simply say “AI wrote it”; the person responsible for the work must put their name to it and take responsibility, the piece says.

The proposed rules are: writers should not send a document unless they have read and understood it from beginning to end; reports should open with a “four-line briefing” covering the purpose, which parts AI drafted and which the writer analyzed, facts and sources checked firsthand, and unverified claims or risks; and AI performance should be assessed not just by the number of items produced but also by supplementary measures such as rejections and rework due to factual errors or poor quality. Marking unverified areas can help a manager focus the review and cut the time from 2 hours to 15 minutes. If too much remains unverified, the report is not ready for approval.

The piece says Lütke is not opposing AI adoption, but calling for the people who produce work to be accountable for it. AI has made writing and coding dozens of times faster, while people’s speed at verifying truth and understanding context has not changed. The piece concludes that unless workplace rules assign responsibility for bridging that speed gap, review work may fall heavily on people making important decisions.