AI & the job market
Lots of layoffs have been announced, but how much is AI to blame?
TL;DR
AI is triggering a rapid reallocation of work.
Jobs are being broken apart and reassembled around new technologies.
Today’s layoffs are cyclical and transitional, not structural collapse.
The tension is between the speed of technological substitution and the slowness of institutional adaptation.
Deflation and abundance will be local.
A bunch of companies have announced large layoffs recently, and people are falling all over themselves to blame AI. These layoff announcements, we are told, augur widespread job loss at the hands of AI. We will soon all be sitting at home, twiddling our thumbs, collecting universal basic income while Sam Altman counts his trillions.
If only it were so easy.
Below, I try to bring some nuance to the debate.
What’s actually happening: reallocation, not collapse
We’re entering an age of reallocation, not an age of unemployment. AI and automation displace some tasks but complement others. Jobs are bundles of tasks, and history shows that when technology eats one part of the bundle, new ones emerge around supervision, orchestration, interpretation, and maintenance.
Industrial revolutions recompose labor. The factory, spreadsheet, and internet all triggered waves of technological unemployment rhetoric that never materialized because productivity gains eventually translated into new demand, industries, and roles. The same dynamics are in play with AI, just faster, noisier, and more cognitively uneven.
The layoff narrative is mostly cyclical
Layoff spikes often follow overheated expanasions, not technological shocks. The post-COVID hiring binge left firms bloated; rebalancing looks scary in the data but rarely signals a permanent structural break. If AI were truly eliminating labor demand en masse, we’d see falling job openings, labor-force participation, and wages across sectors. None of that’s evident yet.
That said, the churn is real. Middle-skill, routine cognitive roles are compressing faster than replacement tasks appear. That creates local unemployment and transitional pain even if aggregate employment holds.
“Compute replaces headcount” misunderstands production
AI infrastructure is not a subttitute for people so much as it is a new industrial stack built by people. Data centers, chips, energy systems, model evaluation, safety, compliance: all these require vast human capital. Capital deepening doesn’t erase labor; it changes who’s valuable and where they cluster. The AI build-out may end up employing more electricians, engineers, and grid operators than it displaces coders.
Deflation and abundance are partial, not universal
Information goods do show sharp cost declines. But the economy runs on atoms. Power, housing, bandwidth, and regulation remain bottlenecks. So we get local abundance (in intelligence, information, creative output) alongside persistent scarcity in the physical world. The result is uneven price dynamics, not uniform deflation.
“Paid to stay home” is politically, not technologically, determined
There’s no mechanism by which AI-generated productivity automatically translates into universal income. Abundance without redistribution simply concentrates surplus. Paying people to stay home presumes a large-scale fiscal transfer regime that doesn’t exist and, historically, faces deep political resistance.
The real tension: speed of reallocation vs social adaptation
The danger isn’t permanent unemployment. It’s frictional lag: how long it takes displaced workers to retrain, relocate, or reorient. When the pace of task substitution outruns the pace of institutional adaptation, you get pockets of despair, political backlash, and productivity paradoxes (gains without felt prosperity). That’s the real risk of the AI era.
What to actually watch
If we were truly entering a jobless future, you’d see sustained declines in job openings and participation, rising long-term unemployment, and wage collapse alongside high productivity. If, instead, we’re in reallocation mode, you’ll see churn, wage polarization, and new roles appearing faster than we can categorize them.
Right now, all evidence points to the latter.
The likely path
Near term (1-3 years). Sharp churn, mid-skill compression, firm-level productivity gains but uneven worker adaptation.
Medium term (3-7 years). Stable employment but polarized incomes; social pressure for benefits portability and retraining.
Long term (7-15 years). Outcome hinges on whether robotics and embodied AI catch up. If they do, a second, deeper reallocation wave hits. If not, abundance stays mostly virtual.
The real future challenge
The question isn’t “What happens when there’s no work?” It’s “How do we allocate meaning, income, and status when cognition is cheap?” Work will persist, but its symbolic and distributive roles will change. That’s where the politics will be, not in some idle utopia.
If you enjoy this newsletter, consider sharing it with a colleague.
I’m always happy to receive comments, questions, and pushback. If you want to connect with me directly, you can:
Great article; a vote for AI as a technology. Which like other technologies will shift net jobs rather than reduce them. Albeit with local pains and gains that are hard to fully predict.