AI-assisted people and learning deliverables

Start with the decision the artifact must support

Name the reader and the next decision before choosing a format. A competency map supports curriculum design. A facilitator guide supports delivery. A development plan supports a conversation between a participant and a manager. Combining all three into one document often hides missing evidence and unclear ownership.

The UNESCO AI Competency Framework for Teachers separates competencies across a human-centred mindset, ethics, AI foundations and applications, AI pedagogy, and professional learning, with progression from acquire to deepen to create (UNESCO teacher framework). That is a useful reminder to define the population, capability dimension, and expected progression rather than treating AI fluency as one undifferentiated skill.

Sources: [1]

Build a source-to-design matrix

For each learning objective, record the evidence source, intended behavior, learning activity, output expected from the learner, reviewer, and boundary. Separate published standards from internal observations and untested assumptions. If an objective has no observable behavior or evidence route, label it as a hypothesis instead of quietly turning it into a scored requirement.

The UNESCO student framework likewise distinguishes a human-centred mindset, ethics, techniques and applications, and system design, with understand, apply, and create progression levels (UNESCO student framework). These categories should be adapted to the workplace task and audience. They are not proof that a particular programme works or that one universal sequence fits every role.

Sources: [2]

Record the AI transformation

Attach a short transformation note to the deliverable. It should identify the source material provided to the AI, the requested operation, important exclusions, substantive edits accepted or rejected, and unresolved questions. A clean final document alone cannot show whether an unsupported activity, fabricated example, or misplaced learning objective entered during drafting.

The note can be lightweight: source IDs, prompt purpose, three material changes, and one reviewer decision. Do not preserve sensitive prompts or personal information merely to create an audit trail. Retain only what the programme's approved information-handling process permits.

Worked artifact: a learning-design evidence card

Consider a synthetic operations cohort learning to verify AI-assisted incident summaries. The design card names the target behavior as tracing every material statement to an incident record. The activity supplies two conflicting logs and an outdated summary. The learner produces a corrected briefing with a source table and an uncertainty note. The reviewer checks five predefined claims and records whether the conflict was surfaced.

The AI may help generate a first-pass exercise and facilitator prompts. The card records that the incident data are synthetic, a subject-matter reviewer approved the conflict pattern, and no production incident or employee record is used. A later transfer check asks whether participants use the same traceability behavior in a new synthetic scenario. Completion and satisfaction are recorded separately because they do not establish transfer.

Verify the deliverable before release

Use four checks. First, trace each learning objective to its source or explicitly marked design hypothesis. Second, test the artifact with a representative user for instructions, timing, and accessibility. Third, have a domain reviewer inspect technical accuracy and realistic constraints. Fourth, confirm that the assessment method creates a fair opportunity to demonstrate the stated behavior without requiring hidden knowledge.

Version the packet when objectives, sources, examples, scoring criteria, or delivery conditions change. A cosmetic edit does not require a new evidence decision, but a changed behavior or reviewer standard does.

Keep interpretation bounded

A people and learning artifact can organize evidence and make review more consistent. It does not establish that training caused performance improvement, forecast performance outside the assessed task, validate a benchmark, or make an employment decision. Two Prune can provide task-bounded assessment evidence from a controlled work simulation; customers retain responsibility for programme design, interpretation, and any workplace decision.