AI-enabled work by role and function

Map tasks before making claims about a role

A job title contains many tasks with different exposure, consequences and evidence needs. Start with recurring units of work such as analyzing a dataset, drafting a policy, reviewing a proposal or preparing a customer response, then describe how AI changes each unit.

The ILO's 2025 exposure research works at task level and finds substantial variation within occupations. Its central distinction between exposure and automatic replacement matters for practical role design: an AI-capable task does not mean an entire role disappears or that every part of the task should be delegated.

For each task, record the objective, inputs, decision points, acceptable assistance, required checks and final accountable owner. This produces a more useful capability map than a list of generic prompts attached to a title.

Sources: [1]

Define evidence and failure modes for the work

The same broad capability looks different across functions. Evidence navigation for a policy analyst may involve authority and effective dates; for a finance analyst it may involve periods, definitions and reconciliations; for a marketer it may involve attribution and denominator choices.

A role page should therefore name representative artifacts and the failures a reviewer needs to catch. It should distinguish an illustrative scenario from a shipped product capability and avoid implying that one benchmark applies equally across all functions.

Skills England's foundation benchmark is deliberately cross-cutting, but role design must add context, task difficulty and consequences. A shared core can support comparison while the scenario supplies the realistic material needed to observe how the behavior appears in practice.

Sources: [2] [3]

Use role maps to design practice and review

A role map becomes useful when it changes learning or workflow design. Select a small number of consequential tasks, create realistic practice with representative evidence, and give feedback on both the delivered artifact and the process used to reach it.

OECD research emphasizes that workplace AI changes the mix of digital, data, managerial and human skills. Development should therefore address framing, evidence, verification and communication alongside tool operation rather than assuming technical familiarity is sufficient.

Interpret the resulting evidence within the task. A strong performance can identify demonstrated behaviors and next challenges; a weak performance can identify follow-up questions and development areas. Neither result should be converted into an automated employment decision.

• Name the task and decision before selecting an AI technique.
• Use representative evidence and a realistic deliverable.
• Specify the principal failure and a practical verification method.
• Keep conclusions bounded to the observed work.

Sources: [4] [1]