Break a role into the work you actually do. Estimate what AI could assist with, what still needs a person, and how your result changes under stronger-capability assumptions. Your ratings—not a hidden database—drive every number.
Exposure is not job-loss probability. This models tasks, not whether a role or person will disappear. It has no employer, labour-market or future-AI data; hiring, adoption, law, cost, accountability and demand are outside the maths. Treat the result as a conversation starter, not career or financial advice.
Add tasks that describe a normal week. There is no upload, account or automatic saving.
* Arithmetic from your own task ratings. These are not probabilities or measured labour-market outcomes.
What if the AI-capability rating you entered for every task were 25% or 50% higher? These are assumption stress tests only; they do not predict when or whether that change will happen.
Per task: AI-assisted hours = hours × your AI-capability rating. Potentially automatable hours = hours × your AI-capability rating × (1 − your human-essential share). The three sensitivity cards apply 1.00×, 1.25× and 1.50× to your capability ratings, capped at 100%; they do not change the human-essential share.
The example is invented to demonstrate the arithmetic, not an occupational study. Your tasks are held in volatile page memory and disappear when you leave. “Download private backup” saves the role and task labels you entered; protect that file. Copying the share text excludes task names and excludes your role title unless you opt in. Nothing is sent to a server.