Purpose and fit

Use this in a developmental conversation when a colleague can produce an AI-assisted output but needs help explaining when to trust, question or reject it. Examine a bounded work example together. The aim is to improve the reasoning and support around the task, rather than reward a polished result or confident presentation. Agree that the conversation is for learning and how its notes will be used.

Before you start

Use a learner-selected example that already exists, with authorised, anonymised excerpts of the brief, output and any available corrections. If no output can be shared, describe the task and judgment question to obtain a general conversation outline. A fictional illustration is optional, not required data. Agree criteria, support and note-use boundaries.

Expect a developmental conversation outline, blank coaching note and proposed practice activity. These are learning aids, not an evaluation of a person; answers, commitments and observed performance must not be invented.

Required inputs

  • A learner-selected, authorised example with whatever brief, source material, output and corrections are available; if sharing is restricted, describe the task to obtain a general outline instead.
  • The intended recipient, acceptance criteria and current tool-use boundaries.
  • A private discussion space, agreed note access and approximately 25 minutes, adjusted for access and communication needs.
  • One feasible practice opportunity and a person available to help before the next conversation.

Practical workflow

  1. Open with the learner's question. Ask, “Which moment required the most judgment?” Agree one moment to examine and invite the learner to correct your assumptions about the task.
  2. Reconstruct the sequence before evaluating it: what the learner knew, what they asked, what the system returned and what happened next. Separate recollection from evidence still available.
  3. Ask for the reasoning at that moment: “What made this input suitable? What would have changed your mind? What did you check directly?” Allow time to inspect the source rather than demanding an immediate answer.
  4. Explore one counterexample together: a convincing unsupported claim, a conflicting source or an unavailable reviewer. Ask what observable signal should change the response. Share your own uncertainty where relevant.
  5. Choose one small practice experiment, such as tracing two material claims before accepting a draft. Specify the support, safe task, evidence to retain and condition for stopping or seeking help.
  6. Close with a learner-written summary and a review date. At the next conversation, inspect what happened and revise the experiment, including any tool or workload problem that prevented it.

Expected output

A short coaching note and one practice commitment. Copy these fields into a document the learner can review:

Example/task: [ ]. Learner’s question: [ ]. Judgment moment: [ ]. Evidence inspected: [ ]. Reasoning described: [ ]. Remaining uncertainty: [ ]. Alternative considered: [ ]. Practice experiment: [ ]. Coach support: [ ]. Observable learning evidence: [ ]. Stop/help condition: [ ]. Review date: [ ]. Note access and retention: [ ].

Useful evidence is specific: “identified an unsupported statement and removed it after checking the source.” Avoid labels such as “AI-ready” that reach beyond the observed example.

Human checks

The learner checks the note for accuracy. The coach distinguishes a knowledge gap from unclear instructions, inaccessible tools or unrealistic workload. A domain specialist handles questions outside the coach’s competence. Do not upload identifiable coaching records to an unapproved service or use generated interpretations to infer motivation, personality or employment suitability.

Limitations

One conversation cannot establish broad competence or explain why every mistake occurred. Hierarchy can inhibit candour; offer another learning contact where appropriate. This is a proposed coaching aid, not a validated assessment or a basis for automated employment decisions.

Starter prompt

Prepare a developmental coaching conversation for [task] using this sanitised example: [brief, sources, output and corrections]. The learner’s question, permitted tools, acceptance rules and available support are [details]. Exclude identifiable employee records.

Provide a 25-minute outline, open questions about reasoning and evidence, and one hypothetical counterexample. Then create a blank note with fields: Task; learner’s question; judgment moment; evidence inspected; reasoning described; uncertainty; alternative considered; practice experiment; coach support; observable learning evidence; stop/help condition; review date; note access and retention.

Suggest one bounded practice experiment. Separate facts from possible explanations. Do not invent learner answers, diagnose the learner or recommend an employment action. Leave commitments and dates blank unless supplied.

Adapt the requested fields to the authorised inputs actually supplied. If corrections or sources are missing, ask what evidence could be examined rather than diagnosing the learner. Keep conflicting recollections separate from records. Tailor timing and questions to supplied access needs, without inferring sensitive personal traits. State the expected output and acceptance criteria before drafting. Ask for essential missing information; where a useful partial draft is possible, label its limits and assumptions. Do not invent facts, evidence, results or approvals.

Data, model variability and your responsibility

Use relevant data that exists or that you provide, with accuracy and completeness sufficient for the task. Different AI models and versions may produce different results, and responses may vary between runs. You are responsible for checking and validating the output against your own requirements, source information and acceptance criteria before relying on it.