Slop Grenadiers

This week I prompted AI to create a research doc.

It delivered in minutes. Sixteen sections, four appendices, an evaluation rubric, 15 open questions, and a recommendation that boils down to "Here's why the hypothesis is right." Every citation points to a real paper. It would survive a skim from almost anyone, and that's the problem.

This is exactly what are known as slop grenades: AI output that someone lobs into a channel, an inbox, or a review, leaving the people around them to hold it. In the last year I've caught plenty. I've also, admittedly, thrown a few of my own and I'll get to those.

I once received a doc not dissimilar to the one that faked for this example. It was shared with no additional context as to what I was to do with it, and the only thing that prefaced the doc was a slight disagreement with my engineer about why more design options wasn't always a better way to start building.

Being the recipient I was confused and frustrated by the expectations of having to read such a lengthy document, not to mention the intent - it was as if the sole purpose was trying to be prove me wrong. It's hard to meet that with grace or thankfulness - and it also didn't help that I received little to no sympathy or resolution from my direct manager when relaying the interaction, but that's a different story.

This is the reality in todays environment as everyone is overwhelmed trying to best figure out how to utilize AI in their daily workflows.

I've put together some thoughts, from experience, that I think would benefit many orgs and individuals alike as they work to understand this new paradigm of AI, and collaborative workflows in any environment.

The doc can wait an hour

Speed is the cheapest thing AI hands you. It's also the thing almost nobody asked for. Very few research docs or competitive analyses carry a deadline measured in minutes, yet we ship them like the building's on fire.

The extra hour is where criticality lives. Read the whole thing, end to end, before anyone else does. Open every source you didn't pull yourself. Check every number, because a number is the easiest thing for a model to invent and the hardest thing for a reader to question.

I'll own my part here. I've passed along AI output I didn't fully validate. It looked right, it sounded like me, and I was moving fast. Nobody caught it, which didn't make it okay. It just meant I got lucky.

In some cases it required me passing it on for validation, because I didn't know the truth to begin with and it's design's job to do so, but that didn't reduce stakeholder frustration upon encountering the hallucinations.

Put your name on it like you mean it

Corporate environments have always rewarded visible output. A 30-page doc reads as proof of work, and for a long time it was, because 30 pages used to cost someone a week. That just ain't true anymore, pardner.

I think what stays scarce is a person willing to stand behind what they share. So build your own standard for the question underneath every send: what value does this bring to my org, my client, my day, my life? Mine runs through a few checks:

• Would I send this if my name were the only thing on it, with no "AI-generated" disclaimer to hide behind?

• Could I defend every claim in a meeting without opening the doc?

• Does it make a decision easier, or does it just make the decision longer?

• Who has to read it, and what does that cost them?

Your list will likely look different. The point is having one, and running it before the send button will save you some pain.

Do it the old way, then compare

Never let speed stand in for expertise. The fastest way to find out where your expertise actually lives is to do the work without the shortcut once in a while, then hold the two versions side by side.

Take a slop grenade doc for example. Anyone who has shipped product could spot the gap in the first page especially if there were no clear users, no business context, no one accountable for the call. The doc mighty bury that absence under principles and laws, and the citations make it look airtight. An expert sees what's missing. A skimmer sees what's there.

Try this with one deliverable a week:

1. Write your own outline first, by hand, before you prompt anything.

2. Generate the AI version.

3. Compare them. What did it add that you'd missed? What did it skip that you knew mattered?

That gap is the job. Sometimes it shows you a blind spot of your own. More often it shows you exactly what the model can't know about your org, your users, and your history, and that's the part people hired you for.

Decide who you want to be in their inbox

Every slop grenade moves work from the sender to everyone else. You save three hours generating the doc. If six people each spend an hour reading it to discover there's nothing to act on, the team just paid three hours for your shortcut. The math gets worse every time you do it.

The bigger cost is quieter. Your peers learn. After the second or third grenade, they start skimming anything with your name on it, and the good work you send later gets the same discount. Trust compounds in both directions.

In my case, I was dependent on the engineer to execute what I built, but at the end of the day it began to feel like they didn't respect design input and that they would rather have any number of permutations just generated to have any result, rather than an opinionated one to test against.

So ask yourself what you want people to think when your name shows up in their notifications. "Oh good, they've thought about this" takes years to earn and one lazy afternoon to lose.

No explosion, same damage

If you pulled the pin above, you saw the ending. The countdown hits zero and nothing blows up.

Slop rarely explodes on a short timer. It just sits in a shared drive, quietly costing people time and costing you a little credibility with each send. That means it can become more unstable and dangerous the longer it goes without being addressed. The fix is unglamorous: slow down, set your own bar, check your work against what you actually know, and remember there's a person on the other end holding whatever you throw.