Have you ever turned up to a place wildly overdressed for the occasion? When I was leaving for university, my father convinced me to wear a suit for a 12+ hour flight. My eventual roommate met me on that flight. He was in a hoodie and slacks. He still loves to remind me how he thought I was a complete basket case when he saw me in my suit, with a briefcase from the ’60s to complete the outfit.

But overdressing is not the only way to make yourself cringe. You might have mistakenly hit “reply all” and sent your unfiltered thoughts to 400 people. You might have called someone by the wrong name with the sort of unshakeable confidence my father had when he told me to don that suit. You might have forwarded an article you clearly didn’t read, only for someone to ask you about it in the next meeting.

Right now, a lot of people are putting on AI the way my father put me in that suit. Only about 15% of people in the U.S. use it daily for work. That number has been climbing fast, but it still means 85% of the workforce — many from GTM — are either not using it or just getting started. And early in a journey is when those cringe-worthy mistakes get made.

So, we put together a guide on the 5 deadly sins of AI use. You probably know most of this already (right?), but you definitely know someone who doesn’t. Send it their way, before they show up for a red-eye in a three-piece suit.

Few things make me cringe the way an obviously AI-written email landing in my inbox does. I get it. Once you discover AI can help you manage your inbox, you’re going to try. But AI written emails come across as far less sincere, and the people reading them can feel it, making them far less effective.

A University of Florida study found that when supervisors used AI to write internal emails, only 40–52% of employees viewed them as sincere. A non-AI email? 83%. Your team can tell when you didn’t write it, and what they feel is “you weren’t worth my time.”

Now, 63.5% of people can’t actually detect AI writing. Which means it’s possible to use AI for email without looking like a robot. The trick is that the people who get away with it aren’t hitting send on what the machine gives them. They’re editing to add in the soul that AI just always misses.

Do that and you’re already ahead of half the users of AI. And remember, if your email starts with “I hope this email finds you well,” know that it didn’t find them well. It found the trash.

This is going to hurt my brand, but I was quite horrendous at math growing up. Not the simple stuff. But things like quadratic equations made me queasy. So, I have nothing but empathy for people who want to use AI for the math that many of us have to do in our jobs. But empathy doesn’t make it safe.

In October 2025, the ORCA Benchmark gave the leading AI models 500 real-world math problems. Things like unit conversions, and calculating probabilities and compounding interest. The best model was wrong 4.5 times out of every 10 problems it solved. In academia, that’s a solid D, which where I grew up got a shoe thrown at you.

For all of us who are blown away by what Claude can do with a financial model, here’s a question: How do you know what you got is right? Unless you go check everything, you don’t really know. And if you have to check every formula and every calculation, that might be more work than doing it manually in the first place.

I’m not saying don’t use AI for these things. But don’t one-shot it and confidently send it off, as if you’ve just solved the Riemann Hypothesis. You probably didn’t.

I just have to say this: Please DO NOT send your AI note-taker to a meeting you’re not actually attending. It’s corny, it’s annoying, and you’re better off wearing white after Labor Day.

Now that I’ve got that off my chest, let’s talk about the meetings you do attend. Note-takers seem like an obvious win — the market’s gone from non-existent to a projected $15B+ by 2032, and for good reason. Jotting down takeaways while also looking present and engaged was always a losing battle.

But 84% of users modify what they say when a note-taker is present. Think about that. You brought a tool to capture an open conversation, and its presence made the conversation guarded. How useful are the notes if the meeting they tracked wasn’t a real one?

Everyone who starts using AI quickly starts using it to help them write. That’s natural. They start posting on social media at the velocity of a paid influencer after being lurkers for the previous decade. They go from taking three days to respond to an email to sending multiple internal memos before the clock strikes 8am. Strategy docs are being churned out and shared around like they’re the next Milton Friedman.

That’s all fine, as long as the content is good and has a soul. We already covered that. But even then, word selection and punctuation could be your downfall.

You probably know about the em dash by now. This poor little punctuation mark — a staple for centuries — was the first casualty of AI. The woolly mammoth has better odds of making a comeback at this point. Then there are words. Many words. Words like “delve” and “realm” and “fluff” are done.

It’s not about what’s made with AI anymore. It’s about what sounds like AI. (See what I did there?) 52% of web articles are now AI-generated, most of it AI slop that gets ignored. So, you need to become a master editor. That’s the skill to master.

When AI first showed up in our organizations, we all took the same approach: Let people experiment and figure it out. That was probably the right call. Today, the average large enterprise uses 2,191 applications, only 61% approved by IT. Half of workers admit to using AI tools at work without getting approval. We definitely experimented a lot. 

The problem with that approach is that everything gets to 90% ready and stalls. That last 10% to make something production-ready is a heavy lift, but that’s where the real alpha lives. You close that gap by centralizing your AI deployment strategy. (Kyle Norton has written a wonderful substack about this.)

Otherwise, you’ll be the leader at the dinner who can talk about all the cool experiments your team is running, but a few months in you try to find their impact and you can’t. Productivity isn’t up. Efficiency isn’t up. Just new spend. Balancing decentralized experimentation (which should still happen to find interesting use cases) with centralized deployment is the only way to get proper ROI in this new world.

None of this is about avoiding AI. Use it. Use it a lot. But don’t abdicate to it. Make sure what it’s saying is correct. Make sure it sounds like you. And always, always give it your human touch before it leaves your hands.

Btw… my roommate turned out to be one of my best friends. He still brings up the suit. But I never made that mistake again. That’s the thing about cringe. Once you feel it, you never forget it.

Become a Topline insider by joining our Slack channel.

We want to hear your feedback! Let us know your thoughts.