Last Tuesday, around 11:15 PM, I was sitting in my home office with a cold cup of coffee and a crushing sense of self-loathing. I had to finish a post-mortem report for our logistics lead about a botched rollout we had in Scranton three weeks ago. I was tired. I was bored. So, I did what everyone does now: I pasted my rough notes into ChatGPT and told it to “write a professional summary with key lessons learned.”
It gave me five beautiful, bulleted points in about four seconds. I read them, thought, “Yeah, that sounds like something a person would say,” and hit send. The next morning, my boss, Sarah, called me. She didn’t yell. She just asked, “Did you actually think about why the driver stayed at the warehouse for six hours, or did you just want to finish the email?”
The AI had smoothed over the actual problem—a specific data entry error in our legacy routing software—and replaced it with a generic platitude about “improving communication channels.” I looked like a hack because I used the tech to bypass the thinking, rather than sharpen it. It was humiliating. I felt like a fraud. And the worst part is, I see people doing this every single day at work.
Stop looking for a ghostwriter and start looking for a fight
The biggest mistake people make with LLMs is treating them like a high-paid intern who is too scared to tell you your ideas suck. We use them to generate “content”—a word I’ve grown to absolutely loathe—instead of using them to find the holes in our own logic. If you’re using AI to save time on the writing, you’re usually just accelerating the production of garbage.
What I mean is—actually, let me put it differently. The value of these models isn’t that they can write; it’s that they have ingested almost every argument ever made. They are world-class at being contrarians if you force them to be.
I’ve started doing this thing I call “The Stress Test.” Before I finalize any big decision or strategy, I paste my argument into the prompt and say: “I am emotionally attached to this idea. Tell me three reasons why it will fail, and be as mean as possible about my blind spots.”
I tested this over the last month with 22 different work scenarios. Out of those 22, the AI found a genuine logical flaw in 4 of them. That doesn’t sound like a lot, but those 4 saves probably prevented about $12,000 in wasted vendor fees and a lot of awkward meetings. Most of your ideas are actually garbage, and AI just helps you package the garbage faster. Don’t be that guy.
My ‘Red Team’ process (it takes 12 minutes)
I don’t have a “comprehensive guide” for you because I hate those. I just have a workflow that works for me. It’s messy, but it keeps me from looking stupid in front of the VP of Operations.
- The Brain Dump: I write my thoughts in a messy, ungrammatical list. No AI involved yet.
- The Steelman: I ask the AI to summarize my argument better than I did. This confirms it actually understands what I’m trying to say.
- The Interrogation: This is the important part. I tell the AI: “Assume I am wrong. What data am I ignoring?”
- The Friction: I spend at least 5 minutes arguing back.
Real thinking requires friction. If the AI is making your life too easy, you aren’t thinking; you’re just delegating your intellect to a statistical average.
I used to think the goal was zero-shot prompting—getting the perfect answer on the first try. I was completely wrong. If you get what you want on the first try, you didn’t think hard enough. You just confirmed your own bias. Total waste of time.
A hill I will probably die on
I know people will disagree with me on this, but I actively tell my friends to avoid Grammarly’s new AI features and Notion AI. I love Notion for notes—I’ve used it daily since 2019—but their AI implementation feels like a desperate, cluttered cash grab. It’s annoying. It constantly tries to “fix” my tone.
I don’t want my tone fixed. I want my tone to sound like me, even if I’m a bit grumpy or I use too many em-dashes. These “writing assistants” are making everyone sound like a polite, mid-level HR manual from 2014. It’s soul-sucking. I’d rather read a poorly punctuated email that has a real spark of an idea than a perfectly polished paragraph that says absolutely nothing.
I refuse to use any tool that tries to auto-complete my thoughts. LLMs are like those funhouse mirrors—they shouldn’t show you reality, they should show you a distorted version of your own assumptions so you can finally see how weird they look. If the tool is trying to be “helpful” by finishing my sentences, it’s just getting in the way of my own brain firing.
The friction is the point
There’s this weird obsession with “workflow optimization.” People track their words per minute or how many emails they cleared. But nobody tracks the quality of the decisions they made.
I’ve noticed that when I use AI to “spar,” my brain actually feels tired afterward. That’s a good sign. It means I’m actually processing. When I use it to just write the email for me, I feel nothing. I’m just a glorified copy-paster.
Anyway, I’m not saying you shouldn’t use it. I use it every day. But use it to make your life harder in the short term so it’s easier in the long term. Ask it to find the flaw in your budget. Ask it why your marketing angle is derivative. Ask it to play the role of your most annoying client and tear your proposal apart.
The goal isn’t to write faster. The goal is to be less wrong.
I’m still not sure if this actually makes me smarter or if I’m just getting better at arguing with a very sophisticated calculator. Sometimes I worry that I’m losing the ability to find flaws on my own without the robot’s help. Is that progress? I don’t know.
Try it tomorrow. Don’t ask for a draft. Ask for a fight.