We've hit collective freeze response

Here’s a strange signal: experts are getting more worried about AI displacing jobs, while regular people are getting less curious about it. That gap is not comfort. It’s a warning.

Through 2025, people noticed. Searches for the practical survival skills — prompt engineering (writing instructions that get useful answers from AI), GitHub Copilot, coding bootcamps, even “technological unemployment” — all spiked. Then, sometime mid-2025, they all peaked and rolled over. Prompt engineering interest is down 36% from its June 2025 high. Coding bootcamp curiosity has cratered 69%. Online learning is off 55%.

The tempting read is relief: maybe the panic was overblown, and people moved on. But look at what didn’t slow down.

The thing they stopped watching kept growing

Developers are pulling AI tools at a staggering clip — over 40 million downloads a week across the major SDKs, with no sign of deceleration. OpenAI and Anthropic libraries alone account for roughly 40 million of those weekly installs. The capability that worried everyone is still compounding. People just stopped paying attention to it.

That pattern has a name worth borrowing: normalization. We notice a threat, we process it (often emotionally, often through entertainment rather than action), and then we fold it into our worldview as “just how things are” — without changing a single behavior. Declining curiosity isn’t the same as solving the problem. Sometimes it’s the sound of a window closing.

The buffers are thinning while attention fades

This would matter less if households were getting more resilient. They’re doing the opposite. The personal savings rate slid to 3.6% in early 2026, down from 4.0% a few months earlier. Revolving credit — mostly credit cards — sits at $1.34 trillion and keeps climbing. People are saving less and borrowing more, exactly when their safety margin matters most.

One reassuring number remains: credit card delinquency is still healthy at 2.92%. That’s the last quiet indicator. It’s also a trailing one — it tells you about damage that already happened, not damage that’s coming.

Why the expert–public gap is the real signal

Watch the divergence. The New York Times ran an escalating drumbeat through spring 2026 — economists newly worried, Congress doing little to prepare. Meanwhile, the trending content people actually chose to watch was a movie trailer, a pop star, and Netflix. Not learning. Not preparation.

When the people closest to the data sound more alarmed while the public grows calmer, that divergence usually resolves the wrong way. The crowd isn’t reassured because the risk fell. It’s reassured because it stopped looking.

What I’d watch next (and what’s speculative)

Let me flag the uncertainty plainly: none of this proves a crash is coming, and the timing is a guess. AI-driven job losses aren’t yet visible in the claims data. But the setup is worth naming because it’s testable.

The likely sequence, if it breaks: unemployment claims climb past 250K while AI tool downloads hold high, and within a few quarters credit card delinquency pushes above 4% — the moment normalization snaps and people scramble. The sharpest early tell would be a sudden surge in people searching “career change” at the same time delinquencies cross 3.5%. That combination means workers are reaching for the exit after the financial damage, not before it.

Why this matters

The cruel thing about normalization is that it feels like wisdom. Calm looks like maturity; disengagement looks like having made peace. But preparation windows don’t announce their own closing.

You can’t control the macro numbers. You can decide not to mistake your own fading curiosity for safety. If the topic feels old and settled, that’s worth a second look — because the capability driving it is still accelerating, and the buffer that would catch a fall is getting thinner every month.