Introducing Atlas of Work
A shared map of occupations, workflows, and the evidence behind AI capabilities.
Ask whether AI can do a job, and the most useful answer often starts with another question: which part of the job?
A title can hide an extraordinary variety of work. The same person may gather information, interpret an exception, coordinate with a colleague, make a judgment, and document a decision. Those activities depend on each other. They also depend on a setting that a short job description rarely captures.
We’re building Atlas of Work to make those connections easier to explore.
Start with the work
The organizing idea is simple: move from an occupation to its workflows, from a workflow to its tasks, and from a task to the evidence relevant to it.
Each level answers a different question. An occupation provides a starting point. A workflow explains how activities fit together. A task brings the discussion down to a concrete action. Evidence helps us examine what an AI system has actually been assessed on.
Moving between these levels matters. A task without its workflow loses context. A benchmark without its task is hard to interpret. A broad claim about an occupation can conceal more than it explains.
Keep the sources close
Atlas is built around a distinction between the structure of work and the evidence about AI capabilities. A benchmark result is evidence about a particular assessment, under particular conditions. Connecting that assessment to a task does not establish that an entire occupation can be automated.
Our aim is to make the connection inspectable. What was the task? What did the assessment require? What source supports the result? How closely does that setting resemble the work someone is trying to understand?
Those questions are useful even when the available evidence is incomplete. Sometimes the most valuable finding is knowing where a conclusion would require more investigation.
Make room for practitioners
People who do the work can see details that a taxonomy misses. They know which exceptions are routine, which handoffs are fragile, and when a procedure depends on judgment that is difficult to write down.
That knowledge belongs in a shared understanding of work. Our approach to contributions is grounded in review: make the proposed change explicit, attach context where possible, and distinguish a suggestion from an accepted update.
A useful atlas should become more precise as people challenge it. It should make uncertainty visible and let readers examine the reasoning behind a connection.
A place to ask better questions
Atlas is for practitioners examining their own field, teams evaluating AI for concrete workflows, and researchers looking for gaps in the evidence.
The common starting point is curiosity about how work gets done. Instead of compressing an occupation into a single capability claim, we want to make its parts and their relationships visible.
We’ll use this space to share the thinking behind Atlas: how to describe workflows, how to read assessments, and how to keep a changing map useful. Begin with why work is more than a job title or our approach to reading AI evidence.