White-collar layoffs do not stabilize the way blue-collar ones did. They feed back. Each round of cuts compresses consumer spending, which compresses revenue, which compresses margins, which justifies the next round of AI capex, which displaces the next cohort. Citrini Research's 2028 Global Intelligence Crisis names the loop and describes it bluntly as having no natural brake.
This is not a forecast about robots. It is an ordinary economic feedback loop running unchecked because every individual decision inside it is rational.
The mechanism, step by step
AI capabilities improve. A late-2025 coding agent actually replaces a junior developer's ticket queue. A customer-service model closes 80% of tier-one tickets. A paralegal model generates the deposition summary that took a person three hours.
Companies need fewer workers. Entry-level postings are down roughly 35% since January 2023. Employment for workers aged 22 to 25 in AI-exposed roles is down 6% since late 2022; young software developers off nearly 20%. White-collar job openings are down 15% year-over-year. New computer science graduates sit at 7.0% unemployment.
White-collar layoffs increase. ServiceNow's projected 15% headcount cut in Q3 2026 is not one company's bad quarter. It is the signature of an entire enterprise-software cohort discovering that AI agents do the work the seats were sold to do. Citrini projects jobless claims surging to 487,000 by Q3 2027 — the highest weekly print since April 2020.
Displaced workers spend less. The top 10% of US earners account for over 50% of consumer spending. A 2% job loss at the top yields a 3 to 4% decline in discretionary spending. A $200,000 layoff removes that household's salary from circulation, and the lifestyle around it.
Margin pressure pushes firms to invest more in AI. With revenue compressed, the only knob a CFO has is cost. AI capex is the cheapest cost to defend, because the alternative is admitting demand has structurally fallen. The next training run gets greenlit. The loop tightens.
Big-four hyperscaler AI capex for 2026 is projected at up to $660 billion. That number does not pause when unemployment ticks up. It accelerates.
Why nothing in the system stops it
A normal recession has stabilizers. Unemployment insurance puts cash back in pockets. The Fed cuts rates. Inventories clear. Wages fall, then production restarts.
The displacement spiral has none of those.
Unemployment insurance was designed for cyclical layoffs by hourly workers. It does not replace a $200,000 product manager's salary, and it expires before that worker finds a comparable seat — because the comparable seat is the role being eliminated.
Rate cuts do not bring back a job that was structurally automated away. Cheaper credit funds more AI capex, the same direction the spiral is already running.
Inventory clearing assumes the displaced worker can be re-employed when demand returns. The spiral's premise is that demand for the worker is permanently lower, because a model does the task at one-tenth the marginal cost. The funnel into junior roles is the part being eliminated first.
Daron Acemoglu calls this so-so automation — technology just barely good enough to replace a worker, not productive enough to grow the pie. Savings flow to capital. His research finds automation explains 50 to 70% of US wage inequality from 1980 to 2016. AI is the same mechanism with the velocity dialed up.
The political brake is weaker than it looks. AI labs have lobbying capital, a "we are building the future" rhetoric, and a national-security framing that aligns commercial with state interest. By the time politics catches up, the headcount cuts have cleared.
The shape of the next two years
Citrini's timeline is specific. Late 2025: the agentic-coding step function. Q3 2026: ServiceNow's 15% headcount cut. Early 2027: consumer agents controlling purchasing decisions. Q1 2027: Mastercard revenue growth slows to 3.4% as agents route around card rails. Q3 2027: jobless claims hit 487K. November 2027: market crash, perception flips from cyclical to systemic. June 2028: unemployment 10.2%, S&P down 38% from its October 2026 peak.
The sentence Citrini uses to describe the loop is the one to internalize: "Every company's individual response was rational. The collective result was catastrophic."
Nobody is the villain. The CFO who approves the AI rollout is doing what fiduciary duty requires. The board that approves the capex is doing what shareholders demand. The engineer who builds the agent is doing what the comp package incentivizes. Each step is a local optimum. The aggregate is the spiral.
The policy response Citrini sketches — a Transition Economy Act, a Shared AI Prosperity Act funded by a compute tax — is not exotic. A feedback loop with no natural brake needs a manufactured one. The question is whether it gets built before unemployment crosses 10%, or after.