I finally have the words for the feeling of being overextended when scaling work with AI: 'Cognitive Debt.'
It's that creeping sensation of losing the plot. You're moving fast and hitting milestones, but you've stopped building any shared understanding of why the work is going where it's going.
In a recent talk titled 'Understanding is the New Bottleneck' Notion's Geoffrey Litt picked up the idea (the term was originally coined by researcher Margaret-Anne Storey). His argument: when AI agents write code faster than anyone can read it, human understanding of that code becomes the weak point.
You don't need to be a software engineer to feel this. I don't write code, and I see cognitive debt everywhere.
Design a customer journey, run the data analysis, build out the product roadmap. Hand any of it to AI without understanding the engine and you end up stranded. Verifying that the output is correct is the low bar. You want to understand it well enough to still have a say in it.
Hold a rich conceptual picture of a project and you can still pivot, recombining ideas into the next move. Let the AI accumulate all the context and you have quietly stopped leading the work.
The fix is boring: demand transparency from your tools. Treat the AI like a collaborator and make it explain its steps and its reasoning. Stay in the loop.