Every number on this site right now was generated by an AI describing jobs to itself. That's a reasonable start — the task decompositions are realistic, the classifications are structurally sound, and the model has read more job descriptions than any human researcher ever will. But it hasn't lived in any of these roles. It doesn't know what it feels like to sit in a meeting where the real decision happened in the ten minutes before the meeting started, or what makes a particular client trust you and not your equally-qualified colleague, or why the quarterly report takes three days when technically it should take three hours. The synthetic data captures the skeleton. It misses the tissue.
That gap is the whole point. We built this to understand what AI can and can't replace — not from the outside, which is how every economic model approaches it, but from the inside. From people who do the work describing what the work actually requires. Every real assessment that comes in adds something no model can generate: ground truth. The difference between what a job looks like on paper and what it takes in practice. That difference is exactly where automation risk is routinely overestimated, and where human advantage is routinely invisible.
This is also why the stakes are collective, not just personal. When you map your role here, you're not just getting your own score — you're contributing to a shared picture of where we're all headed. AI development is moving fast, workforce policy is lagging, and most of the data shaping decisions about both comes from people who aren't doing the jobs. The only corrective is more signal from more workers across more industries, honestly describing what their days actually involve. The larger that dataset grows, the harder it becomes to make sweeping claims about automation that don't account for the informal, relational, and contextual layers that keep showing up as the real moat.
Understanding this is how we get ahead of it. Not by slowing AI down — that's not the choice — but by building an accurate enough picture that the people affected by these transitions can see them coming, name what's actually at risk, and make decisions about what to learn, change, or protect. The map only works if the people who know the territory help draw it.
One more thing. This project is a solo effort, and it costs real money to run — API calls, hosting, the hours that go into building it. I'm funding it from savings because I believe it matters, and I'm at the point where collaboration of any kind makes a genuine difference. If you spot a bug or something that doesn't feel right, report it. If you have an idea for how this could work better, send it. If you find this useful and want to help keep it going financially, that would mean more than I can say. And if you simply tell someone else about it — a colleague who's been wondering about their own exposure, a researcher who works on this, anyone — that helps too. The dataset grows one honest assessment at a time. So does everything else here.
— Robert