---
id: "lib-2026-04-00-the-ai-jobs-transition-framework"
type: "library"
title: "The AI jobs transition framework"
lang: "EN"
org: "OpenAI"
source_type: "Report"
date: "2026-04-01"
week: "2026-W14"
url: "https://nosov.ai/library/lib-2026-04-00-the-ai-jobs-transition-framework/"
original_url: "https://cdn.openai.com/pdf/the-ai-jobs-transition-framework_report.pdf"
tier: "A"
tier_name: "Strong evidence"
tier_reason: "OpenAI Economic Research (Alex Martin Richmond, with a note from chief economist Ronnie Chatterji) adds human necessity and demand elasticity to exposure across 900+ occupations and checks the result against ChatGPT usage data; method is described, but elasticities are GPT-generated and the lab grades its own product's footprint."
summary_basis: "full-text"
original_minutes: "30"
tags: ["task exposure", "transition pathways", "jobs"]
---

# The AI jobs transition framework

Tier A (Strong evidence): OpenAI Economic Research (Alex Martin Richmond, with a note from chief economist Ronnie Chatterji) adds human necessity and demand elasticity to exposure across 900+ occupations and checks the result against ChatGPT usage data; method is described, but elasticities are GPT-generated and the lab grades its own product's footprint.

Summary based on the full text.

## In short

An April 2026 OpenAI Economic Research report arguing that AI exposure scores are too blunt to predict near-term job effects. It combines theoretical exposure (Eloundou et al. 2023) with two new occupation-level measures, human necessity and demand elasticity, and validates the mix against anonymised work-related ChatGPT usage. The result sorts 900+ occupations covering about 152 million US jobs into four archetypes: higher automation risk, reorganize, grow with AI, and less immediate change.

## Key numbers

- 18%: share of jobs at relatively higher short-term automation risk
- 24%: share of jobs expected to reorganize, with declining employment but workers still needed
- 12%: share of jobs that could grow because lower effective cost expands demand
- about 5x: ChatGPT usage in jobs labelled automation-susceptible vs jobs facing less immediate pressure
- 22.8% vs 91.0%: realized vs theoretical exposure for jobs at high automation risk (68.1 pp capability overhang)

## What it reports

- Shares of jobs: 18% higher automation risk, 24% reorganize, 12% grow with AI, 46% less immediate change.
- Human necessity has three types: regulatory or accountability (judges, lawyers), relational (teachers, nurses) and physical (plumbers, physical therapists). Court reporters rank top 5 on some exposure measures but a human must still certify transcripts in most jurisdictions.
- Demand elasticity is the link between productivity and employment. Inelastic examples: firefighters, home health aides (-0.8), childcare workers (-0.7). Elastic examples: graphic designers, software developers.
- Elasticity shows minimal correlation with theoretical exposure, so exposure rankings miss whether productivity gains compress or expand employment.
- Capability overhang by archetype: grow with AI 24.6% realized vs 92.8% theoretical; high automation risk 22.8% vs 91.0%; reorganize 18.4% vs 76.7%; less immediate change 3.6% vs 17.8%.
- Unemployment from 2024 Q1 to 2026 Q1 rose across all archetypes, by +0.2 pp for high automation risk and +0.6 pp for less immediate change, so no displacement signal shows up yet.
- Policy by archetype: transition support and early warning for high-risk jobs; staffing standards and guardrails for reorganizing jobs; capacity-building and procurement reform for growing jobs; monitoring for the rest.

## How they got it

Theoretical exposure comes from Eloundou et al. (2023) task scores weighted by task time. Realized exposure maps second-half 2025 work-related consumer ChatGPT conversations to O*NET tasks and compares usage against expected task share, capped at fully realized. Human necessity uses structured rules on occupation titles and descriptions, with GPT-5.4 categorising occupations. Elasticity asks GPT-5.4-mini for the demand response to a 10% price fall over two to three years, from O*NET profiles. A correction on 30 April 2026 fixed swapped archetype labels for a subset of occupations.

## Limits

- Elasticities are model-generated priors, which the authors call the least directly observed part of the framework.
- Usage data covers consumer ChatGPT only, owned by the report's author organisation, and cannot be checked externally.
- Occupations are treated as task bundles; bottlenecks within workflows, general-equilibrium wage effects and cross-occupation spillovers are not modelled.
- The authors say the categories are not job-loss forecasts. Further robustness work is promised in a technical appendix not yet published.

## Why it matters for companies

The four archetypes give workforce planners a better first cut than exposure heatmaps: ask whether a person must stay accountable, and whether cheaper output would sell more. Roles where neither holds are where headcount pressure comes first; roles with elastic demand are candidates for growth, not cuts. The large gap between theoretical and realized use also says most of the change is still ahead, so plans built on today's adoption rates will undershoot.

Original: https://cdn.openai.com/pdf/the-ai-jobs-transition-framework_report.pdf