---
id: "lib-2025-02-00-displacement-or-complementarity-the-labor-market"
type: "library"
title: "Displacement or Complementarity? The Labor Market Impact of Generative AI (HBS Working Paper 25-039)"
lang: "EN"
org: "Harvard Business School"
source_type: "Paper"
date: "2025-02-01"
week: "2025-W05"
url: "https://nosov.ai/library/lib-2025-02-00-displacement-or-complementarity-the-labor-market/"
original_url: "https://www.hbs.edu/ris/download.aspx?name=25-039.pdf"
tier: "A"
tier_name: "Strong evidence"
tier_reason: "Harvard Business School working paper by Chen, Srinivasan and Zakerinia using near-universe US job postings and a synthetic difference-in-differences design; clear method and checkable estimates, but an unrefereed draft."
summary_basis: "full-text"
original_minutes: "40"
tags: ["displacement", "complementarity", "job postings"]
---

# Displacement or Complementarity? The Labor Market Impact of Generative AI (HBS Working Paper 25-039)

Tier A (Strong evidence): Harvard Business School working paper by Chen, Srinivasan and Zakerinia using near-universe US job postings and a synthetic difference-in-differences design; clear method and checkable estimates, but an unrefereed draft.

Summary based on the full text.

## In short

HBS Working Paper 25-039 (draft dated December 2024) builds separate automation and augmentation scores for US occupations and tests them against job postings before and after ChatGPT's release. Firms posted fewer ads, asking for fewer skills, in occupations whose tasks are highly automatable. They posted more ads, with more and newer skill requirements, in occupations that mix automatable and non-automatable tasks. The authors read this as evidence that generative AI both displaces and complements labour, depending on task mix.

## Key numbers

- 17%: drop in job postings per firm per quarter for top-quartile automation-prone occupations after generative AI
- 22%: rise in job postings per firm per quarter for augmentation-prone occupations
- 24%: drop in generative-AI-exposed skills requested per firm per quarter in automation-prone jobs
- 15%: rise in generative-AI-exposed skills requested in augmentation-prone jobs
- 76,890: firm-quarter observations in the automation analysis sample

## What it reports

- Top-quartile automation occupations: job postings down 17% per firm per quarter relative to controls after generative AI's introduction.
- Augmentation-prone occupations: postings up 22% per firm per quarter.
- Generative-AI-exposed skills requested fell 24% in automation-prone jobs and rose 15% in augmentation-prone jobs; total and newly required skills moved the same way.
- Automation hits structured cognitive-task jobs; augmentation favours roles that combine automatable tasks with judgment or interpersonal work.

## How they got it

Task descriptions for 923 O*NET occupations (19,265 tasks) were scored by ChatGPT for automation potential, following Eloundou et al. (2024); an augmentation score was built from the mix of exposed and unexposed tasks. These were joined to Lightcast US postings from 2019 to June 2024, aggregated by firm and quarter, and analysed with synthetic difference-in-differences comparing top-quartile occupations with lower-quartile controls. ChatGPT also classified required skills as exposed or not.

## Limits

- Postings measure employer demand, not hires, wages or employment.
- Exposure scores come from a language model and inherit its judgment; the treatment date is the public release, not firm-level adoption.
- US only, through mid-2024; other shocks in 2023 (interest rates, tech hiring correction) may affect high-exposure occupations differently.
- Working paper, not yet peer reviewed.

## Why it matters for companies

The paper supports planning by task mix rather than by job title: roles that are mostly structured cognitive work are where hiring will contract, while mixed roles may need more people with broader skill profiles. For hiring teams, the skill-requirement shift is the actionable signal. Job descriptions for augmentation-prone roles should add AI-collaboration skills, and automation-prone roles are candidates for redesign before the next hiring round.

Original: https://www.hbs.edu/ris/download.aspx?name=25-039.pdf