Executive summary

Thesis: Agentic AI could trigger a self-reinforcing automation→demand collapse, where cost‐cutting layoffs depress spending, fueling further automation and amplifying macroeconomic stress through financial and housing channels.

  • Scenario origin: Citrini Research’s two-year pre-mortem models U.S. unemployment roughly doubling to 10–12% and the S&P 500 falling about 38% (to ~3,500), based on automation-driven contraction in outsourced white-collar services.
  • Immediate impact: Outsourced consulting, IT, and procurement providers face margin compression as firms substitute agentic AI for contractor labor, according to Citrini’s assumptions.
  • Scale and speed: Labor’s GDP share falls from 56% in 2024 to 46% by 2028 in the model, accelerating a spiral of layoffs, “ghost GDP,” and reinvestment into additional automation.
  • Financial amplifier: The scenario flags ~$2.5 trillion in vulnerable private credit and concentrated tech-hub housing as channels for cascading defaults, per Citrini’s estimates.
  • Model caveats: All figures are conditional outputs of Citrini’s scenario framework; absent agentic performance benchmarks, these should be treated as illustrative stress-test parameters, not forecasts.

Breaking down the scenario

Citrini’s framework centers on narrow but agentic AI systems capable of autonomously executing multi-step decision workflows. The scenario’s causal sequence is: AI substitutes for outsourced decision labor → firms realize payroll savings → reduced household incomes depress consumption (“ghost GDP”) → margin pressures drive fresh AI investments → system performance improves, repeating the cycle. The report attributes a doubling of U.S. unemployment to 10–12% and a 38% S&P decline by mid-2028 directly to this feedback loop, treating these outputs as scenario assumptions rather than predictive estimates.

The model abstracts from sector granularity, grouping outsourced services (consulting, legal, procurement) into a composite cost center roughly aligned with 8% of U.S. GDP. Citrini’s sensitivity analysis (not published) is said to show that a 5% cost reduction in that segment can trigger a >2% drop in aggregate demand over two years. No public agentic benchmarks validate the pace of cost reduction, so the timeline remains highly uncertain.

Why now

Two factors lend urgency to the scenario’s plausibility. First, major technology vendors have begun piloting agentic prototypes in procurement and back-office workflows, shifting these use cases from research demos to controlled production environments. Second, large enterprises already externalize complex decisions to third-party vendors, so a marginal improvement in AI autonomy could substitute entire contract portfolios rather than augment individual tasks. In this context, even modest performance gains in agentic pilots could shorten the latency between technological advances and measurable economic outcomes.

However, the scenario does not account for potential countervailing forces such as rapid upskilling of labor, emergent business models around human-AI collaboration, or demand creation from new AI-enabled products. These hypotheses underpin mainstream productivity projections that remain plausible given current data gaps.

Risks, uncertainties, and governance considerations

  • Financial amplification: Citrini flags ~$2.5 trillion in private credit exposure as a potential contagion vector; stress scenarios without broad market liquidity could magnify defaults in leveraged deals tied to offshore outsourcing firms.
  • Housing concentration: Tech-hub mortgage stress and commercial real estate in cities with high contractor populations could cascade into regional banking strains, per the scenario’s regional debt‐service sensitivity analysis.
  • Policy mismatch: Monetary tools target interest-rate cycles, not intelligence-driven demand shocks. The model suggests traditional rate cuts may have limited effect once “intelligence scarcity” depresses consumption through unemployment rather than financing costs.
  • Legal-operational risk: Autonomous agents making procurement and contracting decisions risk breaching compliance, cybersecurity, and liability thresholds if human oversight protocols lag behind deployment speed.
  • Model uncertainty: All timing and magnitude estimates derive from assumed AI cost curves and exogenous spending elasticities. Absent open benchmarks on agentic AI performance, the scenario remains a diagnostic tool rather than a projection.

Comparative framing

Unlike standard productivity models that assume AI augments labor, Citrini’s scenario emphasizes substitution and demand destruction. This “death of intermediaries” framing echoes concerns from the early SaaS era—where platform disintermediation reshaped software markets—but extends them to white-collar decision ecosystems. Mainstream forecasts, including a January 2024 Citrini report projecting $420 billion in AI/ML revenues by 2027 at 72% CAGR, illustrate how shifts in capability assumptions can pivot institutional outlooks from bullish to crisis-oriented.

Critics argue that efficiency gains often birth new industries—data annotation, AI ethics services, and human-AI ecosystem roles—that can offset net demand losses. Until agentic benchmarks and adoption rates are empirically validated, these counterarguments retain equal plausibility in macroeconomic forecasting exercises.

Monitoring signals and governance implications

  • Corporate disclosures: Sudden upticks in CAPEX on AI platforms relative to labor line items may signal accelerating substitution trends. Tracking this ratio at S&P 100 firms could function as an early stress indicator.
  • Private credit stress tests: Inclusion of AI-automation feedback loops in credit-risk models would reveal concentration risks in leveraged outsourcing plays. Regulators might consider scenario overlays incorporating 5–10% demand shocks in these portfolios.
  • Procurement automation metrics: The emergence of API-first agentic procurement tools should be monitored as a gauge of in-house substitution velocity, particularly where vendors lack human-in-the-loop controls.
  • Labor share tracking: Persistent declines in aggregate labor compensation as a share of GDP—beyond typical business-cycle swings—could reflect systemic AI substitution rather than cyclical unemployment.
  • Legal and compliance filings: Escalations in litigation or regulatory actions tied to autonomous procurement or contracting outcomes may presage broader governance challenges.

What to watch next

  • Public reports of agentic pilots at scale, particularly in procurement, back-office, and consulting verticals.
  • Quarterly earnings and guidance from major outsourcing vendors such as TCS, Infosys, and Wipro for unexpected revenue contractions or margin pressures linked to AI substitution.
  • Investor discourse led by prominent figures (e.g., Michael Burry’s Stocktwits commentary), which may amplify market perceptions of macro AI risk.
  • Regulatory consultations on autonomous decision-agent frameworks, procurement standards, and labor transition policies in major jurisdictions.

This diagnostic scenario reframes agentic AI from a productivity upside to a potential source of systemic demand risk. Observing these signals and governance levers could prove consequential for understanding how rapid, large-scale labor substitution may reshape macroeconomic resilience.