The cognitive overload from agentic coding is real.

I’ve got a bad case of “just one more prompt” syndrome.

Before work. Before lunch. Before bed.

Agentic coding is incredible leverage. You ask for a feature and come back to a mini PR. But the cost comes later: review, understanding, and maintenance.

At first, I treated generated code like free speed. It isn’t. It’s deferred cognitive load. Every skipped review is future debt when you need to debug or extend something.

There’s also a second-order effect. Models handle the repetitive tasks — boilerplate, styling, glue code. What’s left is the hard part: decisions, architecture, trade-offs.

So your day becomes high-intensity thinking, back to back. Useful, but mentally expensive.

I still love these tools. They make me faster than anything I’ve used before. But they also make it dangerously easy to over-commit and under-process.

Agentic coding doesn’t remove cognitive load. It compresses it — and if you’re not careful, it compounds.