I’ve been tinkering with AI tools a lot lately. AI tools are a constant companion, and I find myself flicking open O365 Copilot for all sorts of little things. The other day, I was staring at a Teams message I needed to reply to. Instead of just writing it, I prompted an AI:

“Draft a polite but firm response to Joe Tester saying his code review is late and I need it by EOD.”

Two seconds later, I had the perfect paragraph. I pasted it, hit send, and felt that little productivity hit. I probably saved 90 seconds. But I’ve been thinking about that moment a lot since then. Did I really save anything? Or did I just spend a little bit of my focus for later?

The New Form of “Cutting Corners”

As engineers, we’re always looking for leverage. We write scripts to automate tasks, we use aliases to speed up our workflow, and now we have AI. We’re not just writing code anymore; we’re “vibe coding”—giving the AI the gist and letting it handle the syntax.

Each time it works, it feels great. It’s a little sugar rush of efficiency. But as I’ve used these tools more, I’ve started to notice a familiar feeling, one that reminds me of shipping a feature right before a deadline. It feels a lot like technical debt, but for our brains. I’ve started calling it Cognitive Debt.

Every time we take an AI shortcut, we add a little bit to that debt. For me, it shows up in a few ways:

1. The “Is This Actually Right?” Tax. You didn’t write it, so you have to audit it. AI suggests a 10-line function that looks right, but is there a subtle off-by-one error in there? That AI-generated email sounds polite, but is it my voice? The mental energy I spend verifying and tweaking someone else’s work (even a machine’s) feels different, and frankly more draining, than creating it myself.

2. The Switch from Creator to Editor. When I’m deep in a problem, I’m the architect. I’m building the system in my head. The moment I ask an AI for help, my role changes. I’m no longer the architect; I’m the building inspector, walking through a structure I didn’t design, looking for cracks. Switching between those two modes is jarring and breaks the deep focus—”the zone”—that’s so hard to get into.

3. Losing the Muscle Memory. I think a lot of us have a certain pride in our craft. The art of debugging, of tracing a problem to its source, is a skill built over years of struggle. Those frustrating moments are where the real learning happens. By asking an AI for the answer, we get the fix, but we miss out on the workout. Our problem-solving muscles get a little weaker.

The 10-Hour Prompt Engineering Fallacy

This really crystallized for me when thinking about a bigger project. You estimate a task will take 20 hours of focused work. Then the idea pops in: “I bet I could spend 10 hours getting the prompts just right and have an AI build the whole thing.”

It’s the ultimate “cut a corner” temptation. But those 10 hours are rarely a clean investment. It’s a messy cycle of prompting, debugging weird outputs, and trying to explain nuance to a machine. You might end up with something that works, but you don’t know it. You didn’t build the intuition for it. Six months down the line, when it breaks, you’re not fixing your code; you’re trying to figure out what a black box was thinking.

So, What Am I Trying to Do About It?

This isn’t a post about ditching AI. These tools are incredible. It’s about being more purposeful, which is something we talk about a lot in our Q4 reflections. For me, that means adopting the old mantra: “slow is smooth, and smooth is fast.”

• Automate the Tedious, Not the Thoughtful. I’m trying to use AI for the truly boring stuff—boilerplate, generating test data, or summarizing a long document. But the moment the task requires core logic or a thoughtful response, I’m taking the controls myself.
• Protecting “The Zone.” That state of deep work is where our best work happens. AI can be a constant interruption, a tap on the shoulder. I’m trying to be more disciplined about using it in batches, rather than as a constant companion.
• The Five-Minute Rule. Before I even think about asking an AI, I’m trying to spend five minutes with a text file, just outlining the problem. What are the inputs? What’s the goal? That small act of “slowing down” almost always makes the solution clearer and, ultimately, makes me faster.

In the end, it’s a shift from thinking about raw velocity—how many tasks we can check off—to building sustainable momentum. That 90 seconds I “saved” on that Teams message cost me 15 minutes of focus later. That doesn’t feel like a good trade.

I encourage you to think about where these tools fit into your own workflow. Are they serving you, or are you serving them? It’s a question I think we’ll all be wrestling with for a while.