What is applied AI capability at work?

Capability is more than prompting

Prompting is one part of AI-enabled work. A capable worker must also choose an appropriate task, supply relevant context, navigate source material, notice uncertainty, test important outputs and integrate the result into a usable deliverable.

The OECD and European Commission describe AI literacy through knowledge, skills and attitudes that support critical, ethical and creative use. Skills England similarly combines technical, non-technical, responsible and ethical workplace skills. Neither model reduces capability to command syntax.

A person can write fluent prompts and still solve the wrong problem. Another can use a simple interaction effectively because they understand the decision, bring the right evidence and perform the necessary checks. Evaluation should keep those distinctions visible.

Sources: [1] [2]

Observe connected capabilities in the work

A practical observation model follows the work from framing through delivery. It examines context framing, evidence navigation, AI orchestration, verification and risk control, judgment and synthesis, and communication and delivery as connected dimensions rather than isolated tricks.

These capabilities interact. Weak evidence selection constrains the best possible synthesis. Verification can reveal that an assumption must change. Communication can expose whether the worker understands the limitations of the result or merely repeats the model's confidence.

The task must create a reasonable opportunity to show each behavior. If a scenario contains no calculation, the absence of spreadsheet checking is not evidence of weakness. Missing opportunities and missing records should reduce confidence in the interpretation.

Sources: [2] [1]

Choose the measure that matches the question

Knowledge tests, self-reports, usage data, work samples and realistic simulations answer different questions. Organizations should combine them deliberately and avoid treating one form of evidence as an automatic proxy for another.

A knowledge test can show whether someone recognizes a principle. Usage data can show interaction frequency. A work sample can show the quality of a result. A realistic task can create a consistent opportunity to examine how the person frames, uses evidence, works with AI and verifies the final output.

No single task establishes a permanent trait. Interpret results within the scenario, state the evidence strength and use follow-up observations when a decision needs broader confidence. This keeps capability measurement useful without overstating precision.

• Define the work context.
• Identify the behaviors that matter.
• Create a fair opportunity to demonstrate them.
• Connect interpretations to observable evidence.
• State missing evidence and limits.

Sources: [1] [2]