We are trading productivity for cognitive debt

We are trading productivity for cognitive debt, and most of us have not noticed the bill.

I have been thinking about this a lot as the recent models have grown genuinely impressive. The MIT Media Lab put a name to it this year: writers who leaned on LLMs showed measurably lower neural engagement. Cognitive debt.

For software engineers and architects, the choice between deep and shallow understanding is not new. It used to be StackOverflow or a blog post, copy-pasted in haste, versus the slower work of understanding the problem space well enough to build something strong and context-aware.

LLMs raise the same choice, but with a twist: their solutions always look credible and optimal. Only a seasoned engineer can specify a problem properly and judge a solution against the criteria that are specific to their company, domain, project and timing.

And as we hand more of that thinking to the machine, the skill quietly erodes. Not overnight, but through a slow loss of comprehension, tooling instinct and programmatic and design fluency that is hard to notice until you need it and find it gone.

So the reminder I keep giving myself: think, evaluate, challenge whatever you introduce into your systems. Keep the mind sharp and the decisions deliberate.

This post was inspired by a great read from Igor Roztropiński: The Joy and Power of Understanding.

Follow-up: Spending judgement where it matters: how we structure AI-native delivery.