How to Deal with AI Cognitive Debt

I work with AI all day, every day, and I'm overwhelmed — I think everyone is. I cope with this across a few verticals, which I'd like to list here in hopes that it's useful to you. I want to focus on the 80/20 of it all.

Personal Use

The biggest problem I've experienced is that our minds seem to treat the AI's responses as our own thoughts. What do I mean by this? If I prompt an agent and we discuss a topic, I subconsciously take ownership of what's said. In practice, my sense of ownership is out of sync with my actual understanding — unless I make the agent go at my pace.

This is the crux of every issue with AI nowadays: we hope it will do our thinking for us, and we end up going too far, too fast. We lose control, and with it, responsibility.

My solution is to go slow and make sure I'm on the same page as my agents. It means I'm not flying at "light speed," but unlike a lot of people, I produce quality work.

Team Management

I've now worked with several people who outsource too much of their thinking to AI. They use skills to write specs and tickets, then work entire sprints on features they don't understand. Later, when their pieces don't fit the bigger puzzle, they start wondering where they went wrong.

These people don't offer much value beyond what the agents themselves offer. On top of that, they don't even realize their predicament until it's too late — and by then they're in technical (and cognitive) debt that's very hard to pay off.

My solution is to hold them accountable by asking cutting questions about what they're building. Nine times out of ten, their inability to explain it produces enough shame for them to close the gap. After three or four rounds of this, they usually adopt better habits and actually start understanding what they're building.

Producing Content

I still write everything by hand and only use AI for editing. I know today's token-maxxers and "full-stack marketers" scoff at this, and maybe for marketing purposes they're right. But I like the personal touch — otherwise I'm just polluting what's left of the internet. It also keeps me from writing things I can't stand behind.

Writing Specifications

AI agents go really fast and really hard when writing specs for my apps. If I let them go all in, I usually end up with slop. Why? Because I lose track of what I'm building.

Instead, I go slow with the agent and talk it through until I myself understand what I want built. Only when I can envision what I'm building — whether on a product or a technical level, depending on the project — do I let the agents go after it. This takes some discipline, but I paid for the lesson with several slop-infested projects I had to burn.