How Managers Can Use AI in Daily Work: A Realistic Weekly Routine

How Managers Can Use AI in Daily Work: A Realistic Weekly Routine

If you are wondering how managers can use AI in daily work, start small: pick a few recurring writing, summarizing and preparation tasks, give each one a fixed day of the week, and decide in advance who reviews the output. The AI produces the draft. The decision, the tone with your people and the accountability stay with you. A simple routine, repeated, delivers more than trying several tools without criteria.

Key takeaways

• Start with tasks you already do every week, such as agendas, meeting minutes, reports and announcements, not with new tools.
• Give each day of the week a function, so that use becomes a habit and not improvisation.
• Define what never goes into a conversation with AI: sensitive personal data, trade secrets and evaluations of people.
• Every output has a human review point, with a simple criterion to approve or redo it.
• At the end of each week, record what worked and save the best prompts for reuse.

Why start with a routine and not a tool?

Most managers who try AI start from the wrong end: they open a tool, ask a random question, like or dislike the result and go back to working as before. The use does not last because it is not tied to any moment of the actual work. When AI has a scheduled slot in the week, it competes with real tasks and saves real effort.

Another reason is quality. A request repeated every week improves with each round, because you learn what context to provide, what format to require and where the result tends to fail. One-off requests never mature.

Finally, a routine makes clear where AI does not belong. That matters as much as the use itself: a manager answers for decisions, and decisions are not outsourced to a language model.

What to keep out before you start

Before the first Monday, set the limits. There are only a few, and they apply to you and to the people who work with you.

• Personal data of customers, candidates and employees does not go into tools the company has not approved, and sensitive data only goes in with a legal basis and clear controls.
• Contracts under negotiation, margin spreadsheets and strategic information stay out of personal accounts.
• Performance reviews, promotions, dismissals and formal warnings are neither written nor decided by AI.
• Nothing generated by AI goes out under your name without a full read.

If your company does not have a written rule yet, read the article on a company policy for employee ChatGPT use first. It helps separate what to allow from what to ban. For the decisions that should stay human, the article on what not to delegate to AI in business complements this list.

In Brazil, the LGPD (the Brazilian data protection law) applies to any personal data you paste into a tool, even a simple name in meeting minutes. If you have doubts about official guidance, the website of the national data protection authority (ANPD) is the place to start. In practice, the safest habit is to anonymize: replace names with roles (for example, agent A, customer B) before pasting the text.

What does the weekly routine look like, day by day?

The routine below is a starting point. Adjust the days to your calendar, but keep the idea: one function per day, one review per output.

• Monday, planning: paste your list of open items and commitments (without sensitive data) and ask for a proposed order of priorities, along with questions about what might be missing. Review: you check that the suggested order respects commitments only you know about.
• Tuesday, meetings and follow-ups: turn your notes into minutes, with decisions, owners and deadlines. Review: each owner and each deadline is checked against what was actually agreed.
• Wednesday, numbers and reports: ask the AI to explain variations, point out questions the report does not answer and organize the presentation text. The calculation itself stays in the spreadsheet or the system. Review: every number cited is traced back to its source.
• Thursday, people and communication: use the AI to draft announcements, prepare the structure of a difficult conversation or check the clarity of a message. Review: tone, facts and fit with your team's culture, read aloud.
• Friday, wrap-up and learning: ask for a summary of the week from your notes and a list of what remains open. Update your prompt library with what worked. Review: you decide what becomes the priority for next Monday.

Notice that no step asks the AI to decide on its own. It organizes, summarizes, questions and drafts. You choose.

How do you write good prompts without becoming a prompt expert?

A good manager's prompt has four parts, and all of them fit in a few lines.

• Role and goal: who the text is for and what it needs to achieve (inform, align, ask for a decision).
• Minimum context: what the AI needs to know so it does not invent, pasted by you and already stripped of sensitive data.
• Expected format: bullet points, a short email, a table of owners and deadlines, whatever is useful to you.
• Honesty rule: ask it to mark what is an assumption and to list what information was missing.

That last part reduces one of the biggest risks of daily use, which is the convincing but wrong answer. When the AI is told to separate fact from assumption, review gets faster, because you know where to look first.

If you want ready-made templates for each department, the article on business prompts by department shows how to adapt them without starting from scratch.

How do you review what the AI delivers?

Review is not skimming. It is a short check, always with the same questions. A checklist that works for most management tasks:

• Do the facts match the source? Were dates, names, owners and numbers verified?
• Is anything stated with certainty that is actually an assumption?
• Does the text sound like you or like a generic announcement?
• Did any confidential or personal information slip into the text?
• If the people affected read this message, is there anything unfair, ambiguous or embarrassing?
• Does the person who will act on this know exactly what to do and by when?

If the answer to any item is no, redo it or fix it by hand. Over time, you learn where the tool most often fails in your context and give those points extra attention. That knowledge is part of the gain from the routine.

For a more formal reference on managing the risks of AI systems, the NIST AI Risk Management Framework organizes the topic into govern, map, measure and manage. It is reference material, and you do not need to apply all of it to individual use.

Illustrative scenario: one week for an operations manager

The example below is an illustrative scenario, not tied to any real company.

An operations manager at a distribution company has an alignment meeting every Tuesday. On Monday, she pastes her list of open items, replacing customer names with letters, and asks for a suggested priority order along with questions about what she might be forgetting. The AI suggests an order and raises a question about a delivery deadline she had not written down. She accepts the question, but changes two priorities because she knows about a commitment to a supplier that the AI did not.

On Tuesday, after the meeting, she pastes in her notes and asks for minutes with decisions, owners and deadlines. During review, she notices the AI assigned a task to the wrong person, because the notes mentioned two names in the same sentence. She fixes it and sends it out. On Thursday, she uses the AI to structure a feedback conversation with a supervisor, but writes the examples and the decisions herself. On Friday, she asks for a summary of the week and saves the prompt that worked best to repeat the following week.

What makes this use healthy is not the tool but the pattern: protected data, reviewed output, and decisions and difficult conversations kept under her responsibility.

How to apply map, limit, pilot and transfer to your personal use

The method we use for AI projects in companies also works at an individual scale, with less formality.

• Map: for one week, note which repetitive tasks consume your time and which involve text, summarizing or preparation. For broader process work, the guide on how to map business processes before applying AI covers the step-by-step.
• Limit: define which data and which decisions are out of the AI's reach, as in the section on limits above.
• Pilot: choose one or two tasks, use them for a few weeks and compare effort and quality with the old way. Without a comparison, the impression of gain is often misleading.
• Transfer: when a task works well, document the prompt and the review rule so another person on the team can repeat it, and decide whether to expand, adjust or stop.

The last point is the central one: a use that only works in your head is not a process, it is a personal habit. Transferring means writing down enough for another manager to reproduce the result.

When should you bring the routine to the team?

Bringing the routine to the team makes sense when three conditions appear together: your individual use is already stable, there is a written rule about data, and someone is responsible for reviewing what goes out. Without these conditions, the risk is multiplying the habit of pasting anything into any tool.

A prudent path is to start with a single shared task, such as the weekly minutes, and observe over a few weeks how many corrections were needed and what kind of errors appeared. With those records, the decision to expand stops being a matter of opinion. If you decide to measure more formally, the article on how to measure an AI automation pilot offers metrics and decision gates.

One last caution: tools connected to email, calendar or files widen what the AI can see and what it can do. Before granting that access, understand which permissions are being given. The OWASP Top 10 list of risks for LLM applications helps you spot problems such as malicious instructions hidden in incoming text.

Frequently asked questions

Do I need a specific tool to get started? No. The routine works with any general-purpose AI assistant, as long as the company has approved its use and you know how the tool handles the data you enter. Check the terms of your plan, especially on use of content for training and on retention, before you choose.

Can I paste customer and employee data into the conversation? By default, no. Anonymize first, replacing names and identifiers with roles, and only use personal data in tools the company has approved, with a clear purpose and attention to the LGPD. Sensitive data, such as health information, requires even more care and generally should not go into informal use.

How do I know if the routine is working? For each task, compare effort and quality with the previous way of doing it, and record how many corrections you had to make during review. If the text still needs heavy rewriting, or the errors keep appearing at the same point, adjust the prompt or remove the task from the routine. Stopping use on a task is also a valid result.

Can AI help evaluate the performance of team members? It can help you organize your own observations and prepare the structure of a conversation, but it should not produce or decide the evaluation. Evaluation involves context, fairness and accountability that belong to the manager, and the data involved is usually personal and sensitive.

Sources

• NIST: AI Risk Management Framework
• ANPD: Autoridade Nacional de Proteção de Dados
• OWASP: Top 10 para aplicações com LLM