There’s a term in aviation called “automation dependency” — when pilots become so reliant on autopilot that their manual flying skills quietly atrophy. The FAA has been worried about it for decades. The plane works great, right up until it doesn’t.
This morning, while doing my thing with Claude, I got the dreaded API Error: 500 - Internal Server Error. A quick check of the status page confirmed a major outage.
And I just... sat there.
Not for a second. For longer than I’d like to admit. There was a low-grade cognitive stall — the same one you get when the Wi-Fi drops mid-thought. Except this wasn’t Wi-Fi. This was me realizing that the first tool I reached for, before I’d finished my first cup of coffee, was Claude. I didn’t have an obvious convenient fallback. I just had the error message and the quiet.
We talk about AI adoption like it’s still a choice. Like we’re evaluating options, running pilots, deciding whether to “integrate.” But adoption has a way of becoming dependency before you’ve decided it has.
That’s not inherently bad. We’re dependent on electricity too. The question isn’t whether to depend on these tools — that ship has sailed for most of us doing serious knowledge work. The question is whether we’re building the muscle to know what to do when they fail. Whether we’re maintaining any manual flying skills at all.
Because the tools will fail. They’ll go down, hit rate limits, and spit out error messages at the worst possible moments.
The pilots who scare me aren’t the ones who use autopilot. It’s the ones who’ve forgotten they’re still flying the plane.