🤖 AI, Automation and Work

About this simulation

Headlines about AI often ask "which jobs will disappear?" — but occupations are bundles of many distinct tasks, and modern automation research (in the tradition of Autor, Levy & Murnane, and more recent large-language-model exposure studies) shows that AI capability spreads unevenly across those tasks. This grid places eight occupations against five representative tasks each, so you can watch automation creep task by task instead of job by job.

How it works

• Every task cube carries an exposure score — the AI capability level at which that specific task becomes cost-effective to automate — and a time-share (its bar height) showing how much of the job it occupies.
• Task-level model — the realistic view. Each cube flips from human (blue) to automated (violet) independently once the capability slider passes its own exposure score, so most occupations end up partly automated, not wholly replaced.
• Whole-job model — the naive view some commentary implicitly assumes. An entire occupation flips at once based on its time-weighted average exposure, exaggerating both job losses and job "safety."
• The side bars chart the time-share automated per occupation, updating live as you move the slider.

Controls

• AI capability — a stand-in for how capable general AI systems become (0–100%).
• Focus occupation — isolate one occupation's task grid and dim the rest.
• Task-level / Whole-job — switch between the realistic and naive automation models to see how much the whole-job framing overstates displacement.
• ↺ Reset — restores default capability and view. Drag to rotate, scroll to zoom.

Did you know?

Research on generative-AI task exposure (e.g. Eloundou et al., 2023) estimates that around 80% of the US workforce has at least one task where AI could reduce completion time by half — yet only a small fraction of workers have the majority of their tasks exposed, which is exactly the task-versus-job distinction this simulation visualises.