Management is a profession where the most important work is also the hardest to start. Writing performance reviews. Having a difficult conversation with someone who isn't meeting expectations. Preparing for a skip-level with a senior leader who will ask pointed questions about your team's performance. Drafting the job description for a role that's been open for three months because every version you've written has felt wrong.
These tasks share a common characteristic: they require sustained writing and thinking, they carry real stakes, and they are easy to defer when there's a meeting to prep for or an urgent Slack message to respond to. The procrastination is rational in the short term and corrosive in the long term. People don't get the feedback they need to grow. Roles don't get filled. Difficult conversations become crises that required a conversation six months earlier.
The core value of AI for managers is not that it makes these tasks easier to do well. It's that it removes the blank-page friction that causes them to get deferred. You can start from a draft instead of from nothing. That shift in activation energy is significant.
Performance Review Language That Actually Means Something
Performance review writing fails in one of two directions. It's either vague to the point of uselessness ("Sarah is a team player who contributes positively to the group") or it's harshly specific in a way that reads as punitive rather than developmental. The goal is the third thing: specific about what happened, clear about impact, oriented toward what comes next.
AI doesn't produce this on its own. It produces it when you give it specific input. The prompt that works:
"Write a performance review section for a [role] who reports to me. Here are specific things they did this review period: [list 4-6 specific examples with outcomes where you have them]. Here are areas where I've observed they need to develop: [specific, behavioral observations, not personality judgments]. Write this in a tone that is direct, specific, and developmental. Not harsh, not vague. The rating is [meets expectations / exceeds / etc.] if that's relevant. Don't use corporate filler phrases. Separate strengths from development areas."
The quality of the output is entirely determined by the quality of your input examples. If you give AI vague observations, it produces vague reviews. If you give it specific examples, the draft is specific. This is also where the value of keeping contemporaneous notes during the review period pays off: "in the Q3 product launch, she identified a data quality issue in the dashboard two days before launch and coordinated the fix across three teams without being asked" is the kind of input that produces a review sentence worth writing.
Review everything it produces before it goes into your HR system. AI sometimes softens critical feedback to the point where the development need is buried, or it writes language that sounds like a specific example but is actually generic. You know which one is happening; edit accordingly.
1:1 Agenda Prep from Prior Notes
The best 1:1s are not status updates. They're conversations about what the person is working through, what's blocking them, where they want to go, and what you can do to help. Getting to that conversation requires that you've thought about the person before the meeting, not opened your calendar at 1:58 PM and started reading the notes from last time.
Here is the workflow. Keep your 1:1 notes in a consistent place, Notion, a shared doc, your own notes app, anywhere searchable. Before each 1:1, paste the last two or three sets of notes into Claude with this prompt:
"I'm about to have a 1:1 with someone who reports to me. Here are notes from our last few meetings: [paste notes]. Based on these, give me: 1) Any open items or things I said I'd follow up on that don't appear to have been resolved. 2) Themes I'm seeing in what they're working on or struggling with. 3) Three specific questions I could ask that would be genuinely useful to them, not generic check-ins. Don't be formal. Write like I'm prepping for a real conversation."
Five minutes of this prep makes the difference between a 1:1 that's transactional and one where the person feels like you're actually paying attention. Over the course of a year, the compound effect on trust and retention is substantial.
Meeting Summary Generation from Raw Notes
After a team meeting, project sync, or strategy session, the summary is what determines whether anyone acts on what was discussed. A summary that takes 30 minutes to write gets deprioritized. One that takes 5 minutes gets done.
The workflow: take rough notes during the meeting, including decisions made, action items with names attached, open questions that need follow-up, and anything notable. After the meeting, drop them into Claude:
"Turn these meeting notes into a clean summary. Format: one paragraph of context and decisions made, then a bulleted list of action items with owner and timeline where I've noted them, then any open questions still unresolved. Don't add information I haven't given you. Keep it under 200 words unless the meeting content requires more. Here are the notes: [paste]."
Edit for accuracy and anything time-sensitive, then send. The team gets a summary that actually captures what happened. You spend 7 minutes instead of 25.
Skip-Level Interview Preparation
Skip-level conversations, whether you're being skipped to by your manager's manager or you're conducting them to get a read on your own team's experience, require specific preparation. The failure mode is generic questions that produce generic answers and leave the conversation feeling like a formality.
AI can generate substantive, non-generic questions when you give it context about the team situation. Tell it: the team's current priorities, any known tensions or changes that have happened recently, what you're actually trying to learn, and any specific dynamics you want to probe without being leading. Ask for 8 to 10 questions that would surface honest information about team experience, management effectiveness, and organizational friction. You'll use four or five of them. The preparation of thinking through what you actually want to learn is as valuable as the questions themselves.
Difficult Conversation Scripts
The difficult conversation that gets avoided is almost always more expensive than the one that happens. Underperformance gets worse. Bad behavior patterns calcify. Relationships that could be repaired through honest feedback instead deteriorate through accumulated resentment on both sides.
The reason these conversations get avoided is not that managers don't know they need to happen. It's that they don't know how to open them, fear the person's reaction, or have tried to start the conversation and then softened it to the point where nothing was communicated. AI can give you a scripted opening that's clear, non-accusatory, and sets the conversation on the right track.
"Help me script the opening of a difficult conversation I need to have with someone on my team. The situation: [describe specifically what happened or what the pattern is]. What I need them to understand: [the core message, stated plainly]. What I want from the conversation: [a behavior change, a shared understanding, an action plan, whatever is true]. Write me a 3-4 sentence opening that: names the issue directly, doesn't accuse or attack, and opens space for their perspective. Then give me two or three follow-up questions for if they get defensive or deflect."
Practice the opening out loud before the conversation. The first thirty seconds are the hardest part. Having a clear, rehearsed opening means you're less likely to hedge it into incoherence under stress.
Job Description Writing and Hiring Rubrics
Job descriptions are one of the most influential documents a manager produces and one of the most neglected. A vague job description attracts a wide and poorly-filtered applicant pool. One that's specific about what the role actually requires, what success looks like in 90 days, and what kind of person thrives in your team's environment self-selects for people who actually fit.
AI produces a solid job description draft when you give it: the actual responsibilities in your own words (not HR boilerplate), the skills and background that are genuinely required versus nice to have, what the person will own versus support, and what the team culture actually is. Ask it to avoid inflated credential requirements and to write the "about us" and "what you'll do" sections in language that sounds like a real company, not a corporate template.
Hiring rubrics are worth building with AI and worth the time even for roles you've hired before. A rubric forces you to define what "great" looks like on each dimension before you're in an interview making real-time judgments. Ask AI to draft a scorecard with specific behavioral indicators for each competency you've identified, then refine it with your team before you start interviewing. Interviewers who use a rubric make more consistent decisions and are less susceptible to halo effects and gut-feel biases.
OKR and Goal Language
Writing objectives and key results that are actually measurable is harder than it looks. Objectives tend toward the vague and inspirational; key results tend toward activity metrics that don't tie to outcomes. AI is useful for stress-testing both. Give it your draft objectives and key results and ask it to flag any that are not measurable, any that measure activity rather than outcome, and any that seem disconnected from each other. Ask it to suggest more specific language for the ones that need it.
This is a good example of using AI as a reviewer rather than a generator. The goals come from your understanding of what the team needs to accomplish. The AI's job is to help you articulate them with sufficient precision that you'll know at the end of the quarter whether you achieved them.