The agentic company and the 2027 manager|INEMA.CLUB-PRO

Lesson 2 · 17 min

Managing an agent looks like managing people

What a company did with a new employee, it needs to do with each agent: give the role, the goal, the tools, the limits and the evaluation. The words change, not the logic. In this lesson you build the translation of a job into an agent, see the two items that people did not have, and learn where the comparison stops.

Start the lesson
In a bright meeting room, a manager and an employee review a printed sheet; beside them, a laptop shows a chat panel with colorful bubbles.

In this lesson you will

  1. 1The six pairs that carry people management over to agent management
  2. 2The two items that people did not have: explicit context and graduated autonomy
  3. 3Whoever posts does not approve: the agent prepares, the person decides
  4. 4Where the comparison stops, and the mistake of treating people like machines

Goal: translate a person's job, responsibility, tools, limits and evaluation into an agent, and say where the comparison stops.

01Before and now, item by item

Each block shows what was done with a person and what is done with the agent. Click to see the example.

Click each block in the diagram to see what it does.

Source of the thesis: INEMA's AI Management area. How to write each piece for a real agent is the G3 course (Agents).

02The same logic in four situations

Three professions where the rule already existed for people, and the point where the comparison stops. The Horizonte examples come from the fictional study company.

1. The same work

Think about what a good manager knows about the team: what the goal is, who does what, what information each person needs, how much autonomy they have, how to measure, and when to step in. Now swap "person" for "agent". It's practically the same list.

Microsoft gave this a name in the 2025 Work Trend Index: the agent boss, a human manager of one or more agents, and the human-agent ratio, the balance between human oversight and agent efficiency. It's the vendor's framing, useful as vocabulary.

At HorizontePatrícia became Hora's boss without anyone telling her. She already knew how to manage Juliana; she just hadn't written down for Hora what she would say to a new saleswoman.

03Template to copy: the before × now table

One page per role. Start with what you would tell a new person in that role and translate it line by line.

Before × now table
BEFORE × NOW TABLE · role: <e.g., salesperson who takes orders> · agent: <name> · <date>

ITEM          | FOR THE PERSON (before)          | FOR THE AGENT (now)                | HOW I CHECK
Role          | <job title>                      | <process + result, 1 sentence>     | <does an outsider understand?>
Goal          | <responsibility>                 | <result you can check>             | <where it is recorded>
Context       | (learned by being around)        | <what it gets in writing>          | <who keeps it up to date>
Tools         | <systems, phone, spreadsheet>    | <look up / create / send ...>      | <which one can change data?>
Autonomy      | <agreed limit>                   | <does alone / prepares / never>    | <ceiling as a number>
Evaluation    | <performance review>             | <indicator + target + when I look> | <this week's number>

PERSON RESPONSIBLE FOR THE AGENT'S RESULT: <name, from the business>
WHAT STAYS WITH THE PERSON (and why): <decision, relationship, exception ...>
WHERE THE COMPARISON STOPS: <what I look after in the person and not in the agent>
Prompt · review the table
You are an experienced manager reviewing the translation of a human role into
an AI agent. Don't rewrite the table: point out gaps.

Table:
"""<paste here>"""

Answer, in this order:
1. Is any agent line an intention ("help", "be quick") instead of
   something you can check?
2. Does the context list what the agent needs to know and a person knew
   without anyone writing it down?
3. Does the autonomy have a ceiling as a number where there is value?
4. Is the responsible party a person from the business, with a name?
5. Is what stays with the person justified, or is it just what was left over?
Simulation: illustrative example

Replenishment saleswoman → Hora · Horizonte Distribuidora, fictional study company

ItemPersonHora
Rolereplenishment saleswomanrecord replenishment orders from registered customers that come in through WhatsApp
Goalhit the monthly targetorder in the app that picking doesn't need to correct
Contextknew who pays latecustomer record, current price list, discount policy in writing
Toolsphone, app, spreadsheetread customer record and price list; create order; never mark as paid
Autonomydiscount "as far as Carlos allows"up to 5% on her own; 5–10% prepares; order above R$ 5 thousand waits for Patrícia
Evaluationconversation at the end of the month% of orders accepted without correction, checked every Friday

Responsible: Patrícia Lima. Stays with the person: negotiation above 10% and new customers.

04Practice

At Horizonte, to practice. Then with a role from your team.

1 · Hora's table

20 min

Read Hora's prompt and configuration in the lab. Copy the template and fill in the before × now table, taking a replenishment saleswoman as the "before". Use this lesson's fictional policy (5%, 10%, R$ 5 thousand). Only then compare with the example.

Done when: all six lines are filled in, each line in the agent column has something another person can check (a record, a number or a rule), the autonomy cites the three numbers of the policy, and the responsible party is Patrícia.

2 · A role on your team

25 min

Pick a repetitive task that a person on your team does today and that an agent could take over in part. Fill in the table, run the review prompt and write the two final lines: what stays with the person and where the comparison stops.

Done when: the table has all six rows, a named owner, at least one thing that stays with the person along with the reason, and one sentence saying what you look after in the person and not in the agent.

Lab piece: Horizonte's WhatsApp agent (prompt, configuration and conversations; fictional study company).

05Quick quiz

5 questions. The explanation appears after your attempt.

Question 1 of 5

What are the two items that managing agents has and managing people did not?

06Summary

Same logicRole, goal, tools, autonomy and evaluation, just like with people.
Two new itemsExplicit context and graduated autonomy, written down.
Prepares, doesn't decideThe agent goes up to the signature gate; whoever posts does not approve.
Where it stopsA person has workload and a future; an agent has accuracy, time and cost.

Going further

Course G3 · Agents: the job sheet, delegation and real authority limitsCourse Companies of 2027: people and agents, side by sideAI Management Area: the thesis and the accounting, legal and health cases
Previous lessonNext lesson