Why can't I explain my own project any more?
When AI does the thinking, the understanding never reaches your head.
The tool you built with AI works. Everyone uses it.
Then someone asks: "How does it work out the March figure?"
You open the folder and realise you don't know. You didn't write it. You approved it.
Technical debt: shortcuts in code that save time now and cost you later, like putting things on a credit card. The code is messy, so every future change takes longer.
Cognitive debt: the same idea, except the debt lives in your head. It's the understanding you skipped. The code might be perfectly tidy, but nobody knows why it's built the way it is.
Margaret-Anne Storey, a software engineering professor, coached a student team that built fast with AI.
By weeks 7 or 8, one team hit a wall. They could no longer make even simple changes without breaking something unexpected.
They blamed messy code at first. The real problem was that nobody on the team could explain why things had been designed the way they were, or how the parts fitted together.
They had accumulated cognitive debt faster than technical debt, and it paralyzed them.
Simon Willison, one of the people who created the Django web framework, is about as experienced as programmers get. Of his own AI-built side projects, he wrote:
I've found myself getting lost in my own projects. I no longer have a firm mental model of what they can do and how they work.
Mental model: the picture in your head of how something works. It's what lets you guess what will break before you change it.
In 2025, MIT researchers had 54 people write essays over several sessions. One group used ChatGPT, one used a search engine, and one used nothing.
They measured brain activity with EEG: a cap with sensors that picks up the brain's electrical activity through the scalp.
It's a small study, but it points the same way as the stories above.
Understanding mostly comes from the struggle: trying, getting it wrong, working out why.
When the AI does the struggling, you see the finished answer. Reading a finished answer feels like understanding, but it isn't the same as being able to explain it.
It's like being driven somewhere twenty times. You'll recognise the street when you get there, but you couldn't give someone directions.
Explain it back. After each chunk of work, ask the AI to explain what it did and why, then say it back in your own words. Ask it to correct you.
Keep a decisions note. One line per choice: what you decided and why. Future you, or your replacement, will need it.
Do the core parts slowly. The bits that define how the whole thing works deserve your full attention, even if the small stuff gets vibed.
If you couldn't explain it to a colleague, you're not finished.