How can managers use AI in their own work?
Managers can use AI for the preparation and writing that surrounds people work, as long as the decisions stay with them. The useful tasks have something in common: the manager already knows the facts and wants help organising or wording them.
Keep personal information to the minimum the task needs, and use only the tools your organisation has approved. Notes about an individual's health, family circumstances or a grievance do not belong in a general assistant.
Prepare a one-to-one
Turn your notes from the last meeting and the person's current priorities into an agenda with two or three open questions.
Structure feedback
Organise specific observations into a situation, behaviour and impact format before a conversation. The observations must be yours.
Summarise team updates
Combine several written updates into one view for your own manager, keeping each person's points attributed.
Explain a change
Draft a team message about a new process, then ask the assistant which questions the team is likely to raise.
Think through a decision
Lay out options and consequences for a resourcing or priority decision, then make the decision yourself.
Plan a meeting
Turn a vague purpose into an agenda with a decision to reach and a time for each item.
What does AI-assisted feedback look like when it is done well?
Done well, the manager supplies specific observations and the AI only helps with structure and wording. Done badly, the AI turns thin notes into confident judgements about someone's character.
A fictional example. Dana manages a small client services team at Copperleaf Bookkeeping. Her notes on one team member, Sam: 'Two client deadlines missed in September (Acme payroll file, Riverside VAT pack). Both clients emailed me. Sam's client calls are excellent, clients ask for him by name. Seems stretched since taking on the Hollis account.'
A loose prompt, 'Write feedback for Sam based on these notes', produced lines such as 'Sam lacks organisational skills and needs to show more commitment' and 'This has damaged our reputation with key clients.' Neither is in the notes. 'Lacks commitment' is a judgement about character the evidence does not support, and the reputational claim is invented.
A better prompt: 'Help me structure feedback for a one-to-one. Use only the observations below. Describe behaviour and its effect, not personality. Include the positive observation. End with one open question about workload. Do not draw conclusions about the person's attitude.' The checked version opened with the two missed deadlines and the client emails, credited the client calls specifically, and asked: 'Since you took on Hollis, how manageable has your workload been, and what would help?' Dana then edited the wording into her own voice before the meeting.
Where should managers keep AI out of decisions?
Keep AI out of decisions about people: performance ratings, pay, promotion, discipline, dismissal and who gets which work based on their behaviour or traits. Use it to prepare and word, not to judge.
The EU AI Act lists, among high-risk uses in Annex III, AI systems intended to make decisions affecting terms of work-related relationships, promotion or termination, to allocate tasks based on individual behaviour or personal traits, or to monitor and evaluate the performance and behaviour of workers. A manager asking an assistant to tidy their own notes is not deploying such a system, but asking a tool to score or rank team members moves towards it.
Monitoring is the related trap. The UK ICO's guidance on monitoring workers says employers must inform workers about monitoring, be clear about the purpose and choose the least intrusive means to achieve it, and notes that the availability of a form of monitoring does not make it the best way to achieve your aims. Reading staff members' AI conversations to see how they use the tool is monitoring. This is general information, not legal advice.
Want to see how this connects to a role?
Explore the planned team trial ↗How do you lead a team that uses AI?
Give the team clear rules, time to practise and examples of good work, then review the work itself. Most adoption problems are not about the tool: people are unsure what is allowed, have no time to learn, or have never seen what good use looks like in their own job.
Say what is allowed
Name the approved tools, what information may go into them and what must always be checked by a person. Put it in writing in plain language.
Start with one task each
Ask each person to choose one repeatable, low-risk task from their own role, such as a first draft or a summary.
Protect a little time
A short, regular slot for practice works better than a single long workshop that people forget.
Share before and after
In a team meeting, have someone show the prompt, the first output and what they changed. The changes teach more than the prompt.
Review outputs, not usage
Check a sample of AI-assisted work the way you would check any work. Ask what was verified and how.
Model it yourself
Tell the team when you used AI for something and what you had to correct. It makes checking normal rather than embarrassing.
How do you know whether your team is using AI well?
Look at the quality of the work and the reasoning behind it, not at how often a tool is opened. A count of prompts tells you a tool is in use. It does not tell you whether a reply was accurate, a summary was faithful or a figure was checked.
Useful signals are concrete: can a team member explain which claim in a draft they removed and why? Do AI-assisted documents still need heavy correction from you? Are people using AI for the tasks you agreed, inside the approved tools?
This is the design principle behind the planned Tosheo.ai manager dashboard: managers see skill progress, completion and a weekly digest, with a reminder option for inactive learners, but not raw learner chat logs. It is visibility into learning, not surveillance of conversations. The product is pre-build, and the final fields will be documented before launch.
What should AI training for managers cover?
Training for managers should cover their own use and their leadership of others, and the second part is usually missing. A useful programme includes:
Their own practical use
Drafting, summarising and preparing, with the same checking habits expected of the team.
Reviewing AI-assisted work
How to question a draft, spot invented detail and ask a team member what they verified.
Policy and data boundaries
What the organisation allows, why, and how to answer the team's questions about grey areas.
People decisions
Where AI must stay out, and why fairness and data protection matter more in people management than elsewhere.
Coaching
How to help a reluctant or over-confident team member find a sensible starting task.
When is this guide, or Tosheo.ai, not the right answer?
If you are a senior leader setting AI strategy for an organisation, deciding investment, governance structures and risk appetite, an executive or business school programme is a better fit than a practical guide. Our AI governance and policy guides are a starting point for the smaller-company version of that work.
Tosheo.ai is being built for hands-on practice in everyday work. Its planned preset roles are functional, such as Marketing, Sales, Operations, Support, HR and Recruiting, Finance, Product and General; there is no dedicated people-manager path in the planned launch list. A manager would take the path for their function or the General path, and the planned Growth plan adds custom roles. Managers also get the planned dashboard for their team.
Sources and scope
UK Information Commissioner's Office, Employment practices and data protection: monitoring workers ↗