Move AI work from the manager's screen into the team's practice

You showed the team how you work with AI: you shared your screen, took a real work task and got a result. Your employees were impressed and everyone agreed to give it a try. A week later, documents are prepared the old way, data is collected by hand, and people come back to you for help with AI.

For this to become team practice, hand over not the tool but a way of doing a specific piece of work — and test it by putting the employee at the keyboard.

You yourself, meanwhile, are already at home in Codex, Claude or another AI agent: you work through big tasks with it, assemble documents, come up with automations, and things that used to be put off for weeks move forward in an evening. An understandable frustration sets in: I showed them, it works for me.

But it is worth looking at what exactly you managed to hand over. Your employees saw what you can do. Did they get everything they need to repeat your work on their own?

More is working on your screen than it seems

When you solve a familiar task with AI, you may simply not notice part of the work. You know which document to open and which one has long been out of date. You understand what the client meant, even though they phrased it badly in their email. You can see where the model suggested suitable wording and where it confidently wrote something the company cannot deliver.

As the conversation goes on, you refine the request, add missing information and discard what is unnecessary. Sometimes one short remark is enough, because you already picture the result you need.

An employee watching from the side sees a simple sequence: the manager wrote a request, AI answered, the document is ready. But behind it are your experience, your access to the materials and dozens of small decisions. If these stay in your head, a single ready-made AI request will not be enough for a colleague. Together with the tool, you need to hand over a way of doing the specific work: where to get the input data, which decisions they can make themselves and how to tell that the result is good.

“Use AI” is too big a task

A typical situation: the manager has learned to prepare a draft sales proposal with AI. They take the client’s request, add information about the company, attach a suitable earlier document — and after a few refinements get a basis that only needs checking. After the demonstration, the sales team is asked to do the same.

The first salesperson attaches an old proposal with outdated terms. The second pastes in only the client’s email and gets a text that is too general. The third sees a promise in the answer that was not in the source materials and does not know whether it can stay. The fourth decides it is quicker to use the familiar template: at least with that it is clear what to check.

It is easy to conclude that the team is resisting change. But first check whether the employees had the same conditions. Did they know where the current template was? Did they have an example of a good result? Did they understand which information must not be made up and who confirms disputed terms?

The instruction “use AI” leaves too many questions open: a person has to learn the tool, invent a sequence of steps and take responsibility for an unfamiliar result, all at once.

For a first attempt, narrow the task: prepare a draft of one section of a proposal from specified materials, check it against clear criteria and pass it on for approval.

Hand over one workflow the employee needs

A manager may have a whole personal lab: several agents, projects, instructions, connected services and their own habits. All of this grew around their tasks and interests. The employee has tasks of their own, and one clear way to handle an incoming request may be more useful to them than a tour of every capability of your system.

Start with work that a particular person does regularly and understands well: then they will judge the AI’s result on its merits, notice a mistake and explain where the new way helps and where it adds hassle. For the proposal example, agree on the following:

• when to use the workflow: when preparing the first draft of a proposal for a new request;
• which materials to use: the client’s request, the current description of services and the approved template;
• what to give AI: building the structure and the draft, and listing the missing information separately;
• what the employee checks: fit with the request, facts, scope of work, no unconfirmed promises;
• when to ask for a decision: if the price, deadline or non-standard terms are unclear;
• where to save the result: in the project folder, with a link from the task;
• who accepts the work: the person responsible for the proposal, before it is sent to the client.

This is a working agreement for one workflow — test it on the next task and adjust it where it is awkward. Along with it, give an example of an accepted result and explain why it works. “Make a good proposal” hardly helps anyone check their work, whereas a specific document with comments shows what you value and which mistakes you consider significant.

The most useful demonstration is when the employee is at the keyboard

Once you have shown your way of working, swap roles. Let the employee take the next real task and go all the way through it alone: find the materials, write the request, assess the answer, make edits and save the result where someone else can pick it up. Your job is to observe and notice where you have to step in.

It is very tempting to prompt them: “open the other file”, “ask it differently”, “drop this paragraph”. But each such intervention shows what is missing from the way of working you handed over:

• the employee cannot find the source document — sort out access and where the materials are kept;
• they don’t understand what counts as a good answer — they need an example and an explanation of the criteria;
• they are afraid to correct the model — go through one unsuccessful attempt together and show how to continue the conversation.

Sometimes it turns out that the workflow simply does not fit their work well: the draft comes out quickly, but then a lot of time goes into transferring it into the mandatory template. That is a reason to improve the workflow, not to persuade the person to try harder. Every time they get stuck without your hint marks a specific point that needs to be explained, simplified or changed. After an attempt like this, the instructions become much more useful: they contain answers to the questions that actually came up.

Give the new way a place in the ordinary working day

Even a successful joint attempt does not mean the employee will use this way of working tomorrow. The working day is full of urgent tasks, familiar templates and established routines. If the new workflow remains something you “can try in your spare time”, it is easy never to get round to it.

Agree on which upcoming tasks it will be used for and when you will look at the results together. For example, the salesperson prepares the next few proposals the new way and notes where they needed help or had to fall back on the old routine. Set aside working time for this trial: if a person has to handle the same workload as before and learn a new process on the side, you have in effect given them an extra duty whose benefit has not yet been proven.

Also name the person who will maintain the workflow: update the template, collect questions and deal with recurring difficulties. In a small team this may be the manager, but the role must be explicit. Otherwise every change will again end with the message “I got something weird, can you take a look?”.

Check whether finishing the work has become easier

The question “so, are you using it?” says little about the result. An employee may open AI regularly and spend more time on the task than before. A good-looking draft may need lengthy checking, and a quickly prepared document may come back for rework several times. Assess the whole process — from preparing the materials to an accepted result:

• how much time goes into preparation and checking;
• which mistakes and redos keep recurring;
• where the manager’s help is needed;
• whether the employee can repeat the process on the next similar task.

To begin with, short notes on a few attempts are enough — what matters is seeing where the benefit appears and at what cost. Perhaps AI is good at helping to break down a client’s request, but it is too early to hand it the whole proposal. Then keep the part that works and refine the rest later. That is also a result: you have found work the team can do better and understood the limits of the approach.

The manager will have to change their role too

When you enjoy working with AI yourself, it is easy to become, without noticing, the person through whom all complex requests pass. An employee brings a task, you quickly solve it in your configured environment, show the result and go back to your own work. Today that may genuinely be faster than explaining and waiting. But if you always do it this way, the team will learn to bring you tasks for AI, and you will end up with yet another queue of requests.

Handing over a practice takes time: watching how another person works, allowing them to go more slowly, going through mistakes and adjusting the instructions.

An employee may choose a slightly different sequence of steps that suits them. Keep the source materials, the requirements for the result and the mandatory checks common to everyone, and within those boundaries give them room for their own way of working. You will recognise the goal in a simple moment: the next task is finished, the result is accepted, and you did not have to explain the whole process again.

From personal practice to the way the company works

Different speeds of adopting AI are understandable: employees have different tasks, experience and time for experiments. One impressive demonstration will not close this gap.

To start: choose one task that you already know how to solve with AI and that someone on the team does regularly. Prepare the source materials, an example of the result and a short sequence of steps for that person, and then suggest they work through the next real task on their own. Write down every point where you had to step in: this list will show what to improve before the next attempt. Once the approach starts to repeat without constant prompting, you can test it with more employees and tasks.

Read next: if tasks keep coming back to you because agreements are missing — how to hand over a task so that it doesn’t all stay in the manager’s head; once the workflow has taken root and you want to hand the process to an agent — how to prepare a work process for an AI agent: ten checks.

If you already do a lot with AI but the team is still working the old way, we will look at one of your workflows at a free 30-minute meeting with Simplit: what it will take for an employee to use it — from the source materials to checking the result. Discuss your task