The executive who made AI use mandatory is now reckoning with the messy side effects of unchecked AI-generated output
Tobias Lütke spent the last year telling Shopify employees that AI wasn’t optional. Now he’s dealing with the consequences of everyone actually listening.
The Shopify cofounder and CEO has coined a new term for what happens when workers lean on AI without bothering to review the output: “slop grenades.” These are unexamined emails, code snippets, and documents that land in a colleague’s inbox or codebase, shifting the burden of quality control to whoever happens to open them next.
From mandate to mess #
In April 2025, Lütke issued what amounted to an internal edict: AI use would be treated as a “baseline expectation” at Shopify. Teams were told to prove they couldn’t accomplish a task using AI before requesting additional headcount or resources. AI proficiency was woven into performance evaluations.
Roughly 50% of production pull requests at Shopify now originate from interactions with River, the company’s internal AI agent, across somewhere between 7,000 and 10,000 Slack channels. Speaking on The Knowledge Project podcast, Lütke described a new failure mode that his company hadn’t anticipated when it went all-in on AI. Workers were generating outputs at a pace that outstripped anyone’s ability to meaningfully review them.
Lütke put it bluntly: machines don’t take responsibility. When an AI-generated code commit introduces a bug, or an AI-drafted email miscommunicates a key detail, the AI doesn’t own the fallout. A human does. And if that human never properly reviewed the output before shipping it, the accountability chain breaks down entirely.
AI, tech, and the markets they move—in one daily briefing.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
The overproduction problem #
The “slop grenade” framing is deliberately vivid. A grenade doesn’t just affect the person who throws it. It creates a blast radius. When one employee fires off an AI-generated document without reviewing it, the colleague who receives it has to spend time figuring out whether it’s accurate, whether it’s complete, and whether any of its claims are hallucinated. The labor doesn’t disappear. It just moves.
This dynamic is particularly acute in software development, where Shopify’s 50% pull request figure becomes both impressive and alarming. Code that ships without proper human review can introduce subtle bugs that compound over time, creating technical debt that dwarfs whatever productivity gains the AI delivered in the first place.
A wider reckoning for tech leadership #
The Shopify story is worth watching because the company went further than most in making AI adoption mandatory. By tying AI proficiency to performance reviews and requiring teams to exhaust AI options before hiring, Lütke created powerful incentives to use the tools. What he didn’t build, at least not quickly enough, were equally powerful incentives to maintain quality standards while doing so.
Lütke appears to be recalibrating rather than retreating. The mandate to use AI hasn’t been rescinded. But the expectation that employees take ownership of what the AI produces is now being stated with equal force.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our