Impact of Agentic AI on the world#

• Source 1: https://alapshah1.substack.com/p/the-global-intelligence-crisis
• Source 2: https://www.citriniresearch.com/p/2028gic

Step-by-step logic#

Phase 0: Foundation#

1. Core assumption of the economy: Human intelligence is the scarce, expensive input that turns raw materials into goods and services.
2. Agentic AI: Not chatbots — systems that do multi-step, autonomous work. METR: unaided task duration doubling every ~6–7 months (~14.5 hours now → trend to ~1 month by mid-2028).
3. Substitution, not complement: Past tech augmented humans; agentic AI replaces cognitive labor. So “tech destroys jobs then creates more” may not hold — new roles still need humans; AI is getting good at those too.

Phase 1: Rational adoption#

4. Per-firm logic: Adopt AI → cut headcount → margins and earnings up. Firms that don’t adopt lose on cost and competitiveness. So adoption is rational for each company.
5. Early evidence (Article 2): Late 2025 agentic coding step-change → “replicate mid-market SaaS in weeks.” Procurement in 2026 asks “build in-house?” → vendor pricing power collapses (e.g. 30% discount; long-tail SaaS crushed).
6. ServiceNow reflexivity (Article 2): ServiceNow sells seats. Clients cut 15% headcount → cancel 15% licenses. So the same AI-driven layoffs that help customer margins mechanically destroy vendor revenue. Workflow-automation vendor is disrupted by better workflow automation → cuts headcount and invests in that same tech. Collective result: every dollar saved on labor flows into AI that enables the next round of cuts.

Phase 2: Negative feedback loop — no natural brake (both)#

7. Loop: AI improves → companies need fewer workers → white-collar layoffs → displaced workers spend less → margin pressure → firms invest more in AI → AI improves. No self-correcting mechanism.
8. OpEx substitution (Article 2): It’s not “less CapEx.” Spend shifts: e.g. $100M labor + $5M AI → $70M labor + $20M AI. Total spend falls, AI spend rises. So demand can fall while AI buildout continues.
9. “Ghost GDP” (Article 2): Output and productivity rise in national accounts, but machines don’t spend. Velocity of money and the human-centric consumer economy (70% of GDP) weaken. Single GPU cluster replacing 10,000 Manhattan workers = “economic pandemic” more than panacea.

Phase 3: Spending concentration (both)#

10. Who gets displaced: White-collar, high earners — the same people who drive spending. Article 1: top 20% ≈ 65% of US consumer spending; Article 2: top 10% >50%, top 20% ≈ 65%.
11. Leverage of job loss on demand: A small % drop in white-collar employment → large % drop in discretionary spending (e.g. 2% employment → ~3–4% hit to discretionary spend). Plus lag: high earners use savings for a few quarters, then spending breaks (Article 2: initial claims 487k, then S&P -6%).

Phase 4: Intermediation layer dismantled (both)#

12. Moat = human friction: Decades of value built on: limited time, patience, habit, willingness to accept bad prices. Trillions of enterprise value on that.
13. Agents remove friction (Article 2): Subscriptions, travel, insurance (passive renewals), financial advice, tax, routine legal, real estate (commissions 2.5–3% → <1%, “agent on agent”). “Human relationships” often = “friction with a friendly face.”
14. Habitual intermediation: DoorDash-style moat (“app on home screen”) doesn’t exist for agents; they compare all options. Coding agents also lower entry barriers → many thin-margin competitors → margins crushed.
15. Payments (Article 2): Agents optimize away fees → 2–3% interchange targeted → stablecoins / instant settlement, fractions of a penny. Mastercard Q1 2027: “agent-led price optimization”; card-centric banks and mono-line issuers hit (Amex, Synchrony, Cap One, Discover). Moat = friction; friction → zero.

Phase 5: Labor market and wage cascade (both)#

16. Core white-collar jobs (Article 1): Stagnant/declining since 2023; +4% vs pre-pandemic (6 years) vs +5% population, +11% real GDP. Information sector already ~8% below peak.
17. Displaced downshift (Article 2): Senior PM $180k → Uber $45k. Overqualified labor floods service/gig → wage compression there too. Then AVs and autonomous delivery hit that gig layer. Sector-specific disruption → economy-wide wage and job pressure.

Phase 6: Financial daisy chain (both)#

18. Private credit (Article 2): >$2.5T by 2026; big share in software/tech LBOs underwritten on ARR that stays recurring. AI makes ARR not recurring (e.g. Zendesk: support automated, no tickets). Marks 100→92→85 while public comps 50. Moody’s downgrades; Zendesk covenant breach → largest private-credit software default. “Permanent capital” = annuity policyholder money in same paper; when that paper defaults, life insurers (Apollo/Athene, etc.) and offshore SPVs create opacity and regulatory stress. Recognition of losses, not just losses, turns it systemic.
19. Mortgages (both): ~$13T. Underwriting assumes stable employment/income for 30 years. Borrowers are prime (780 FICO, 20% down). 2008: loans bad at origination. Here: loans good at origination; world changed after. Income expectations structurally impaired → “are prime mortgages money good?” Delinquencies rise first in tech/finance-heavy ZIPs (SF, Seattle, Austin, Manhattan). Trajectory is the risk; full crisis still avoidable if policy and income stabilize in time.

Phase 7: Fiscal trap (both)#

20. Revenue = tax on human work. As payrolls and white-collar income fall, receipts drop (Article 2: 12% below CBO). Labor share of GDP: 64% → 56% → 46% in four years. Output no longer flows through households → not through IRS.
21. Outlays rise (unemployment, transfers) exactly when receipts fall. Automatic stabilizers assume temporary job loss and reabsorption. Displacement is structural; many won’t be reabsorbed at prior wages. Government must transfer more at the moment it collects less.

Phase 8: Policy and time (both)#

22. Policy lags and is politicized: “Transition Economy Act” (transfers + tax on AI compute), “Shared AI Prosperity Act” (public claim on AI returns). Splits: redistribution vs growth, taxing compute vs “handing lead to China,” regulatory capture, deficits, GFC-style austerity. Real constraint: AI and markets move faster than institutions and ideology.
23. Conclusion of both: This is the unwind of the intelligence premium. Economy can find a new equilibrium, but only if we build new frameworks (tax, safety nets, labor, credit). Article 2’s twist: “You’re reading in February 2026” — canary still alive; time to prepare.

Combined key points (concise)#

Combined action items (integrated)#

Personal / career

• Treat your job as replaceable by agents in a 12–24 month window; shift toward judgment, relationships, accountability, and tasks agents don’t yet do well.
• Use agentic AI daily (coding, research, analysis) so you’re on the multiplier side and your skills stay relevant.
• Stress-test your industry: Is it intermediation (friction), “seats,” or ARR that assumes human labor? If yes, assume repricing and plan exit or pivot.

Household / financial

• Reduce dependence on “prime borrower forever”: Less leverage; don’t assume current income for 30 years; larger emergency buffer.
• Diversify income (skills that work in essential services, gig, or roles less exposed to agent substitution).
• Assume a possible prolonged demand shock and repricing of risk (equities, credit, real estate in tech-heavy metros).

Investing / portfolio

• Audit “daisy chain” exposure: Revenue and margins that assume white-collar employment, ARR stickiness, interchange, or intermediation margins staying intact.
• Hedge or underweight: Private credit/PE software, card-centric issuers, intermediation-heavy business models, real estate in tech/finance-heavy ZIPs.
• Differentiate “AI infrastructure” (convex to adoption) vs “economy that depends on human spending” (concave).

Business / leadership

• Model reflexivity: If your customers cut headcount (or reduce friction), what happens to your “seats,” ARR, or take rate? Don’t assume you’re only a beneficiary of AI.
• Plan for both productivity gains and demand shock: slower top-line, margin pressure, and political/regulatory risk (transfers, compute taxes, public claim on AI).

Civic / policy

• Support and debate serious proposals: transfers for displaced workers, funding (e.g. deficit, compute tax, or shared returns from AI), and automatic stabilizers redesigned for structural displacement.
• Vote and advocate for candidates who treat this as a structural transition, not a normal business cycle.

Mindset

• Combine both articles: One is “from the future” (2028); the other is “from 2026.” Use the combined step-by-step logic above as your mental model: adoption → reflexivity → loop → intermediation → labor → financial daisy chain → fiscal trap → policy race.
• Use “canary still alive”: If you treat the present as “before the loop has fully kicked in,” the priority is assessment and preparation (career, leverage, portfolio, policy) rather than panic.