Labor economists have said the same thing for years: AI doesn't replace jobs, it replaces tasks. A job is a bundle of tasks, and automating a few doesn't mean the whole bundle goes away.
But OpenAI and Anthropic had been focused on only capability - look what the model can do - and how it threatened jobs. Until the last few months. They've quietly started measuring economic impact as well, and are slowly arriving at the same conclusion: AI replaces tasks, not jobs.
So we rebuilt our AI-Proof Score, and captured a new AI Impact component of the score that better reflects economic dynamics like human necessity and demand elasticity. The new score is live on every page on this site, and the methodology page walks through how it works.
The score now follows OpenAI's AI Jobs Transition Framework. Instead of asking only what AI can do, it weighs three plain questions about a job:
How much of the day-to-day work a model can handle, and is already handling today.
Whether a person has to be there - by law, in the room, or to be the one accountable when something goes wrong.
When AI makes a job's output cheaper, demand sometimes grows instead of shrinking. More of the work to go around, not less.
AI being able to do part of your job does not mean the job is going away.
A role can score high on that first question and still be steady, because someone has to sign off on the result, or because cheaper work cuts both ways - AI does more per person, but the lower cost can raise demand enough that the role still grows.
For a while the loudest takes were gut reactions. A demo goes viral, and everyone extrapolates from it to the whole economy. That is starting to give way to slower, more careful work. Two pieces of it matter here, and they come at the question from opposite ends.
A framework that works out what makes a job hard to automate, from the three questions above.
OpenAI: The AI Jobs Transition Framework (April 2026)Millions of real Claude conversations, showing what people actually hand to AI, task by task, and whether it’s automating the work or assisting with it.
Anthropic Economic Index (March 2026)Our new methodology now uses both. The framework gives us the three questions to ask of every job. The Economic Index gives us the real usage data to answer the first one, which tasks AI actually handles and how well it does them. We score all three questions for each occupation and combine them into the AI Impact component, then weigh that against the labor-market data we already track.
We also moved to the newest Anthropic Economic Index (March 2026). It looks at how people use Claude Opus 4.5 and 4.6, and it adds learning curves - a measure of how much better AI gets at a task as people learn to use it well. The task-by-task breakdowns on every career page now reflect that newer data.
There is a "Give Feedback" button on every career page. The data can tell us a lot, but it cannot tell us what is happening in your specific role, at your company, this month. If a score looks wrong, or your situation has shifted, tell us. Real stories are what keep this honest.
If you want to see how the score is built, the methodology page walks through all three questions and the math behind each one. Start there, and check how your own role scores.