< Your Brain on GPS

I can remember the first time I used SatNav in a car. I'd rented one in Berlin and was driving through Eastern Germany and Austria. It had an interface which was a single line text screen and a knob to select the character[1]. I was dubious. But after navigating through a maze of one-way streets and lanes in Dresden, I'd ditched the mapbook and was fully dependent on it.

The snag? I didn't realize the maps only covered Germany. When we hit Austria the screen went blank. I had no idea where I was.

I'm sure my navigation skills have fallen off a cliff since. Research backs this up -- studies show GPS users have reduced hippocampal activity and worse spatial memory than those who navigate "traditionally". The skill atrophies when you don't use it (or indeed, even develop it in the first place).

Now I'm watching the same pattern with AI coding assistants. I've been having an extremely productive run with them. But. Over the last few days I've (personally) noticed a steep degradation in how effective Claude Code (Opus 4.5) is for me[2].

Just a few weeks earlier I'd been leaning heavily into being an AI-coding-maximalist. Suddenly my day was upended as I wasn't able to get through my AI-coding-maximalist expected tasks.

I'm more than sure Claude will leapfrog again and this whole thing will be moot[3], but it was a bit of a sense check at the same time. We're integrating this to our infrastructure and processes at a rapid rate; and we should not expect it to be perfect. I'm not sure we're considering that we still need those fallbacks.

Turns out the map can go blank at any time. Best to keep a vague idea of where you are.

I write about AI, organizations, and engineering leverage: find out about me and subscribe here.

Discuss and share via the meta page . Filed under AI, Code, People, and 100PR.

Footnotes

  1. The way it worked is you selected a character and then it narrowed the selection based on the next possible steps. Was surprisingly effective.

  2. This is purely subjective at the best of times, but corroborated somewhat by Reddit FWIW and Margin Lab.

  3. And, indeed, the Opus 4.6 Release did seem to resolve it.