AIPE Atlas
Chapters
Act I · The question

This is not a thought experiment.
The leading edge has already landed.

Aggregate employment is holding. But the displacement is real, uneven, and front-loaded onto one cohort: the young. Firms aren't firing juniors — they're quietly not hiring them, removing the bottom rungs of the career ladder.

Fig. 01
−13%
Early-career employment, AI-exposed jobs
Relative decline for workers aged 22–25 in the most AI-exposed occupations since late 2022, in ADP payroll microdata — while older workers in the same jobs held steady. Young software developers: nearly −20% from peak.
Stanford, "Canaries in the Coal Mine?" 2025–26
Fig. 02
40–60%
Of jobs exposed to AI
IMF: ~40% of global employment is exposed, ~60% in advanced economies — roughly half complemented, half facing downward pressure on wages and demand. "In most scenarios, AI will likely worsen overall inequality."
IMF Staff Discussion Note, 2024
Fig. 03
52 / 45
Augmentation vs. automation, real usage
Across millions of real AI conversations, augmentation-style use (~52%) still edges out automation (~45%) — but the automation share is large, rising, and dominant in enterprise API traffic.
Anthropic Economic Index, 2025

The deeper worry: if AI does the apprentice-level work through which juniors become seniors, the pipeline that produces future experts is being dismantled — quietly, and first.

Act II · The shift

Automation is a rising tide on the ladder of task complexity.

Everything below the waterline becomes nearly free. Everything above it earns a premium — until the water reaches it. The century turns on one question economists cannot yet answer: does the ladder have a top?

??? — is there a top rung?
Data entry & routine clerical the consensus first casualtypremium
Customer service & call centers agents augmented, then thinnedpremium
Junior coding, drafting, analysis the "canary" cohortpremium
Routine professional output standard legal, accounting, contentpremium
Structured physical work warehouses, assembly, logisticspremium
Expert cognitive work diagnosis, senior engineering, research supportpremium
Unstructured physical dexterity trades, repair, care settings — Autor's refugepremium
Trust-based interpersonal work care, therapy, teaching, leadershippremium
High-stakes judgment & accountability bearing responsibility no machine can carrypremium
Novel problem generation asking questions no one has askedpremium
Fig. 04 — The task ladder, after Korinek & Suh's complexity model. Rung order is indicative, not a forecast.
Set the era
Reading · 2025

The water laps the clerical floor

Office and administrative support, data entry, and routine customer service sit below the line — the strongest cross-source consensus (IMF, McKinsey, Goldman, OECD). Nearly everything else still earns a human premium. Displacement is real but narrow, and lands on the entry-level cohort first.

WEF: +78M net new jobs by 2030 Juniors hit first — quietly
Branch A · the ladder is endless

There is always a higher rung

Autor and Noah Smith's bet: human task-complexity is unbounded. New, harder, currently unimaginable work keeps appearing above the waterline — as it has for two centuries. Output and wages can rise forever. "We will run out of workers before we run out of jobs."

Branch B · the ladder has a top

The tide closes over the last rung

Korinek's warning: if human-performable complexity has a ceiling, wages can collapse below subsistence even while output explodes — and collapse can come before full automation if automation outruns capital accumulation. Horses had a comparative advantage too, until their "wage" fell below the cost of their feed.

Atlas hub Two Horizons →
This document is confidential and not printable. (R² Labs · AIPE · 2026-06-24)