An AI agent helps a manager with the preparation and writing that management runs on: one-to-one plans, status updates, delegation briefs and notes after difficult conversations. You paste in your notes or a thread, and it gives you a structured draft that uses what it already knows about your team and your formats. It doesn't manage anyone, and it can't see your inbox, calendar or your team's work.
Most of this work gets done in the gaps between meetings, or skipped. The one-to-one you meant to prepare and the update you write from memory at six are where an agent earns its keep. What you get out depends on what you put in: give it two hurried lines about a person and it will still write a confident appraisal of someone it has never met, so the evidence has to come from you.
Preparing one-to-ones and feedback
Most one-to-ones go wrong because nobody prepared, and the preparation is usually twenty minutes you don't have. Paste in what you noted about the person this week and ask for a plan:
"Notes on Priya this week: two wins, one missed deadline, a question about the rota. Build a 25-minute agenda and suggest the question I should open with."
You get an agenda with timings and an opening question. Because the agent remembers earlier weeks, it can also point out a concern you've mentioned twice without raising it.
The second job is turning a vague impression into feedback someone can act on. "Disengaged" is a label, and nobody can do anything with a label. Tell the agent what you've seen (late to stand-up three times, quiet in reviews, work still on time) and ask for two observations you could say out loud, plus a question that would test whether your read is right.
Appraisals are the same problem at a larger scale. The hard part is the evidence, and the prose is easy by comparison. Give the agent the year's projects, what was delivered, where things went wrong and a couple of examples, then ask it to tie every claim to an example and list any claim you can't support. That second list is often the most useful thing you'll get.
Writing for your director
Senior people read updates in a particular shape, usually headline first, then risks, then the decision they need to make. An agent that remembers your director's preferred format turns a two-week chat thread into one page in that order, with your requests at the top where they won't be missed.
"Thread from the last fortnight on the migration is below. Status update: what's done, what's late, what it costs, what I need from you. One page, no adjectives."
Explaining a delay is harder, because the temptation is to defend yourself. Ask for a short note that says what caused the slip, what you've changed and what it means for the date, without blaming anyone. Reading the cause written out in neutral language also helps you check whether you've understood it.
When you're asked for a business case with a day's notice, describe the workload, the backlog and what the delays are costing. The agent can lay out the problem, options, cost and recommendation on one page, and list the questions finance is likely to ask so you can answer them before they come up.
How do you delegate work so it comes back right?
Delegated work usually comes back wrong because the brief was three lines in a chat. A delegation brief needs the outcome, the constraints, what the person can decide alone, what has to come back to you, and a date.
"I'm handing the supplier review to Dan. He's capable but new to this. Write the brief, with his decision rights spelled out, and list the questions he'll probably ask."
The same thinking applies after meetings. Paste your notes and ask for three lists: decisions made, actions with an owner and a date, and things discussed but not decided. Unowned actions stand out immediately, which is the point of doing it. Send the result to the room the same day.
For decisions that someone will question in six weeks, ask for a short decision record: the context, the options you considered, what you chose, what you accepted as a trade-off, and what would make you reopen it. It takes a few minutes and saves an argument later.
If a project has slipped, paste the original plan and what's been done so far, then ask for options (cut scope, add people, move the date) with the cost of each. You still choose, but you choose from a written comparison instead of from memory in a meeting.
Hiring on evidence
Bad hires often start with a vague role. Before you write an advert, describe what you need the person to take off your plate and ask for a role brief: outcomes for the first year, what the person must already have, and what you can teach.
From the brief, the agent can build an interview scorecard: six questions that test those outcomes, what a strong answer contains and what a weak one sounds like. Use the same questions for everyone. After the interviews, paste your notes and ask for a comparison against the brief, based only on what each candidate said, along with anything you didn't test. Without that step, most people hire the candidate they liked.
When the new starter's date is set, ask for a 30-60-90 day plan with what they learn, what they own, when they should be left to get on with it, and the access and introductions you need to arrange before day one.
How can AI help you prepare for a difficult conversation?
A performance conversation goes better when you've worked out the opening line, the points you have to make and how you'll end it. Give the agent the evidence and what you've already said, and ask for that structure along with the reactions you're likely to get.
Then rehearse it. Ask the agent to play the other person:
"Play Dan. I'll open. Respond as someone who thinks he's being treated unfairly, push back hard, then tell me where I got defensive."
Saying the difficult sentence once, even to a chat window, makes the second time easier.
Afterwards, paste a rough account of what was agreed and ask for two notes: a short one to the person confirming it, and a factual one for your file. People remember conversations differently, and a written record protects both of you.
The agent can also help when you have to announce a change you didn't choose, such as a merger of two teams in April. Tell it what you know and what you don't, and ask for what you'll say on Monday: clear about what stays the same, open about what's still undecided, and with a list of questions people will ask that you can't answer yet.
A weekly rhythm that doesn't rely on memory
A scheduled check-in is a message the agent sends you at a set time. For a manager it does the remembering, which is the part that slips first in a busy week. A useful set:
- Friday, 15:00, next week. "Ask me what's happening next week, who I'm meeting one-to-one and what I owe people. Name the two things most likely to be dropped." Monday morning is too late to change much.
- Weekdays, 08:15, the people list. "Ask me who I need to speak to today and what I owe them. Five lines, nothing I haven't told you."
- Wednesday, 12:00, project pulse. "Ask for one line on each project I'm running and flag anything where the date or scope has moved since last week." A slip spotted midweek can still be fixed.
- Monday, 09:00, feedback log. "Ask what feedback I gave last week and to whom. Keep a list by person and tell me if someone hasn't had any for a month." By appraisal time, you have a year of notes instead of a blank page.
- First working day of the month, the team check. "Ask me about each person on my team in turn: workload, development, anything unresolved."
Start with two and add more once they're useful. The guide to scheduled check-ins has more on wording them.
Saving your formats as skills
Your team notices when your one-to-ones and updates change quality depending on how busy you are. A skill fixes the format so it doesn't. Work through a task once, then ask the agent to save how you did it. The ones that repay the effort for managers:
- Your one-to-one format: the agenda shape, the questions you always ask, how you take notes, and the rule that every meeting ends with one commitment.
- The weekly update: headline, progress, risks and decisions, in your director's order, so the tenth update reads like the first.
- The delegation brief: outcome, constraints, decision rights and check-in date. This one has the biggest return, because a good brief prevents a week of rework.
- The difficult conversation: opening, evidence, the request, the ending, and the follow-up note. Build it before you need it.
Which decisions about people should stay with you?
An agent can lay out a judgement for you to examine, but in management the judgement is the job.
Anything that affects someone's livelihood. Promotions, managing someone out, who gets the stretch project, changes to hours. Let the agent draft paperwork if it helps, then make the decision yourself and be ready to explain it.
Your read on a person. The agent sees the two paragraphs you typed. It doesn't know someone has been quiet since a bereavement, or that the missed deadline came with an apology at 11pm. Test your read in person.
The conversation itself. You can prepare with help, but you have to be in the room. A manager reading a script badly does more harm than one speaking in their own words.
Decisions with no good option. Redundancies, budget cuts, closing a team. The agent can help you say it clearly once you've decided.
Anything your employer restricts. Some organisations ban pasting work material into outside tools, and some roles are regulated. Check your policy first; AI at work: what your employer allows explains what to look for.
If a performance issue needs a formal process, bring in HR. A good conversation plan doesn't replace their advice, and doing things in the wrong order can make a fair decision unsafe.
Other people's data
Managers handle information about other people all day, so take more care here than on most jobs. This is general guidance, not legal advice.
Performance concerns, health issues, grievances and salary details should stay out of a chat unless you have a clear reason and your employer allows it. Usually you don't need them: "late three times in a month" is enough to get help with the feedback, and a placeholder such as [EMPLOYEE A] can stand in for the name, the team and the client.
Assume the agent's memory keeps what you mention in passing. Ask it now and then what it remembers about your team, and delete anything out of date. Treat any draft appraisal as a document about someone's employment: check every claim against evidence and be ready to defend each sentence out loud.
Personal data about employees falls under the GDPR (or the UK GDPR) in Europe and under your local data-protection law elsewhere, and your employer's policy decides what's allowed. What happens to your data lists the questions to ask a provider.
Getting started: our part, then your first week
We set Able° up for you and guide you through the few steps on your side, mainly connecting it to Telegram or LINE. After that, give the first week one job a day:
- Tell it about your team. Roles, what each person is working on, who's new, what you're trying to change this quarter, and who you report to and how they like updates. Leave names out if you prefer. This context is what stops answers sounding like a textbook.
- Prepare one real one-to-one from the notes you have. If it suggests a question you wouldn't have asked, ask it.
- Write one status update from a long thread, the kind that usually takes forty minutes, and compare its structure with what you'd have sent.
- Turn your last meeting's notes into decisions, owners and dates, and send the result to everyone who was there.
- Set two check-ins: the Friday week-ahead and the Monday feedback log.
At the end of the week, keep the task that saved the most time and drop the others until that one is a habit.
Common questions
Can it sit in my meetings and take notes?
No. It has no access to your calls or calendar. You paste your own notes afterwards and it turns them into decisions, owners and dates, which often makes a better record than a transcript because you've already chosen what mattered.
Will my team know I used AI to prepare?
Only if you tell them. What they notice is whether you were prepared and whether your feedback was specific. A structure you drafted with help, filled with your own evidence, comes across as good management.
Is it worth it if I only manage two or three people?
Possibly not. If you talk to everyone daily, preparing in a chat may cost more time than it saves. The tasks that still help with a small team are the ones you tend to avoid, like the difficult conversation or the business case.
How is this different from the AI in our HR system?
HR systems record and administer what has already happened. An agent helps beforehand and afterwards, with the preparation, the rehearsal and the follow-up note, and it remembers your own habits across roles in a way a system built for the whole organisation doesn't.
Will it make my writing sound generic?
It will unless you correct it. Give it examples of your own writing, ask it to keep your phrasing, and point out anything that reads like a press release. Saved as a rule, two or three corrections usually stick.