Lazy work used to mean too little output. Now, with AI, it often means too much, and more work for everyone else. Shopify CEO Tobi Lütke calls that a slop grenade in a clip Shane Parrish shared from The Knowledge Project.
What is a slop grenade?
A slop grenade is when you let AI produce the work and pass it on without adding value, including checking it. Someone else has to wade through it, catch the mistakes, and clean up. You save time and look productive. They pay for it.
The work looks finished. A brief comes back as a long memo. A research ask turns into five tabs of summary. A first pass shows up as a full draft with headings and a conclusion, in a voice that isn't yours.
The problem isn't that the model wrote something. The problem is hitting send before you did any work on it.
Tobi's examples are familiar. A developer tells an agent to go nuts, rubber-stamps the pull request, and leaves the review for a teammate. Or someone uses a model to turn a short point into a long email, and the person who gets it pastes the email into another model just to recover the point. He asks why anyone invented decompression and recompression. If you're already using a model, use it to say the thing in fewer words.
noslopgrenade.com puts it another way: if someone wanted an AI essay, they could have asked the model themselves. They asked you because they wanted your judgment. A wall of text in Slack or email also breaks the medium. Nobody writes essays there. There's nothing left to reply to, push back on, or clarify.
Why this is new
Lazy used to look like missed deadlines and thin notes. AI changed the signal. A long document can look like care. Speed can look like ownership. Volume can look like effort.
The person who threw it often feels busy. They prompted, pasted, and forwarded. Their week is full of drafts they "shared." The person who gets it still has to do the real work, except now it's buried under confident paragraphs.
Before you send
Ask yourself: if they take this at face value, do they get hurt? If yes, you're holding a slop grenade.
Other signs:
- You haven't read the whole thing.
- You couldn't defend a claim if someone asked where it came from.
- You wrote "thoughts?" because you don't have any yet.
- The draft is longer than the decision it supports.
- Names, dates, numbers, or quotes look specific, and you haven't checked them.
If you'd be embarrassed to put your name on it, don't put someone else's day under it.
When one lands on you
They often arrive with cover language: "quick first pass," "for reaction," "I had AI take a swing." Treat that as a warning, not a finished task.
Look for the holes. The sentence that sounds right and isn't. The recommendation that ignores a constraint you actually have. The claim that would be expensive if it were true.
Send it back with a short list of what's missing. Don't rewrite the whole thing if they still own the work. Cleaning it up for them teaches the team that throwing these works.
What to do instead
Open the links. Check the numbers and names. Cut anything the model invented.
Most AI drafts are longer than they need to be. Delete the throat-clearing and keep the claim.
Say what you think should happen, and what you wouldn't do. A draft with no point of view is still a prompt.
Put your name on what you send. If you won't stand next to it, it isn't ready.
Team habits that help
You won't fix this with a setting. You fix it with what the team accepts.
Don't send "thoughts?" on something you haven't read. If you didn't review it, say that and keep it.
Label AI-assisted work honestly. "Model draft, I haven't checked sources" is fine. "Here is the plan" is not, if you haven't made a plan.
The sender owns the first cleanup. Reviewers should edit judgment, not padding.
If the ask was a recommendation, send a recommendation, not a white paper.
If you lead the team, say this out loud. People copy what gets praised. Praise volume and you'll get more slop grenades.
Common questions
What is a slop grenade? AI-generated work you pass on without reviewing it or adding your own judgment, leaving someone else to catch the mistakes and clean it up. You save time and look productive; they pay for it.
Where does the term slop grenade come from? Shopify CEO Tobi Lütke used it on The Knowledge Project podcast with Shane Parrish to describe forwarding unchecked AI output that creates more work for everyone downstream.
How do you avoid throwing a slop grenade? Read the whole thing, check every name, number, quote, and link, cut anything the model invented, and add a clear point of view before you send. If you won't put your name on it, it isn't ready.
AI made it cheap to produce work that looks done. It didn't make it cheap to be right or useful. If cleanup is getting expensive on your team, that's usually process, not the model. We help organizations keep the work clear when the tools get noisy. If you want a second set of eyes, start a conversation.
