"AI will change work" is a big statement. "Which tasks are people actually using it for?" is a much more useful question.
What the first index measured
Anthropic launched its Economic Index on February 10, 2025. Its initial analysis used anonymized Claude conversations to study work-related tasks. The company reported concentrated use in software development and technical writing, with collaboration more common than fully automated task patterns in that dataset. The original announcement
This is evidence from one platform's users, not a census of all workers or proof that a particular job has disappeared.
My reading: break the job into pieces
Imagine an illustrative customer-support role. It might include finding policy details, drafting replies, handling emotionally difficult situations, making exceptions, and coordinating with another team.
A drafting assistant could change one of those tasks without taking responsibility for the whole role. That distinction matters when a headline turns a task-level result into a prediction about everyone's career.
For a team evaluating AI, I would write the task inventory before the productivity pitch:
| Task | A useful pilot question |
|---|---|
| Find information | Can the answer point to the correct current policy? |
| Draft a reply | Does the reviewer spend less time reaching an acceptable result? |
| Handle an exception | Who is authorized to decide, and is that decision recorded? |
These are hypothetical evaluation questions, not measured results from the index.
Do not confuse adoption with improvement
My own test would include review effort and mistakes, not only whether people clicked an AI button. A tool used frequently might still create extra checking work. A tool used rarely might be extremely helpful for a particular difficult task.
The most useful response to this kind of research is curiosity rather than panic. Ask what was measured, who was represented, and what changed in the workflow. Then compare that evidence with a small, clearly scoped experiment in your own context.
The job title is the headline. The task is where the analysis gets interesting.
