1 Introduction: A Defensible Classroom Pilot

Start with a learning decision you can examine.

1.1 The problem this book addresses

A business educator may want students to practise difficult decisions more often than a timetable allows: asking a fair question, checking a recommendation, explaining an uncertainty or revising an unsupported conclusion. AI can help generate practice and feedback, but it can also make weak reasoning look fluent.

This book offers a way to design and evaluate that practice. The aim is to help you decide where assistance is useful, where checks are needed and how the learner will demonstrate capability. More AI use is not by itself the measure of success.

You bring disciplinary and teaching judgement. A model may contribute alternative wording, questions or a simulated role. Neither its confidence nor the length of the resulting conversation establishes what a student knows.

1.2 A small example of the difference

In the fictional pilot, a junior adviser meets an employee whose manager has proposed a performance plan. A polished response announces the plan, invents an earlier warning and offers to approve support arrangements. It sounds empathetic, but conflicts with the supplied records and authority.

A shorter response checks the missing evidence, preserves the opportunity to respond and refers the decision appropriately. It leaves the matter unresolved. That is the stronger first response because its reasoning fits the facts.

The pilot pack provides both responses, flawed feedback, a correction and a changed case. You do not have to invent a scenario or trust a live model to work through the book.

1.3 The reader and the outcome

The main reader is an educator teaching business or a related professional discipline. You do not need to code or purchase an AI subscription. You do need enough disciplinary knowledge to recognise the task’s standards and arrange expert review where the material exceeds your competence.

Student-facing prompts and instructions are labelled as such. They are examples to adapt, not institution-wide policies or ready-made legal advice.

By the end, you should be able to:

Specify a bounded capability and the evidence needed to assess it.
Select permitted data and an approved or non-AI route.
Design practice and feedback with checkable inputs.
Inspect a response and justify a correction or warranted agreement.
Include an appropriate independent demonstration.
Evaluate access, workload and learning evidence before expanding a pilot.

A model’s success at producing a handout is not the same as your success at designing an assessment. The latter requires these decisions to fit together.

1.4 Conversation as a working method

The companion Conversation, Not Delegation develops the broader method of thinking with AI. PDP uses a short practical version: clarify the question, consider alternatives, check consequential claims and make a decision you can defend.

Its VET reminder is useful here: Verify the claim, Explain your reasoning and Test it against a changed condition. The classroom application adds a further question: which part of the student’s performance was assisted, and which part demonstrates independent capability?

A conversation can support that work. It can also become a long rehearsal of unsupported assertions. Do not reward the conversation itself without examining the decisions it contains.

1.5 Three design commitments

First, establish safety before use. The data gate covers permission, account features, handling and alternatives. Fictional material is the default for practice; a real placement case is not automatically suitable because names were removed.

Second, make feedback contestable. A model can misapply a rubric or invent a fact. Students need supplied criteria and examples that let them check its comments. They should be allowed to retain correct advice as well as reject errors.

Third, preserve the learner’s opportunity to act independently. The final changed case cannot be replaced by a reflection that merely says learning occurred. Choose a demonstration appropriate to the outcome and access needs.

1.6 The route through the book

Begin with Understanding AI and Ethics and Data Governance. Then try Getting Started, choose a prompt pattern and practise a checked handoff.

The next sequence develops critique, simulation, self-assessment and assessment design. Small content activities and the virtual company then lead into whole-unit design.

The closing chapters address adaptation, implementation and a pilot decision. The appendices supply the cases, rubric, stress test and workshop resources. These support a standalone route; the other books are optional extensions.

1.7 Activity: choose one capability

Allow eight minutes; no AI required. Pick a task you teach. Complete:

After this activity, the student can ___ using ___ evidence, under ___ conditions.

Avoid “understand AI better” or “produce a professional report” without saying what would demonstrate that outcome. In the pilot, the target is to plan and justify a fair first response using supplied records. It does not claim to certify a competent workplace investigator.

Keep your sentence. Each later chapter should help you make a concrete design decision about that capability.

# Introduction: A Defensible Classroom Pilot {#sec-introduction} > Start with a learning decision you can examine. ## The problem this book addresses A business educator may want students to practise difficult decisions more often than a timetable allows: asking a fair question, checking a recommendation, explaining an uncertainty or revising an unsupported conclusion. AI can help generate practice and feedback, but it can also make weak reasoning look fluent. This book offers a way to design and evaluate that practice. The aim is to help you decide where assistance is useful, where checks are needed and how the learner will demonstrate capability. More AI use is not by itself the measure of success. You bring disciplinary and teaching judgement. A model may contribute alternative wording, questions or a simulated role. Neither its confidence nor the length of the resulting conversation establishes what a student knows. ## A small example of the difference In the fictional pilot, a junior adviser meets an employee whose manager has proposed a performance plan. A polished response announces the plan, invents an earlier warning and offers to approve support arrangements. It sounds empathetic, but conflicts with the supplied records and authority. A shorter response checks the missing evidence, preserves the opportunity to respond and refers the decision appropriately. It leaves the matter unresolved. That is the stronger first response because its reasoning fits the facts. The [pilot pack](../appendices/pilot-pack.qmd) provides both responses, flawed feedback, a correction and a changed case. You do not have to invent a scenario or trust a live model to work through the book. ## The reader and the outcome The main reader is an educator teaching business or a related professional discipline. You do not need to code or purchase an AI subscription. You do need enough disciplinary knowledge to recognise the task's standards and arrange expert review where the material exceeds your competence. Student-facing prompts and instructions are labelled as such. They are examples to adapt, not institution-wide policies or ready-made legal advice. By the end, you should be able to: - Specify a bounded capability and the evidence needed to assess it. - Select permitted data and an approved or non-AI route. - Design practice and feedback with checkable inputs. - Inspect a response and justify a correction or warranted agreement. - Include an appropriate independent demonstration. - Evaluate access, workload and learning evidence before expanding a pilot. A model's success at producing a handout is not the same as your success at designing an assessment. The latter requires these decisions to fit together. ## Conversation as a working method The companion *Conversation, Not Delegation* develops the broader method of thinking with AI. PDP uses a short practical version: clarify the question, consider alternatives, check consequential claims and make a decision you can defend. Its VET reminder is useful here: **Verify** the claim, **Explain** your reasoning and **Test** it against a changed condition. The classroom application adds a further question: which part of the student's performance was assisted, and which part demonstrates independent capability? A conversation can support that work. It can also become a long rehearsal of unsupported assertions. Do not reward the conversation itself without examining the decisions it contains. ## Three design commitments First, establish safety before use. The data gate covers permission, account features, handling and alternatives. Fictional material is the default for practice; a real placement case is not automatically suitable because names were removed. Second, make feedback contestable. A model can misapply a rubric or invent a fact. Students need supplied criteria and examples that let them check its comments. They should be allowed to retain correct advice as well as reject errors. Third, preserve the learner's opportunity to act independently. The final changed case cannot be replaced by a reflection that merely says learning occurred. Choose a demonstration appropriate to the outcome and access needs. ## The route through the book Begin with [Understanding AI](understanding-ai.qmd) and [Ethics and Data Governance](ethics-integrity.qmd). Then try [Getting Started](getting-started.qmd), choose a [prompt pattern](seven-techniques.qmd) and practise a [checked handoff](managing-context.qmd). The next sequence develops [critique](critique-toolkit.qmd), [simulation](flight-simulator.qmd), [self-assessment](self-assessment.qmd) and [assessment design](assessment.qmd). Small content activities and the virtual company then lead into whole-unit design. The closing chapters address adaptation, implementation and a pilot decision. The appendices supply the cases, rubric, stress test and workshop resources. These support a standalone route; the other books are optional extensions. ## Activity: choose one capability **Allow eight minutes; no AI required.** Pick a task you teach. Complete: > After this activity, the student can ___ using ___ evidence, under ___ conditions. Avoid “understand AI better” or “produce a professional report” without saying what would demonstrate that outcome. In the pilot, the target is to plan and justify a fair first response using supplied records. It does not claim to certify a competent workplace investigator. Keep your sentence. Each later chapter should help you make a concrete design decision about that capability.