# Slop Grenade

> The workplace anti-pattern of copying a coworker's question into an LLM and pasting the raw, unread output back at them instead of an actual answer -- a thoughtless lob of unverified AI text that the recipient now has to clean up.

Source: https://adipod.ai/glossary/slop-grenade/
Related episodes: 27

---
## Context

The term comes from [noslopgrenade.com](https://noslopgrenade.com), which made the rounds in May 2026 and got covered in Dan's rant on [Episode 27](/episodes/27-openai-beats-musk-gemini-3-5-flash-and-ai-burnout-mitigation/) of the ADI Pod. The scenario is specific: a coworker asks you why you chose a particular architecture four months ago, and instead of answering, you paste their question into a codebase with Claude running and lob the raw model output straight back at them. You haven't read it, you haven't summarized it, you haven't added your judgment — you've thrown a slop grenade.

The site frames itself as the spiritual successor to [nohello.com](https://nohello.com), the long-running etiquette PSA against sending a bare "Hi" in a DM and then making the other person wait for the actual question. Both name a small, common rudeness that the medium makes easy: nohello externalizes the cost of starting a conversation; the slop grenade externalizes the cost of reading and verifying AI output.

## Why It Matters

The slop grenade is an etiquette problem that doubles as a trust problem. The whole point of asking a human is that you want their judgment — their one-line take, their reasoning, the "plus my professional opinion" subtext that an LLM dump strips out. Pasting unread output says, in effect, "I valued your question so little that I wouldn't even read the answer before forwarding it." Worse, the recipient now inherits the verification burden the sender skipped: they have to figure out whether the wall of generated text is correct, which is often more work than answering from scratch would have been.

It has a sibling failure mode the episode flagged: the "let me Claude that for you" move, where the asker could have pointed the model at the repo themselves — the AI-era descendant of "let me Google that for you." Both are about who does the work of turning a tool's output into an answer.

The fix is not "never use AI to help respond." It is to apply the same courtesy that already governed the medium: lead with your one-line take and a couple lines of reasoning, then attach the full output for anyone who wants the complete kaboom — and flag that it's AI-generated. On the show, Shimin's proposed UX patch was a chat-client toggle for "show my AI research," analogous to a model's show-thinking view, so the take and the supporting generation are visibly separated rather than mashed together.

## Related Concepts

- [Verification debt](/glossary/verification-debt) -- the slop grenade is verification debt handed to someone else: the sender skips the review, the recipient pays it
- [Cognitive surrender](/glossary/cognitive-surrender) -- the decision-making failure mode behind the grenade, where the human stops adding judgment and just relays model output
- [Benchmaxxed](/glossary/benchmaxxed) -- another case of plausible-looking AI output standing in for the real thing