Shopify CEO warns AI 'slop grenades' can create more work - what AI builders should learn

Shopify CEO Tobi Lütke has a name for the worst way AI can backfire in an engineering organization: "slop grenades." In a recent interview on The Knowledge Project podcast, he described how AI tools make it easy to produce large volumes of material quickly, but without any built-in check on whether that material is useful or correct. The result is unvetted, overly long outputs that get tossed at colleagues, who then have to clean up or summarize the mess. For AI builders shipping agentic workflows, internal tools, or code generation products, this failure mode is a direct threat to the productivity gains AI promises.

The 'slop grenade' problem every AI-powered team will face

Lütke's specific example is telling: someone writes an unnecessarily long AI-generated email, and the recipient has to use another large language model to summarize it. He calls this "decompression and recompression" of communications. The AI expands a simple thought into verbose text, and then another AI compresses it back down. The net result is negative productivity, not positive.

This problem scales beyond email. At Shopify, the internal AI assistant River handles a large share of production code pull requests, possibly as much as half. When AI-generated code PRs are shipped without proper human evaluation, the review burden shifts to teammates. Lütke's framing cuts through the usual hype: AI is most valuable when it clarifies and concentrates human thinking, and least valuable when it adds noise to someone else's desk.

Why this matters for AI builders

Shopify has been one of the most aggressive adopters of AI in enterprise. Lütke has made AI a "fundamental expectation" for employees and told teams they must prove AI cannot do a job before asking for more resources. So when the CEO of a company that runs on AI agents warns about slop grenades, the signal is worth taking seriously.

For builders shipping AI products into enterprise workflows, the implication is direct: your tool's success depends not just on raw capability but on how well it integrates with human accountability. If your agent or assistant produces long, unvetted outputs that require extensive review, you are creating slop grenades for your users. The product design challenge is to encourage concise, verified outputs and to make the human review step fast and effective.

Practical takeaways: avoiding the decompression trap

The core insight from Lütke is that humans take responsibility. AI can inform better decisions, but it cannot substitute for judgment. In practice, this means:

• Build review into the workflow. If your AI generates code, emails, or documentation, make it easy for the user to verify, edit, and approve before sending. Do not encourage one-click generation followed by blind forwarding.

• Optimize for conciseness. Longer outputs are not better. Design prompts and post-processing to keep results tight. The "decompression and recompression" cycle is a sign that your tool is creating overhead, not eliminating it.

• Track the net time saved. If your AI feature requires a colleague to spend more time reviewing or fixing than it saved the original author, you have shipped a slop grenade generator.

• Maintain human accountability. Lütke's statement "Humans take responsibility. Machines can help us take more responsibility" is a useful design principle. The AI should empower the human to make better decisions, not offload all effort.

Caveats and open questions

The term "slop grenades" is a coined phrase from the interview, not a formal category. There is no independent measurement of how much of Shopify's AI usage falls into this trap. The effectiveness of tools like River depends heavily on team discipline. Lütke himself positions AI as a net positive when used correctly, and Shopify continues to invest heavily in AI agents. The warning is about misuse, not about AI itself.

For builders, the lesson is not to slow AI adoption but to design for human oversight from the start. The companies that treat AI as a tool for better thinking, not just faster output, will avoid the slop grenade trap.