Guiding without giving the answer
How can you create an AI tutor that helps the learner progress without systematically solving the exercise for them?
1960 € per session, excl. VAT
In short
This 7-hour day, excluding breaks, enables you to build an AI tutoring prototype for a specific professional skill. You define the guidance stages, write the tutor’s instructions and test its responses. You leave with a usable scenario, a quality assessment grid and a supervised trial protocol.
Why this training?
An assistant that immediately provides the correct answer can conceal misunderstandings. Tutoring, by contrast, requires an initial diagnosis, targeted questions and a gradual withdrawal of support to make the reasoning observable. Fluent exchanges prove neither the accuracy of the explanations nor that learning has taken place. The training therefore includes checking references, analysing errors and an unassisted assessment, while keeping pedagogical responsibility with a human. Running case study: The fictitious organisation « Valdrise Archives » is preparing its new staff to classify files. Using a fictitious procedure, the tutor must ask learners to justify each choice, offer graduated hints and consult a designated adviser when a rule is missing. Transfer to the workplace: Trial the scenario for a comparable skill with a pedagogical adviser: observe requests for help, check the answers, then compare success with tutoring against performance on a new independent exercise before extending its use.
Objectives of this training
- Defining a skill and the evidence of mastery
- Building a progression of hints tailored to common errors
- Writing the tutor’s instructions and defining its limits
- Testing dialogues with several learner profiles
- Measuring pedagogical quality and preparing a supervised trial
Who is it for and what are the prerequisites?
Roles and tasks concerned
- Instructional designers and digital learning pathway designers
- Vocational trainers and tutoring programme managers
Prior skills
- Be able to structure a request to generative AI and verify its response.
- Have a thorough command of a professional skill being taught and its main learning difficulties.
What you will work on
Defining the scope of learning
Objective: Define what the staff member must be able to accomplish independently. Observable skill, starting level, scope of documentation and limits of tutoring.
Mapping the difficulties
Objective: Link errors to pedagogical interventions. Confusion between categories, a forgotten rule, insufficient justification; diagnosis without lasting labels.
Organising support
Objective: Avoid providing the correct answer prematurely. Open question, rule reminder, targeted hint, analogous example and withdrawal of support.
Assembling the prototype
Objective: Translate the scenario into testable instructions. Role, reference procedure, limits, feedback format and escalation to a human.
Putting the dialogues to the test
Objective: Identify undesirable pedagogical behaviours. Request for a direct answer, persistent error, ambiguity in the documentation and attempted circumvention.
Correcting and retesting
Objective: Improve guidance based on evidence. Assessment grid for accuracy, gradual progression, relevance of feedback and adherence to limits.
Preparing the trial
Objective: Distinguish successful assistance from genuine skill acquisition. Unassisted assessment, human observation, minimisation of recorded interaction data and stopping conditions.
What you take away
- Pedagogical specification: skill, errors and expected evidence
- Tutoring scenario and version-controlled instructions
- Test set with a completed quality assessment grid
- Protocol for trials and unassisted assessment
Method and assessment
- Short instructional inputs followed by immediate practice using the running case study.
- Individual work followed by work in pairs, with peer review of the outputs.
- Guided group discussion: what works, what needs correcting and what remains a human decision.
- Preparation: Prepare a fictitious or anonymised procedure, together with an exercise and its validated model answer. Check access to a tool approved by the organisation and the rules applicable to dialogue records.
Observable criteria: Each stage asks the learner to take action or provide a justification. ; Hints are graduated before any explicit correction is provided. ; Explanations are consistent with the procedure; gaps in it trigger referral to a human. ; Tests document deviations, corrections and the results of further trials. ; A new task without AI checks skill transfer..
Frequently asked questions
Are programming skills required?
No, the prototype relies on instructions and dialogues; no application integration is required.
Can real exchanges be used?
Only after anonymisation and authorisation; the exercises can remain entirely fictitious.
Does the tutor replace the trainer?
No. The designated lead validates the content, addresses ambiguities and assesses the knowledge and skills acquired.
Is the setup ready for a full rollout?
No. The deliverable prepares the way for a limited trial; its wider rollout depends on internal testing and validation.
Last updated: