Sooner or later someone asks: "Will AI replace me?" At Horizonte it was Juliana, at the end of the workday. There are two easy answers, and both fail: promising that nothing will change or saying that's just the market. In this lesson you practice the third: telling what changes, what doesn't change and what isn't known yet.
4Close with a commitment you can keep, with a date
Goal: handle the "Will AI replace me?" conversation honestly: what changes, what doesn't change and what we still don't know, without promising.
01 · The honest conversation
Five parts, in this order. Click each one.
Click each block in the diagram to see what it does.
Structure recommended by INEMA for the conversation (what changes, what doesn't change, what we don't know). It is not an HR protocol or legal advice.
02 · The conversation in four steps
What to say, what to avoid and what Patrícia did.
1. Listen before you explain
Someone who asks whether they'll be replaced isn't attacking the change; they're asking for information no one has given. Thank them for the question and ask back what they've seen.
Avoid the two quick exits: "don't worry" (a promise) and "that's the market" (abandonment). Both end the conversation and leave the fear untouched.
At Horizonte (fictional study company)"Ever since Hora came in, I only answer what she doesn't know." Juliana brought up a fact: the easy part of her work is gone. That's where the conversation starts.
2. What changes, what doesn't change
Bring the matrix from lesson 2. Show the tasks that left, the ones that came in and the ones that stay with the person, with the reason. A written decision carries more weight than any comforting phrase.
Don't sugarcoat: if Hora already handles half of the questions, say so. Trust comes from the person recognizing in your words what they see in their day.
At HorizonteLeft: catalog and delivery deadlines. Came in: checking new customer records and the exception queue with Patrícia. Stays: the angry customer, the customer who only talks to her.
3. Say what you don't know
No one knows for sure what roles will look like a year from now. The Stanford Digital Economy Lab (08/2026) recorded a relative drop of about 19% in employment among young people aged 22 to 25 in the occupations most exposed to AI, without claiming that AI is the cause.
This is how you talk about a hard data point: what it shows and what it doesn't prove. Do the same with your own case: say what you know, what you don't know and who decides the rest.
A phrase that works"I can't guarantee the role a year from now: it doesn't depend only on me. What I can do is tell you first, myself, and not through the hallway."
4. Close with a commitment
A good commitment depends on you and can be checked: one conversation a month, the person in the testing of the agent's changes, the unanswered questions answered by a set date.
Also give the person a concrete task. Someone who takes part in how the agent changes has less reason to fear and more reason to contribute: the person on support knows where the agent gets it wrong.
At HorizonteTwo weeks noting every case Hora got wrong or passed to Juliana; on Friday, the two of them look at it together. And Juliana joins the testing of Hora's next changes.
03 · Simulation: "Is Hora going to take my place?"
Simulated conversation for practice. Juliana comes to see you at the end of the shift; choose each answer. The comment appears right after.
Juliana · customer service (simulation)Horizonte Distribuidora, fictional study company
Choose your answer. The comment appears right after, and the conversation continues based on your choice.
04 · Templates to copy
The script for the 30-minute meeting with the team and a prompt that looks for promises. The tab next to this one shows Patrícia's.
Meeting script · 30 min
MEETING: WHAT CHANGES WITH <agent name> · team <name> · <date and time> · led by: <name>
00-05 OPENING
Why we're here: <one sentence>. I'll tell you what I know, what changes and what I don't know.
05-12 WHAT CHANGES (from the work-split matrix)
Left your plate: <tasks> · Came in: <tasks> · Who decides what: <...>
12-17 WHAT DOESN'T CHANGE, AND WHY
<customer relationship, decisions, ...> · because: <written decision>
17-22 WHAT WE DON'T KNOW
<what nobody knows yet> · who decides: <name> · when I'll come back with an answer: <date>
22-28 QUESTIONS
Write down all of them. The one I can't answer: I'll answer by <date>, via <channel>.
28-30 COMMITMENTS (only what depends on me)
<commitment 1, with date> · <commitment 2, with date>
Everyone's task by <date>: <...>
Next conversation: <date>
PHRASES THAT HELP: "that's a fair question" · "I don't know that, and I'll tell you when I do"
· "that decision is <name>'s; I'll bring the answer by <date>"
PHRASES TO AVOID: "don't worry" · "nobody is going to lose their job" (if it wasn't decided)
· "that's the market" · "I can't talk about that" (without saying when you can)
Prompt · look for promises
You review the script of a meeting in which a manager will talk to the
team about the arrival of AI agents. Don't rewrite it: point out risks.
1. List every sentence that sounds like a promise about employment, position,
salary or savings, and say whether the script shows who decided that.
2. List generic sentences ("that's the market", "it will get better") with no
concrete task behind them.
3. Say whether the script has the three parts: what changes, what doesn't change and
what we don't know.
4. Point out commitments without a date or that don't depend on the person leading.
Script:
"""<paste here>"""
Simulation: illustrative example
Meeting: what changes with Hora · sales · Friday, 9 a.m. · led by: Patrícia · Horizonte, fictional study company
Opening: "Hora has been with us for a month and nobody has sat down to talk about it. I'll tell you what I know, what changes and what I don't know."
What changes: catalog and delivery deadlines are no longer Juliana's job; she checks new customer records and the queue with me. Discounts of 5% to 10% and orders above R$ 5,000: Hora prepares, I approve.
What doesn't change: upset customers, visits and negotiation stay with you, because that's relationship work.
What we don't know: how customer service will look in a year. Carlos decides on roles; I'll take the question to him and come back by 11/13.
Questions: written down in the team channel.
Commitments: a monthly conversation; nobody hears about a change in the hallway; Juliana on the testing of Hora's changes. Task: two weeks noting down Hora's mistakes.
Promising a job. If the company hasn't decided in writing, don't promise. Say who decides and when you'll come back.
Only what changes. Without what doesn't change, the team imagines the worst.
Skipping the "we don't know." It's the part that gives credit to the rest.
A question with no owner. A question you couldn't answer needs a date and a channel.
Closing without a task. Without a next step, the fear comes back with the next piece of news.
05 · Video: the resistance that starts with the manager
Nei talks about the other kind of resistance: the manager who says "that's an IT thing." Before you lead the conversation with your team, take a look at your own.
Practice in the simulation, write Patrícia's script, and then your own.
1 · The conversation with Juliana
10 min
Do this lesson's simulation. If you don't reach the top score, read the comments, start over and try again.
Done when: you finished the simulation with the top score and can say, in one sentence each, what changes, what doesn't change and what isn't known, for Juliana.
2 · Patrícia's script
20 min
Using the matrix from lesson 2, write the script for Patrícia's 30-minute meeting with the sales team. Run the promises prompt and fix what it flags.
Done when: the script has all six blocks with minutes, at least 3 tasks that change and 2 that don't, a "we don't know" with who decides and a reply date, and no promise about jobs or roles.
3 · Your meeting
30 min + the meeting
Write the script for your team and schedule the meeting. After it, note the questions that came up and answer the ones left open by the date you promised.
Done when: the meeting happened on the scheduled date, the questions are written down, every unanswered question has a date, and the dated commitment was kept or is on the calendar.
07 · Quick test
5 questions. The explanation appears after your attempt.
Question 1 of 5
Juliana asks whether Hora is going to take her place. What is the best first answer?
Correct
Promising that nobody loses their job is not your decision; saying it's the market abandons the person. The right answer acknowledges the question and starts by listening.
Not that one
Promising that nobody loses their job is not your decision; saying it's the market abandons the person. The right answer acknowledges the question and starts by listening.Review: Simulation: "Is Hora going to take my place?"
Question 2 of 5
Why say out loud what is still not known?
Correct
Someone who only brings certainties sounds like they're hiding something. Admitting the uncertainty, with who decides and when you'll come back, makes the rest get heard.
Not that one
Someone who only brings certainties sounds like they're hiding something. Admitting the uncertainty, with who decides and when you'll come back, makes the rest get heard.Review: The honest conversation
Question 3 of 5
How do you cite the Stanford study in the conversation?
Correct
The study shows a drop and doesn't state the cause. Repeating only the number, as if it proved the cause, scares people and teaches the team to read data wrong.
Not that one
The study shows a drop and doesn't state the cause. Repeating only the number, as if it proved the cause, scares people and teaches the team to read data wrong.Review: The conversation in four steps
Question 4 of 5
Which of these is a commitment Patrícia can make?
Correct
A good commitment depends on who is promising it and can be checked. Job security is not her decision alone.
Why close the conversation with a task for the person?
Correct
People who handle customers know where the agent makes mistakes. Noting down the cases and joining the testing of the changes turns fear into contribution.