About
A clearer way to understand AI and work
The Task Graph helps people see how the work inside a role may change. It connects standardized roles to concrete work activities, matches public research at the task level, and keeps every source, gap, and limitation visible. The same foundation is being developed for organization-level workflow, skills, and responsible automation questions.
Individual pilot
Understand your role through the work you actually do.
Pilot participants describe their recurring work, review the proposed public matches, and receive a private report. The aim is a clearer view of what may change, what needs context, and which workflow or skills questions to explore. Places open in small groups and include a short feedback exchange.
Planned organization product
Build a work baseline before choosing an AI strategy.
The planned organization product would map organization-provided roles and recurring tasks to a comparable public baseline. With complete workload data, it could show where change is concentrated and help frame workflow, automation, oversight, and capability questions. It is not available yet, and personal report data would not be reused without clear permission and a separate agreement.
Who it is for
People who want to understand how their role may evolve and make more grounded choices about workflows and skills while keeping human context, judgment, and accountability visible.
HR, operations, transformation, and business leaders who need a sourced picture of roles, recurring work, and workload before setting AI and capability priorities.
Researchers and practitioners who need occupational data, task-level reasoning, and source provenance to remain visible.
Editorial boundary
The public role number is the OpenAI paper's beta technical- exposure score. E1 tasks count fully, E2 tasks count half, E0 tasks count zero, and core tasks count twice supplemental tasks. It is not a probability, measured share of work time, automation rate, or prediction about a person or job.
Public occupation data provides a shared baseline, not a complete description of every employer, workflow, tool, or local policy.
Modeled estimates remain estimates. We show the method, matched and unmatched tasks, original sources, and known limitations.
Limitations
How we handle uncertainty
Different public sources describe roles and AI use at different levels of detail. We preserve those differences instead of presenting them as one interchangeable signal.
ESCO · ESCO limitations
ESCO enriches The Task Graph with occupation mappings and skills/competences, but crosswalking between ESCO and O*NET is inherently imperfect. Match confidence is therefore surfaced in diagnostics and only strong links are treated as resolved coverage.
EconEvals · EconEvals limitations
The June 2026 EconEvals task estimates come from synthetic worker prompts and model-based task walkthroughs. They cover text-only chatbot use, exclude agentic, multimodal, and tool-using systems, and are grouped into broad ranges. The Task Graph therefore shows the source band, trial coverage, release, and exact task mapping rather than treating the estimate as observed automation.
O*NET · O*NET limitations
O*NET provides detailed occupation content and is the main role-level source in The Task Graph. It is a strong structured foundation, but it reflects the O*NET-SOC framework and should not be assumed to describe every country or employer equally.
OECD PIAAC · PIAAC limitations
PIAAC in The Task Graph is restricted to broad occupation-group indicators such as reading at work or ICT use at work. Those indicators add useful context, but they are not displayed as precise measures of a single named job title.