Our recession alarms are going silent

Five economic instruments just started disagreeing with each other, and the disagreement is the real story. Americans are paying their bills more reliably (delinquencies down 4.6 percent), unemployment sits at a calm 4.3 percent, and yet consumer sentiment fell to 49.8, one of the lowest readings ever recorded. Numbers like these usually move together. Right now they are pulling apart, and AI is sitting in the gap.

Here is the pattern I think we are watching, which researchers have a name for at the individual level: normalized affective discomfort. That is the state where a person keeps feeling uneasy about a technology while using it more every day, until the unease stops registering as alarm. Seberger and colleagues documented exactly this in 2022 for individual users. The new wrinkle is that the same shape now shows up across a whole economy.

When people stop looking, the gauge breaks

We built our economic warning lights on attention. When people worry about losing their jobs, they search, they read, they click. So it matters that Wikipedia traffic to the ‘Unemployment’ article crashed to 44 percent below its peak, dropping below its own historical average. The intuitive reading is that things must be fine. A better reading is that people stopped looking, while the underlying distress kept going.

That single fact undercuts a tool we lean on hard. Attention-based early warning assumes worry is loud. If anxiety goes underground while daily life keeps functioning, the alarm reads silence and calls it safety.

Sentiment is measuring a different feeling now

For decades, economists treated low consumer sentiment as a signal of weak demand and a coming downturn. That model assumes the gloom is about spending power. But spending power is improving on paper, and the gloom is deepening anyway. The most plausible explanation is that sentiment has started capturing something more structural, a low background dread about where AI takes work and meaning, rather than the usual question of whether wallets feel thin this quarter. If that is right, the old ‘low sentiment leads to recession’ rule of thumb is measuring the wrong fear.

Adoption has come loose from approval

Meanwhile, people keep building. Combined AI SDK downloads hit 49.6 million per week, up 24 percent, even as public mood sinks. Those two lines used to rise and fall together, because enthusiasm drove use. Now developers and companies adopt AI because their jobs require it, regardless of how anyone feels about it. Adoption has come loose from approval, which is why you can have a booming tool and a sour public at the same time.

Culture is processing the same knot. The most-watched show of the moment, FROM, drops its characters into a town that traps everyone who enters and forces them to fight just to hold onto normalcy. That is a precise emotional rhyme for what the data describes: stay functional, keep going, and quit expecting an exit.

Why this matters

If I am right, the dashboards we use to spot economic trouble are calibrated for a world where distress is visible and attention spikes before a crisis. AI-era strain looks different. People adapt, keep working, and stop sounding the alarm, so the instruments report calm while the pressure builds underneath.

I’m flagging the harder claims as interpretation. The five-domain convergence is in the data; the causal story tying them together is my best read of it and could be wrong. What I am fairly confident about is narrower and more useful: when a warning system depends on people paying attention, it fails exactly when people get used to the threat. We should stop trusting the quiet.