Intelligence Disruption Index
Measuring how fast AI is disrupting human labor
The Intelligence Disruption Index tracks whether AI is displacing human workers — not just getting smarter, but actually taking jobs, compressing wages, and automating real work. It aggregates 19 signals across 4 categories into a single 0–100 score, updated monthly.
4 Categories, 19 Signals
Job Displacement
The core question: are jobs disappearing in sectors directly exposed to AI? Tracks cumulative employment decline in 9 AI-exposed occupations, Big 4 firm headcount (1.5M knowledge workers — first combined decline in 2025), real-time hiring demand from Indeed, information sector unemployment (5.0% vs 3.2% national), real wage compression, and business formation as a counter-signal.
BLS OEWS, Big 4 annual reports, Indeed Hiring Lab (CC-BY-4.0), FRED (BLS CPS, CES, Census BFS)
AI Capacity & Cost
How capable and affordable is AI? Two forces enable automation: intelligence (can AI do the task?) and cost (is it worth automating?). METR's time horizon measures the longest tasks frontier agents can reliably complete — currently ~870 hours of human-expert work, doubling every ~4 months. Meanwhile, frontier model prices collapsed from $120/MTok to $0.15/MTok in 6 years. AI is getting smarter AND cheaper simultaneously.
METR (metr.org/time-horizons), sanand0/llmpricing (OpenRouter data)
AI Penetration
Is AI being used in real work — and is it replacing or assisting? Over 50% of new web articles are AI-generated. Enterprise AI seats are growing exponentially. The Anthropic Economic Index reveals ~45% of AI interactions are full task automation (not just Q&A). When that crosses 50%, AI is doing more replacing than assisting.
Anthropic Economic Index, Microsoft/OpenAI/Google quarterly earnings, Graphite/Surfer, Ahrefs, arXiv REST API
Physical Automation
Are robots and autonomous vehicles replacing physical human workers? Amazon's robot-to-employee ratio went from 0 to 642 per 1,000 in 13 years. CA autonomous vehicle miles hit 4.5M in 2024. Warehouse and transport employment shares are tracked against their 2019 pre-AI baselines.
Amazon data, CA DMV, FRED (2 BLS series)
Scale
Methodology
The IDI aggregates 19 signals across 4 categories. Each signal is chosen for its proximity to AI-driven labor displacement. Some signals (like total job postings or sector unemployment) include broader economic context — they're weighted lower but help distinguish AI-specific trends from macro noise. Two counter-signals (wage growth and business formation) measure economic resilience against displacement.
All four categories are equally weighted at 25%. Our opinion lives in signal selection and normalization — not in category weights. Each category captures a distinct dimension: outcomes (Displacement, Physical Automation) and leading indicators (Capacity & Cost, Penetration).
The composite formula: IDI = 0.25D + 0.25C + 0.25P + 0.25A. Each category is equally weighted. Signals are normalized to 0-100 with economically meaningful anchor points, then weighted within their category. A 6-month EMA is applied for smoothing.
All data sources are free and public: BLS OEWS (9 occupations), FRED (BLS CPS unemployment), Indeed Hiring Lab (CC-BY-4.0, 7 AI-exposed sectors + total US), METR time horizons (frontier agent benchmarks), arXiv, Amazon press releases, CA DMV, Counterpoint/Omdia (humanoid robots), and published LLM pricing. Zero proprietary data.
Embed
Add the IDI widget to your website. Updates automatically.
Licensed under CC BY 4.0. Free to share and adapt with attribution — please link back to this page.