About this project

Why this exists

There's no single place where people can see how exposed their job is to AI. This project was built to change that: one application that maps AI capability across the entire workforce and tracks how it shifts over time, giving a clearer picture of where automation is actually heading.

What it does

This tool analyses every job in the U.S. economy, nearly 1,000 occupations broken into 19,600+ individual tasks, and scores each one for AI automation capability. Data comes from the U.S. Department of Labor's O*NET database, enriched with employment and wage data from the Bureau of Labor Statistics (BLS), and scored by AI. Updated monthly.

How scores work

Each task within an occupation is scored on a 3-point scale:

• 0Still Human - AI cannot do this task. It requires physical presence, complex judgement, or deep human interaction.
• 1Human in the Loop - AI can partially do this task but still needs human oversight.
• 2Fully Automatable - AI can perform this task end-to-end without human involvement.

The overall AI exposure score is a weighted average using task importance ratings from O*NET. A score of 100% would mean every task is fully automatable; 0% means none are.

What the score means

A higher score means a larger share of tasks in that occupation appear technically automatable by current AI tools. It does not mean those tasks are already automated in real workplaces.

Use these scores as a signal for exploration, not a definitive verdict.

Data sources

• •O*NET 23.1 - U.S. Department of Labor's occupational database providing task definitions, skill requirements, and importance ratings
• •BLS OEWS - Bureau of Labor Statistics employment counts and wage data (May 2024 estimates)
• •AI scoring - OpenAI GPT-5 Mini, re-scored monthly to track evolving AI capabilities

What it doesn't do

This is not a prediction tool and does not forecast job losses. Scores reflect current AI technical capability only: what AI could do, not what it is doing. Many factors beyond technology determine actual workplace adoption: regulation, cost, trust, organisational inertia, and human preference.

Known limitations

• •Capability is not adoption - Real-world uptake depends on cost, regulation, risk tolerance, workflow redesign, trust, and customer preference.
• •Task interpretation uncertainty - Some task descriptions are broad or context-dependent, and scoring can vary by interpretation.
• •Model sensitivity - Scores depend on model behaviour and prompt design; different models or rubric choices may produce different outputs.
• •No firm-level context - The index does not include company-specific systems, internal controls, legal constraints, or local labour conditions.
• •Occupation-level abstraction - Roles vary across industries and employers; a single occupation score cannot capture every real job variant.
• •U.S.-centric data - Results are based on O*NET and BLS data and are most valid for U.S. occupational structures.
• •Point-in-time measurement - Scores reflect capability at the time of scoring and may change as models and tools improve.

How to use this responsibly

Use this index to:

• •Identify where AI pressure may be increasing
• •Compare occupations at a high level
• •Guide deeper investigation

Do not use it as a sole basis for employment, compensation, or policy decisions.

Built independently

This is an independent project, built and maintained as a public resource. The dataset is re-scored monthly, and the methodology will continue to evolve. If you spot an issue or disagree with a score, please share feedback: critique improves the index.

Who built this

Hi, I'm Kyle, a software consultant based in the Algarve, Portugal. I currently consult for an AI company that exclusively uses AI tools to build its products. This project started from a simple question: how exposed is my own job to AI? When no single tool could answer that clearly, I built this one.

If you have any questions or feedback, you can reach me directly at kyle.bowden@bittwisted.co.uk.