Don't Be a Slop Cannon
Don't Be a Slop Cannon
AI for knowledge work unfortunately feels like a part of the hyper gamblification of everything. Now instead of actually thinking it's too tempting to press the AI slot machine button again and again. It's really easy in this day and age to develop bad habits around instant gratification. Which is why we haven't seen an explosion in great / new / novel software and way worse system uptimes generally. I think the arbitrage right now is having the patience and determination to compound daily, think long term, and to do hard things. It's easier than ever to slop 10k LOC per day but the bar for finding product market fit hasn't moved an inch
Skip the thinking. Press the button until something looks right enough.
Amplify, don't decide
You already know what you want to say, or think, before you open the AI. It can help you say it better or move faster. What it can't do is make the decision and still leave it yours.
That holds whether you're writing an email or making a call at work. Skip your own judgment and take the output anyway, and you've handed over the whole decision.
Expertise pays off with AI
A chef doesn't follow a recipe. He tastes as he cooks and knows the second something is off, because years in the kitchen gave him that instinct. Someone who has never cooked can follow the exact same recipe, step by step, and still end up with a dry, gray Beef Wellington. They did everything the recipe said. They just couldn't tell the pastry was too warm, or the meat needed two more minutes, because no recipe teaches you to notice that.
The same gap shows up with AI. A senior engineer can read code an AI wrote and catch the one function that doesn't actually exist, in seconds, because they've built that kind of system before. A junior engineer sees code that compiles and looks reasonable, and ships it. AI didn't treat them differently. Their own experience decided whether the mistake got caught.
Use AI hardest in the places where you already have the instinct to catch it being wrong. Slow down everywhere else.
The ownership test
Someone selling prints at an art fair gets asked why they picked that color, or what the piece means. They generated the image, liked how it looked, and printed it. No answer. They never asked themselves the question first.
That's the test for anything AI helped make. Can you explain the choices in it? What you kept, and why. If you can, the work is yours, even though AI helped build it. If you can't, you weren't the author. You just typed the prompt.
Build wide, not narrow
The instinct is to hand-build a rule for every case you can think of: do this, then that, check this, then that. It feels careful. But narrow setups stay narrow. They don't get better when the model underneath them does. A system built on a few general moves, read a file, run a command, gets stronger every time the model improves. A hundred specific rules just get more out of date.
The same holds outside of software. A habit tied tightly to today's AI quirks will need rebuilding soon. A habit built on judgment, knowing what good looks like and checking the work, holds up no matter what the tool becomes.
The bigger picture
We already live surrounded by things made to be thrown away. A shirt that pills after five washes. A phone that quietly slows down right when the new model ships. AI applies that same pressure to something new: intelligence.
No one takes pride in slop. Not the person who made it, not the person who gets it. The satisfaction was always in the friction: deciding, and standing behind it because you made it.
Use AI. Use it a lot. Just don't let the speed replace the part of you that decides.
Don't be a slop cannon.