---
type: source
title: "Atlanta Fed / FRBSF: AI Productivity Gains of 0.8% in High-Skill Services vs 0.4% in Low-Skill — Gains Expected to Double in 2026"
author: "Federal Reserve Bank of Atlanta / San Francisco Fed"
url: https://www.atlantafed.org/research-and-data/publications/working-papers/2026/03/25/04-artificial-intelligence-productivity-and-the-workforce-evidence-from-corporate-executives
date: 2026-03
domain: health
secondary_domains: [ai-alignment]
format: research
status: unprocessed
priority: medium
tags: [ai, productivity, workforce, economic-research, high-skill-concentration, federal-reserve]
intake_tier: research-task
---

## Content

Federal Reserve Bank of Atlanta / FRBSF research paper "Artificial Intelligence, Productivity, and the Workforce: Evidence from Corporate Executives" (March 2026 — companion to NBER Working Paper 34836).

Key sector-level findings (2025 actual data, not executive predictions):
- High-skill services and finance: ~0.8% labor productivity gain from AI
- Low-skill services, manufacturing, construction: ~0.4% gain
- Knowledge-intensive industries with AI job posting surges accounted for 50% of real GDP growth in Q3 2025
- Total factor productivity increases associated with innovation and demand-oriented channels (not capital deepening)

FRBSF Economic Letter (Feb 2026) additional data:
- Most macro-studies find limited evidence of significant AI effect in aggregate productivity statistics
- AI's GDP contribution is currently flowing through INVESTMENT (AI capex) not productivity gains
- "Solid, above-trend growth" expected for H1 2026 partly from AI-related investment

AI adoption concentration pattern (IMF Jan 2026 / PWC data):
- Higher education levels significantly more likely to demand AI-related skills
- Young workers' employment more concentrated in occupations with high AI exposure AND low complementarity to AI → higher displacement risk
- Areas with higher literacy, numeracy, and college attainment see more AI skill demand
- Entry-level positions facing pressure from AI in highly exposed occupations

San Francisco Fed Mary Daly (Feb 2026): AI productivity gains moving "under the hood" — present but not yet visible in standard productivity statistics.

## Agent Notes

**Why this matters:** This is the supply side of the AI-vs-chronic-disease argument. The Fed data shows that where AI gains ARE happening, they're concentrated in exactly the sectors and workers LEAST burdened by chronic disease (high-skill, finance, knowledge workers). The 0.8% vs 0.4% sector split is small but the directional signal is consistent: AI productivity accrues to already-healthy, already-productive workers.

**What surprised me:** Knowledge-intensive industries drove 50% of real GDP growth in Q3 2025 despite being a minority of employment. This is the AI productivity flying through the high-skill conduit while the rest of the economy sees 0.4% or nothing. The GDP numbers look good but the distribution is highly unequal.

**What I expected but didn't find:** A direct comparison of AI productivity gains among workers WITH vs WITHOUT chronic conditions. This is the research gap — we have sector-level data (high-skill vs low-skill) as a proxy, but not direct health-status-segmented data.

**KB connections:**
- Companion to NBER 34836 (80% no AI gains)
- Strengthens Belief 1 disconfirmation target: AI gains concentrated where chronic disease is least, chronic disease concentrated where AI is least — non-overlapping
- The 50% of GDP growth from knowledge-intensive industries creates a paradox: population health (which is declining) may not be the binding constraint on GDP in the near term if capital and knowledge work can decouple from population health status
- HOWEVER: this decoupling is temporary if knowledge workers eventually age and become chronically ill without prevention

**Extraction hints:**
- This source is better used as supporting evidence for the NBER claim than as a standalone claim
- The most extractable finding: "AI productivity gains concentrate in high-skill sectors at 0.8% vs low-skill sectors at 0.4% — a 2x differential that mirrors the chronic disease burden distribution"
- OR: flag this as the GDP paradox — short-term AI can inflate GDP growth measures even as population health declines, which may create a false signal that health is not a binding constraint

**Context:** Fed research has high methodological credibility. The FRBSF economic letter (shorter format, policy-oriented) and the Atlanta Fed working paper are companion pieces — both using the same underlying executive survey.

## Curator Notes (structured handoff for extractor)
PRIMARY CONNECTION: Companion to NBER 34836 on AI-vs-chronic-disease interaction for Belief 1
WHY ARCHIVED: Provides the sector-level quantification (0.8% vs 0.4%) and the GDP growth concentration finding (50% from knowledge-intensive industries). Together with NBER 34836, this builds the case that AI productivity is a high-skill phenomenon that doesn't compensate for low-skill chronic disease burden.
EXTRACTION HINT: Use as supporting evidence for the NBER 34836 claim rather than standalone. The 50% GDP growth concentration finding is the most surprising data point.