How to Run Your Team's First AI Training Session on Real Work Tasks

A first AI session is more useful when it starts with tasks the participants already do. The aim is for each person to leave with a reviewed template for their own process and a clearer sense of where the tool helps or falls short. The sequence below has the team document its current method, try AI, then check the output against the source.

What to Prepare Before the Session

The organisational work done beforehand largely determines the quality of what comes out.

Tasks. Each participant picks one recurring work task in advance - preparing a summary, responding to a standard request, drafting an initial version of a document. The task should be real but safe to use in the room: customer personal data, commercial terms, and internal financial figures need to be removed or replaced with fictional stand-ins. This is called anonymisation, and it is not a formality. Data accidentally sent to a cloud service cannot be retrieved.

Documenting the current approach. Before the session, each person writes down two things: how they currently carry out the task, and what tells them the result is good. These answers become the baseline. Without them, there is no way to know whether AI actually helped or just created an illusion of speed.

Shared minimum context. Before the practical work begins, set aside time for a short briefing: what AI tools do well (generating text following a sample, structuring, paraphrasing, extracting key points), where they go wrong (precise figures, references, specialist terminology without context), and why it matters to check what goes into a prompt before sending it. This is not a lecture on how neural networks work - it is a working briefing.

Session Structure: Five Blocks

Block 1. Unpacking the Task Before Touching AI

The participant explains their task out loud: what goes in, what should come out, and what success criterion they wrote down beforehand. The group asks clarifying questions. The point of this block has nothing to do with AI: the person formulates the task more precisely than usual because they have to explain it to someone else.

Format: 5-7 minutes per participant in a small group (3-4 people). Materials: the pre-filled task description sheet with criteria. Success criterion: the task is described clearly enough that a colleague could carry it out independently.

Block 2. First Attempt with AI

Each person writes their own prompt and gets a response. No hints from the facilitator at this stage - the goal is to record what worked first time and what did not.

Format: individual work, 10-15 minutes. Materials: the anonymised version of the work task, access to an AI tool. Success criterion: there is a first AI response and the participant's initial assessment of whether it meets the criteria written before the session.

Block 3. Cross-Check

The participant passes their result to a colleague to check the structure and any obvious discrepancies with the source material. Facts and specialist criteria are then confirmed by the process owner or the relevant subject-matter person. This is where you see how AI can be convincingly wrong - the wording sounds professional but does not match the source.

Format: pair work, 10 minutes. Materials: the original document or data against which the response can be checked. Success criterion: the colleague records visible discrepancies and open questions; the process owner checks specialist facts separately.

Block 4. Iteration and Template

After the cross-check, the participant refines the prompt, removes the typical errors, and writes up the final version with notes: what needs to change for a specific case, and what stays fixed. This is the template - not a universal one, but one tied to a specific person's specific process.

Format: individual work with discussion, 15 minutes. Materials: a prompt template form with fields for "fixed part", "variable part", and "what to check in the result". Success criterion: the template is clear to a colleague without any further explanation.

Block 5. Shared Observations List

At the end of the session the group compiles three lists: recurring AI errors on team tasks, things AI did consistently well, and limitations that require a different solution. Not every task is solved with a prompt - sometimes you need a data integration or a separate process. These lists form the basis for the next step.

Format: group discussion, 15-20 minutes. Materials: a shared board or document accessible to all participants. Success criterion: each item on the list is tied to a specific task and to a specific participant as the process owner.

Teaching Example: an Anonymised Work Report

The following is a fictional example for illustration purposes only, not a real case.

A session participant prepares a weekly project-status summary for their manager. Current approach: manually pulling data from three spreadsheets, writing the text, formatting it. Success criterion: all projects are mentioned, each project's status is unambiguous, the total length does not exceed one page.

On the first attempt, AI generates a coherent text from the pasted data but adds evaluative phrases - for example, "the project is progressing well" - that are not in the source data. A colleague spots several of these during the cross-check. After iteration, the participant adds an explicit constraint to the prompt: "use only facts from the table, do not add evaluations". The final template records this instruction as a fixed part.

What this example shows is not that AI succeeded or failed - it is that the team now knows a specific risk point in a specific process and knows how to control it.

What the Team Takes Away

Three tangible outcomes should exist after the session.

The first is a set of tested templates tied to real tasks, each with a result-checking instruction. The second is a list of potential use cases with a named owner for each process - the person responsible for the quality of the result when AI is involved. The third is clarity on which tasks need more than a template: a data connection, process configuration, or formal acceptance criteria.

Attending the session does not confirm that the skill has been acquired. Verification works differently: after some time, the participant independently completes a new task of the same type and presents the result together with a completed checking instruction. If the result meets the criteria, the skill can be considered embedded.

Training and building a solution are different things. Some tasks will be ready for independent use with a template after the session. Others will require a full project: with data, a named owner, and formal sign-off. Conflating the two is costly - in time and in team expectations. For more on how to structure the progression from early experiments to a working solution, see AI team training and team AI pilot project.

Frequently Asked Questions

Does the session need a specialist facilitator, or can someone internal run it?

An internal person can run it, provided they have already used AI on real work tasks themselves and understand the difference between a convincing answer and an accurate one. The key requirement for the facilitator: do not offer ready solutions during the first-attempt block, and do not skip the cross-check block.

What if participants' tasks are too different for the group to work together effectively?

Different tasks can be discussed in the same group as long as the inputs and criteria are written down for each example. A colleague from a different process can spot an unclear instruction or a gap - they just cannot be expected to verify specialist facts. Those are confirmed by the process owner or the relevant subject-matter person.

How do you know the session went well on the day itself?

Two indicators: every participant leaves with a written template they are ready to use in their next piece of work, and every participant can name at least one risk point in their process. If neither is true, the session is worth repeating with different tasks.

Is it worth running this session if the team is already using AI?

Yes - especially then. Informal use accumulates habits that may be inefficient or unsafe. A structured session lets you examine what the team is actually doing, document the good practices, and correct the problematic ones.

If you would like to discuss how to adapt this format for your team's tasks, get in touch on Telegram @shimaoz or by email at hello@majento.ai.

Source

Majento consultations and training courses.