Someone drops a 1,400-word AI response into the Slack thread. It has six headers, three bullet lists, and opens with "Certainly! Here's a comprehensive overview..." Nobody reads it. The question that prompted it needed two sentences.
This behavior - pasting raw AI output directly into human conversations - now has a name: the slop grenade. A sharp post at noslopgrenade.com makes the case against it, and the argument lands because most of us have been on both ends of it.
The piece isn't anti-AI. It's about the difference between using AI to help you think versus using it to avoid thinking. When you paste an unedited model response into a conversation, you're telling the other person: I didn't read this, I'm not sure what the relevant part is, and I've decided your time is cheaper than my effort.
The False Authority Problem
AI text carries borrowed credibility. Six headers and 800 words implies someone worked hard on this. The reader has to decide whether to trust that impression or spend their own time verifying it. Most take the easiest path: ignore it and move on.
Models like ChatGPT and Claude are trained to sound comprehensive, not concise. They hedge because they're designed to cover all scenarios. They add caveats because hallucination risk is real. The output that comes back is almost never what you'd write yourself - it's a first draft optimized for completeness, not for your specific situation or audience.
For people doing knowledge work, this becomes a credibility problem over time. When teammates or clients can't tell where you've added judgment versus where you just hit send on raw output, your actual thinking becomes invisible. They can't distinguish your expertise from a free API call.
The 90-Word Rewrite
The fix requires no new tools. Read the output. Pull out the two or three sentences that actually answer the question. Add context only you have. Delete the hedging language the model put in because its training data demanded caution. Cut the sections that apply to everyone and therefore no one specific.
That takes five minutes. A 90-word message that directly answers a question is more useful than a 900-word document that forces the reader to do the extraction work you avoided.
One observation the piece makes that cuts deep: when you pass AI output to someone else without editing it, you're not saving them time. You're offloading your cognitive work onto them. That's not a productivity gain for the team - it's a cost transfer.
AI tools are most useful as a draft or thinking aid. The gap between what a model outputs and what you should actually send is where your value as a communicator lives. The slop grenade lands in that gap and wastes it.