Notes: Prepare for conversations you cannot afford to avoid

Start with Clara's case. She manages a customer-service team in a prepared example, and she has already moved her 1-on-1 with Liam twice.

From the records, she knows two things:

• Tuesday's and Thursday's shift notes left out 2 open billing problems.
• Both customers called back the next morning and explained their problem again from the beginning.

What she does not know is why Liam left those details out.

Many leaders delay a talk like this because they feel they need a complete verdict and a polished speech that answers every excuse. Walking in with all of that usually makes you sound like a prosecutor, and the other person gets defensive even when your words are polite.

There is a lighter way in. Clara arrives ready to state the one observation she knows for certain, and to ask an open question she genuinely wants Liam to answer. She no longer has to pretend she knows what was in his head at the end of the shift.

That draws a line between two jobs:

• Preparation separates proven facts from frustration and puts your opening question into fair, calm words.
• The conversation itself is the only place you learn what obstacle the other person actually faced.

You can follow Clara's case all the way through without sharing any personal details from your own work.

Clara's first thought on Friday morning is: "Liam is rushing out the door at shift close and does not care about leaving a mess for the morning shift."

If she opened the 1-on-1 with "You need to care more about the morning team's workload", Liam would spend the meeting defending his character instead of fixing the shift notes.

So she does not ask AI to soften the accusation. She uses feedback-coaching to examine her own starting material, and she tells the agent to ask questions before it proposes any wording.

The agent asks 4 questions, and her answers come straight from the records:

1. Observed action. The Tuesday and Thursday shift notes, with the open-issues section left blank.
2. Verified impact. The Wednesday and Friday call logs, where 2 customers called back to repeat their billing issue.
3. Prior expectation. The team's handover rule already says every shift note lists any open customer issue.
4. Feasible support. A 3-item end-of-shift checklist, or pairing Liam with a senior rep to review how complex billing problems are logged.

When the agent asks whether the notes prove why the section was blank, Clara sees that "rushing" and "not caring" are guesses.

Her prepared opening names the 2 notes and the 2 repeat calls, then asks: "What made capturing those open issues difficult at the end of those shifts?" That question leaves room for an answer she does not have yet.

Whenever you do this with a real colleague, remove identifying personal details and give AI only the non-confidential facts of the work.

Clara keeps the opening she just prepared and asks the agent to play Liam across 3 short rounds. With the microphone on, she speaks each reply instead of typing it.

Each round tests a different reaction:

1. Liam is unsure what counts as an unresolved issue.
2. He feels squeezed by end-of-shift call volume.
3. He disagrees that the missing note caused the repeat call.

After each round the agent checks whether Clara:

• States the concern clearly.
• Asks an open question.
• Attends to what the simulated Liam just said.
• Separates a newly surfaced fact from an assumption.

Round 2 is where she slips. The simulated Liam says he got stuck on an escalated billing call 5 minutes before the end of his shift on both days, and he has to log off right on the hour. Clara answers: "Everyone gets late calls, so you still have to fill out the shift note before logging off."

The agent quotes that line back and points out that she answered with a rule before understanding the obstacle. A late escalation against a strict log-off time is a real conflict, and repeating the rule does not fix the handover.

Her second attempt calls the late escalation important context, asks how he is handling the overtime limit, and asks what would let him flag the open issue in one line for the morning shift. The standard stays, and now she is listening to the new fact.

If you stop rehearsing as soon as the simulation agrees with you, you have only practised the easy version.

Darius, a team lead in a prepared example, asks Tessa to create an onboarding guide for new starters arriving on the first of next month. He has the rough onboarding notes and the confirmed start date.

The outcome he needs is a guide that lets a new starter finish every first-day essential without hunting down a manager for basic setup steps.

His first draft contradicts itself. It opens with "Tessa, you fully own creating the new-starter onboarding guide", then dictates 5 section headings, the font layout and approval after every page. That turns Tessa into a typist and Darius into a bottleneck.

The repaired assignment has 4 parts:

1. Desired result. A new starter completes the first-day essentials without chasing a manager.
2. Decision freedom. Tessa chooses the layout and proposes the order of first-day tasks. Any change to which security permissions new starters get on day one comes back to Darius.
3. Support. The existing onboarding notes and a named IT operations contact for setup questions.
4. Check-in. One complete working draft 10 days before the new starters arrive, or sooner if she is blocked.

Then Tessa raises a problem. The IT contact is on leave this week, so she cannot check the VPN setup steps until next Tuesday.

Taking the IT section back would be the easy move. Darius adjusts the support instead: he confirms a backup IT contact by tomorrow afternoon, and Tessa keeps drafting and keeps owning the whole guide.

A good test for any assignment you write: your colleague could say in their own words what result they are aiming for, which choices are theirs, and when to come back to you.

Across this section, three things changed in the words:

• Clara's accusation about not caring became 2 incomplete notes, 2 repeat calls and an open question about what made the end-of-shift note difficult.
• Her rehearsal showed that she answered a new fact with a rule. Her second reply explored the overtime obstacle and kept the handover standard.
• Darius's assignment went from formatting rules and page-by-page approvals to a clear outcome, real freedom over layout and order, a named support contact and a return rule for security permissions.

The same principle sits under all three. Useful preparation separates what you already know, like Clara's 2 shift notes, from what only the other person can tell you, like why those notes were left blank.

Even a thorough rehearsal does not mean you understand the real colleague across the table. A simulated exchange is practice, not knowledge.

So before any feedback, coaching or delegation conversation, name one question that must stay open:

• For Clara: what actually happened at the end of Tuesday's and Thursday's shifts?
• For Darius: what obstacles does Tessa see once she inspects the onboarding notes?

The final section uses Darius's assignment as its lens. When you hand routines and multi-step workflows to AI agents, the same 4 parts decide whether delegating saves you effort or gives you a second job supervising and fixing the software.