The technology leaders who once touted the use of AI as the key to unlocking the full potential of every employee are now changing their minds.
This includes Shopify Co-founder and CEO Tobias Lütke. He told employees last year that using AI is “a fundamental expectation” and they should first prove that they “can’t achieve what they want with AI” before asking for more resources. He now says Shopify employees create unverified emails and code with AI and take no responsibility for the sloppy output.
“We call these ‘slop grenades’ that people throw at each other,” he said during an interview The knowledge project Podcast on Tuesday. “This is definitely a bad thing.”
Lütke said it’s easy to make AI “crazy,” but misusing it creates even more headaches for the people receiving and evaluating the work. For example, he suggested that the use of AI should help summarize points in an email, rather than turning it into a “big missive” that wastes time.
“You don’t really read it, and now it has to be reviewed by your colleagues and they say, ‘That doesn’t look right,'” he said. “You just let the AI do the work for you.”
Luis Von Ahn, CEO of Duolingo, has also backed down. He announced last year The company would put “AI first,” meaning employees would be evaluated on their AI usage, human contractors would be replaced with AI, and headcount would only be increased if a team cannot automate the work required. But in May he told Fast company that he got carried away by the AI demonstrations as he wrote, but said they ultimately wouldn’t match the creativity of Duolingo’s people when scaled.
“Maybe we have to write 1,000 different stories for people to learn a language, then you’ll find that 20% of the things were just nonsense,” he said. “Whenever we scale a lot [of] When it comes to things with AI, we have to be really careful that no dirt gets through.”
The rise of “Workslop”
Researchers have coined the term “workslop” for the phenomenon Lütke describes: polished-looking AI output that ultimately hurts productivity because it needs to be reworked.
BetterUp Labs and Stanford’s Social Media Lab questioned This year, 962 full-time American desk workers found that more than half (52.7%) reported sending worklops to colleagues, and that this was more common among people whose organizations promoted the use of AI. Over a third (38%) said they received Workslop, estimating that revisions cost an average of 3.4 hours per month.
What sets Workslop apart from shoddy human work is that Workslop looks legitimate at first glance, but it lacks the components that would make it useful. Examples of work errors cited by survey participants included well-structured emails with broken links or code that was more complicated than it needed to be.
The survey found that relationships are also affected when workslop is suspected. Employees said they viewed their colleagues who sent the message as less competent and less friendly. Of those who received Workslop, over a third (36%) also said they wanted to avoid working with these colleagues in the future.
The cleaning time of 3.4 hours is an increase from last year’s survey, as 40% reported I came across Workslop and said they needed to spend two hours revising it. The figure associated with Workslop’s 2025 overhaul was $186 per month for individual employees and up to $9 million per year in lost productivity for a company with 10,000 employees.