Skip to main content

Structured institutional evidence

AI Workforce Transition Tracker

See how AI is changing real work, how organizations are responding, and what the evidence actually supports.

For the conceptual framework behind these evidence cases, read the workforce transformation framework.

Dataset context

41 published evidence records

For workforce leaders, policymakers and researchers: compare documented changes in work and the limits of available evidence.

How we verify evidence

Current findings

What the reviewed evidence shows

Reviewed finding

AI is changing work before it clearly changes workforce size

Across the current evidence base, AI is already compressing, substituting and reorganizing specific tasks, while corresponding changes in workforce size are often not directly evidenced. Workflow change and headcount effects should be treated as separate questions.

7 supporting evidence cases

Explore evidence
Evidence boundary

This does not establish that AI generally has little employment impact, or that workforce reductions will not occur. It means only that within the current published evidence base, task and workflow changes are often more directly evidenced than net workforce effects.

Reviewed finding

Measured AI gains are clearest at the workflow level

The strongest quantified outcomes in the Tracker are tied to concrete work units—coding, review, planning, reporting and logistics. Institutions evaluating AI should therefore measure changes in specific workflows, not rely only on enterprise-wide adoption metrics.

6 supporting evidence cases

Explore evidence
Evidence boundary

These metrics come from different organizations, functions and measurement methods. They must not be aggregated into a common productivity rate or treated as an expected return from AI generally.

Reviewed finding

Human authority remains explicit in some high-stakes AI workflows

Several healthcare and public-sector cases combine AI delegation with explicit human control points. Where the evidence is strongest, humans retain confirmation, review or final-judgment authority.

5 supporting evidence cases

Explore evidence
Evidence boundary

Do not generalize this into a claim that human oversight is the norm across AI deployments. Human decision rights are not documented for many Tracker Signals.

Evidence explorer

Browse evidence cases

5 of 41 visible

Filter the evidence

Active

Mengniu

China Dairy and agriculture

Mengniu dairy employee builds herd-care advisory tool

Explicit outcome claims

AI mechanism → work change

AI-assisted work Task Compression

How the organization responded

Not yet evidenced

What changed

A dairy employee without programming background assembled a knowledge-based tool that helps frontline farm workers seek herd-health guidance.

Key outcome

Reported per-farm diagnostic/advice turnaround — Reported operational figure from a multi-farm test; no independent audit or causal study identified. — Before 1.5 days reported elapsed time · After approximately 2 hours reported elapsed time

Tencent

China Technology & software services

CodeBuddy cuts coding time across engineering workflows

Explicit outcome claims

AI mechanism → work change

AI agents execute multi-step work Task Compression

How the organization responded

Not yet evidenced

What changed

CodeBuddy is embedded across Tencent engineering workflows and materially reduces coding time.

Key outcome

Overall coding time — Reduced by 40%

NTT DATA

Japan Technology & software services

Agent-supported RFP diagnostic workflow

Explicit outcome claims

AI mechanism → work change

AI agents execute multi-step work Task Compression

How the organization responded

Not yet evidenced

What changed

A specialized AI agent performs initial expert-style review of difficult non-functional requirements and surfaces risks and action plans before human delivery work continues.

Key outcome

RFP-review lead time — Reduced by approximately 60% while maintaining or improving quality

Walmart

United States Retail & e-commerce

AI-supported frontline shift planning

Explicit outcome claims

AI mechanism → work change

AI-assisted work Task Compression

How the organization responded

Not yet evidenced

What changed

AI recommends and prioritizes frontline tasks, reducing the time team leads spend building overnight stocking plans.

Key outcome

Team-lead shift-planning time — Reduced from 90 minutes to 30 minutes — Before 90 minutes per planning cycle · After 30 minutes per planning cycle

The Permanente Medical Group / Kaiser Permanente

United States Healthcare

Ambient AI drafts visit notes while physicians retain clinical review

Explicit outcome claims

AI mechanism → work change

AI-assisted work Task Compression

How the organization responded

Not yet evidenced

What changed

Ambient AI listens to physician-patient conversations and creates draft visit summaries, reducing manual documentation work while physicians retain review and editing authority.

Key outcome

Physicians using ambient AI documentation — 7,260 physicians used the tool between October 2023 and December 2024

Evidence landscape

Each case is counted once by its primary work change. Evidence stages describe what sources report, not independent proof of causation.

Work changeIn operationObserved work changeExplicit outcome claims
Task CompressionAI reduces the time or effort required within an existing task.0 cases2 cases5 cases
Task SubstitutionAI performs a task that was previously performed by people.1 cases1 cases4 cases
Workflow RedesignThe sequence, handoffs, or coordination of work changes around AI.3 cases6 cases8 cases
Role RedesignThe responsibilities or boundaries of a role change as work is reorganized.1 cases3 cases7 cases

Methodology and updates

The Tracker reviews public sources and distinguishes operational evidence, observed work change and reported outcome claims. Claims are not necessarily independently verified or caused by AI. Records are reviewed periodically as evidence becomes available.

Read methodology v1.0