---
id: "lib-2026-04-00-ai-index-report-2026"
type: "library"
title: "AI Index Report 2026"
lang: "EN"
org: "Stanford HAI"
source_type: "Report"
date: "2026-04-01"
week: "2026-W14"
url: "https://nosov.ai/library/lib-2026-04-00-ai-index-report-2026/"
original_url: "https://hai.stanford.edu/assets/files/ai_index_report_2026.pdf"
tier: "S"
tier_name: "Shapes the field"
tier_reason: "Stanford HAI's annual AI Index, the field's reference dataset, steered by researchers including Erik Brynjolfsson, Jack Clark and James Manyika and built on data from Epoch AI, LinkedIn, Lightcast and McKinsey."
summary_basis: "full-text"
original_minutes: "600"
tags: ["ai state of field", "economics", "policy", "benchmarks"]
frames: ["verification-capacity"]
---

# AI Index Report 2026

Tier S (Shapes the field): Stanford HAI's annual AI Index, the field's reference dataset, steered by researchers including Erik Brynjolfsson, Jack Clark and James Manyika and built on data from Epoch AI, LinkedIn, Lightcast and McKinsey.

Summary based on the full text.

## In short

The ninth edition of the AI Index, published in April 2026 by the Stanford Institute for Human-Centered AI (co-chairs Yolanda Gil and Raymond Perrault, editor-in-chief Sha Sajadieh). Across about 425 pages and nine chapters (research, technical performance, responsible AI, economy, science, medicine, education, policy, public opinion), it compiles independently sourced data on AI in 2025. Its overall finding: capability and adoption are scaling faster than evaluation, governance, education and measurement can keep up with. New this year are standalone chapters on science and medicine and new estimates of labour market effects.

## Key numbers

- 88%: organizational adoption of AI in 2025
- nearly 20%: fall in employment of US software developers aged 22 to 25 since 2024, while older developer headcount grew
- 12% to ~66%: jump in AI agent task success on OSWorld, real computer tasks across operating systems
- $285.9 billion: US private AI investment in 2025, more than 23 times China's $12.4 billion

## What it reports

- Generative AI reached 53% population adoption within three years, faster than the PC or the internet; organizational adoption reached 88%. Singapore is at 61%, the UAE at 54%, and the US ranks 24th at 28.3%.
- Measured productivity gains of 14% to 26% in customer support and software development, with weaker or negative effects on judgement-heavy tasks. AI agent deployment remains in single digits across nearly all business functions.
- US developers aged 22 to 25 saw employment fall nearly 20% from 2024, while headcount for older developers kept growing.
- Agents went from 12% to ~66% task success on OSWorld but still fail roughly 1 in 3 attempts on structured benchmarks. SWE-bench Verified performance rose from 60% to near 100% of the human baseline in one year.
- Jagged frontier: Gemini Deep Think won IMO gold, but the top model reads analog clocks correctly 50.1% of the time. Robots succeed in only 12% of household tasks.
- Documented AI incidents rose to 362 from 233 in 2024; responsible AI benchmark reporting by frontier developers remains spotty.
- US private AI investment hit $285.9 billion, 1,953 newly funded AI companies; but AI researchers moving to the US dropped 89% since 2017, 80% in the last year.
- As of March 2026 Anthropic's top model leads the best Chinese model by just 2.7%. Industry produced over 90% of notable models.
- 73% of experts expect AI to improve how people do their jobs, against 23% of the public.

## How they got it

A Stanford team aggregates third-party data (Epoch AI, GitHub, LinkedIn, Lightcast, McKinsey surveys, Quid, Zeki and others), benchmark leaderboards, incident databases, government sources and public opinion surveys, plus its own analysis. Raw data and charts are published openly.

## Limits

- Mostly secondary data; the adoption figures depend on how each source defines "use", and some (McKinsey surveys) come from firms with a stake in AI.
- The report itself notes that benchmarks are saturating and that labs disclose less, so capability comparisons are less reliable than before.
- The entry-level employment figure is US software developers only and shows correlation, not cause.

## Why it matters for companies

Adoption numbers are high but agent deployment is still in single digits per function, so most companies are early regardless of what vendors claim. The clearest productivity evidence is in customer support and software, and those are the same areas where entry-level hiring is falling, which matters for workforce and graduate pipeline planning. A one-in-three failure rate on structured agent tasks means agents need checking and fallbacks before they touch real processes. The gap between expert and public expectations (73% versus 23%) is also a gap a company should expect inside its own workforce.

Original: https://hai.stanford.edu/assets/files/ai_index_report_2026.pdf