Can AI automate Epoch?
Summary
Epoch tests six models on 11 tasks drawn from its own work, including design, data analysis and research planning. Frontier systems handle many defined steps but miss organizational standards and research judgment needed to deliver complete work. The tests distinguish capable assistance from autonomous replacement.
Why it matters
End-to-end automation depends on the weakest parts of a workflow, including tacit standards and decisions about what is worth doing. Success on coding or other isolated steps does not establish that a whole job can be automated.
What to keep in mind
Tasks come from one organization, with one run per model-task pair and one human grader. Rubrics include subjective criteria, and models use different harnesses. Results do not measure economy-wide job exposure or the productivity of people working with agents.