Why start with familiar work?
Start with familiar work because the reviewer already knows what good looks like. If a manager can judge a client email or a handoff note without AI, they can explain the review criteria before anyone opens a tool, and the conversation afterwards is about decisions rather than about whether the tool is impressive.
A task that depends on specialist knowledge nobody in the room has is much harder to assess responsibly. If the reviewer cannot tell a correct answer from a confident one, the exercise teaches people to trust fluency, which is the opposite of the habit you want.
What makes a good first practice task?
A good first task passes six checks. If it fails one, choose another task or change it until it passes.
Familiar
It is a deliverable the team produces already, such as an email, a summary or a short brief.
Low risk
Nothing bad happens if the first attempt is wrong, because it is not going to a customer or regulator.
Bounded source
The allowed facts fit on one page, so an invented claim is easy to spot.
Clear reader
Someone specific will read it, which gives the learner something to aim at.
Named reviewer
A person who knows the work agrees to review it within a day or two.
No confidential data
It uses fictional or approved material, following your company's AI policy.
Which first tasks work well for different roles?
The table gives one starter task per role. Each has a source the reviewer can check quickly and an obvious place where an AI draft is likely to overreach.
| Role | First task | What the reviewer checks |
|---|---|---|
| Marketing | An event invitation from a one-page brief | No invented attendance, results or scarcity |
| Sales | A follow-up email from three meeting notes | Every commitment appears in the notes |
| Support | A reply to a complaint using a pasted policy | No promise the policy does not allow |
| Operations | A new-starter summary of a short SOP | No safety or logging step dropped |
| Finance | A two-line explanation of a change in a small table | The figures match and the cause is not invented |
Want to see how this connects to a role?
Explore the planned team trial ↗What does a worked first task look like?
A fictional consultancy is running a 60-minute online workshop on clearer project briefs, with a worksheet and a live question period. Those three facts are the whole source. The learner's prompt: "Write a short invitation email to operations leads for this workshop. Use only the facts below. Keep it under 120 words with one clear call to action."
A typical first draft is fluent and overreaches. It says the workshop is "trusted by hundreds of teams", that "places are filling fast" and that attendees will "cut project delays in half". None of that is in the source. A reviewed version keeps the three facts, explains why an operations lead might care about clearer briefs and ends with a single sign-up line.
The learner's review note is the useful part: "Removed the attendance claim, the scarcity line and the result promise because the brief does not support them." That note gives the manager something to discuss. The same pattern of a bounded source and an obvious trap works in any role.
Which first tasks should you avoid?
Avoid anything that uses customer personal data, anything with legal, financial or safety consequences and anything without a named reviewer. Those may be good later tasks, but they make a poor first one because a mistake is costly and the review becomes about risk rather than learning.
Also avoid open-ended tasks such as "use AI to write our strategy". There is no bounded source, no clear reader and no way to tell a good answer from a long one. A first task should be finished, reviewed and discussed within a week.
What should the first discussion cover?
Ask three questions: what context the learner supplied, which claim they checked and what they changed before sharing. Keep the source, the first draft and the final edit side by side so another person can follow the reasoning.
A first task should open a conversation, not label someone an expert or a beginner. Use a second, different example before drawing any broader conclusion about a person's skill.
How does this connect to Tosheo.ai?
The planned Tosheo.ai diagnostic and lessons follow the same principle: short practical tasks with a supplied source, a visible rubric and a note on what changed. You do not need a product to run a good first task, though. A brief, a reviewer and a fifteen-minute conversation are enough to start, and Tosheo.ai is pre-build in any case.