Design role-specific AI learning paths around the work people must perform and the decisions they own. Give everyone a common foundation, then choose a recurring task for each responsibility, show what good work looks like, and let the learner practise with feedback. Before they work independently, ask them to handle a different example and explain their checks.
For an L&D or adoption lead, the useful output is a path you can run with a team: an approved task, a realistic practice case, a named coach and evidence of what the person can now do. A course catalogue can supply material, but you still need to connect it to the job. The wider people and change guide covers how learning fits into adoption.
Keep a shared foundation and vary the practice
Start with the employee AI literacy baseline: when to use AI, which information is permitted, how to check a result and when to involve someone else. Those expectations should remain recognizable across teams.
Skills England's employer guide to AI upskilling recommends practical learning rooted in workplace tasks, with accessible provision and room to adapt a consistent core. It is guidance from a UK research program, rather than proof that one curriculum works everywhere.
Its Roche case study describes a shared AI foundation for technical and nontechnical employees, alongside function-level champions and sessions using colleagues' examples. The case is an organization-validated interview account, with self-reported outcomes. It illustrates how common expectations and local practice can coexist.
A role in a learning path is a responsibility. One person may use an assistant, manage colleagues who use it, configure a workflow and review its output. Assign the relevant practice to that person, even if their job title sits in only one department.
Give each responsibility a different piece of work
Suppose a facilities service desk receives maintenance emails for an office building. Jamie Lee, a coordinator, needs a concise ticket summary. Claire Bennett, the manager, decides who should handle the request. A builder configures an approved assistant to suggest ticket fields. A reviewer checks whether the suggestion accurately preserves the request and follows the desk's rules.
Use the same sample email across these paths so people can see how their work connects. Change the output and the judgment expected of each learner.
| Responsibility | Practice output | What the coach checks |
|---|---|---|
Employee using AI | A draft ticket summary that preserves the location, reported problem and unanswered questions | The summary matches the email and adds no invented diagnosis, urgency or repair promise |
Manager supporting use | A decision about the handoff, review and capacity needed for this request | The manager assigns accountability and follows the existing escalation process |
Builder configuring the workflow | A proposed mapping from email details to ticket fields, tested with incomplete and conflicting requests | Missing information stays missing; the configuration does not silently send or approve a ticket |
Reviewer accepting the result | An accept, revise or escalate decision with the reason recorded | The reviewer checks the source and desk rules, rather than judging only how polished the summary sounds |
Design the practice and its permission boundary together. Someone who can review a ticket summary should not automatically be allowed to alter the assistant that produced it.
Supply the actual local rules with the exercise. You are teaching people to use their organization's process, including when to stop using the assistant. An AI-written summary should not become the authority for handling a hazard.
Let people see judgment before asking them to repeat it
Start with a coach doing the task aloud. Show the source material, the request made to the assistant, a disappointing response and the correction. The learner needs to hear why a detail matters. A flawless demonstration hides much of that judgment.
In his account of learning at Shopify, Tobi Lütke describes a support engineer watching a backend colleague use the company's River agent to find a log query, then trying the method the next day. He argues that visible work can teach without a formal curriculum. This is his account of one company's practice, rather than an independent assessment of learning.
That is a useful challenge to a long course-first path. You can include observation and apprenticeship in a short path. Use an approved example and a suitable group, then ask each learner to attempt the work. Company-wide visibility is not necessary, and sensitive requests need their normal access boundaries.
Move from demonstration to a new task
Watch the judgment
The coach compares an AI ticket summary with the original email and explains a correction.
Practise together
The learner handles another request, with feedback on the decisions they own.
Try a different case
The learner works without prompting from the coach, including a missing detail or exception.
Agree the next use
Record the permitted task, required review and remaining learning need.
The independent attempt should test the same responsibility without repeating the demonstrated answer. For Jamie, change the problem and location details. For the builder, add conflicting descriptions. For Claire, change the staffing constraint. Keep the exercise within the approved scope; passing it does not grant broader permissions.
Make supporting the team a manager's learning task
Managers need practice in making learning possible and responding to problems. Include a conversation about a rejected output, a decision about review workload and a plan for protected practice time. These tasks deserve attention even when the manager rarely writes a prompt.
In a Reddit discussion, a workshop designer described running critical-thinking practice alongside tool webinars and local policy training. They wanted greater emphasis on:
developing and encouraging curiosity and experimentation, and coaching managers to support their teams.
The author also described fear and concerns about AI's effects. This is one person's account of their rollout, not a measured comparison of training methods.
For the facilities manager's path, ask Claire to respond when Jamie says, “The summary looked good, but it changed the reported hazard.” She should help examine the attempt, decide the safe next step and make room for another practice case. A response that blames Jamie for slowing the queue teaches a different lesson.
Agree how to protect learning time alongside delivery targets before inviting people. Provide the approved tool and exercise materials, check accessibility needs, and tell learners who can help when they get stuck.
Keep a small record of what the path teaches
For each path, record these details where the coach and learner can find them:
- Task and boundary. Name the recurring work, approved inputs and actions that still require a person.
- Practice material. Keep a typical case, an exception and the local rules used to judge them.
- Evidence of learning. Retain the learner's output, corrections and explanation of an independent attempt.
- Next use and support. Agree where they may apply the skill, who reviews it and what still needs practice.
- Owner and review trigger. Name who updates the path when the tool, policy or workflow changes.
Avoid treating attendance or enthusiasm as proof of transfer to work. In the same Reddit thread, another workshop organizer clarified that their pilot was only about a month old and follow-up surveys were still ahead. Interest in the workshop was observable; sustained adoption remained an open question.
The content also needs to move with the job. In an account on LinkedIn, author and community founder Jacob Morgan reports a financial-services learning leader's shift from prompting toward context, evaluation and workflows. It is a secondhand account with an unnamed employer, but it identifies a useful review question: does the path still teach the decisions people now make?
Start with one repeated task and the people responsible for it. Review their attempts, repair the exercise where it was unclear, and decide what to add next. You can build a usable learning path before building a large curriculum.
Questions and answers
Should AI learning paths follow departments or job titles?
Use departments to find realistic examples, then assign practice by responsibility. A facilities coordinator and manager may work with the same maintenance email while owning different decisions. Someone who also configures the assistant needs builder practice too. Start by naming who uses, manages, builds and reviews the workflow, then use the responsibility table to choose their outputs and checks.
Does every employee need to learn coding?
No. An employee using an approved assistant needs to understand its permitted use, inspect the result and recognize when to ask for help. Coding belongs in a path when the person's work requires it, such as building or maintaining an integration. Begin with the common AI literacy baseline, then add the technical skills required by the actual responsibility.
How long should a role-specific AI learning path take?
Set the duration from the task, the learner's starting point and the practice needed to handle exceptions. There is no single training-hour target in this guide. Schedule a demonstration, supported practice and a different independent attempt, then review what remains difficult. Give employees protected time and adjust the plan from their work, rather than assuming a certificate or a short workshop establishes readiness.
How do you know someone has completed the path?
Ask the learner to perform a new example of the approved task, explain their checks and handle an exception. The coach should review the output and reasoning against the local rules, then record the permitted next use and any continuing review requirement. Completion applies to that task and responsibility. A successful ticket-summary exercise does not establish competence to configure an automated ticket workflow or make its approval decisions.
