How I use AI for management

The feedback I give and receive in 1:1s used to rely heavily on memory. Whatever stood out in the last week or two would dominate, and the slower patterns from a month ago would slip out of view.

Now I connect Claude to Slack, Asana, Gmail, and Drive and ask it to pull the real evidence. What did this person work on in the last month? Where did they shine? Where did they get stuck? The output is a draft of feedback grounded in what happened, not in what I half-remember.

I still go through it manually and pick what lands, and I still use my own pattern recognition for the things only I would notice. But starting from real evidence makes the feedback more objective.

The other thing I love about this: if both of you do it, you sometimes surface different things, which is interesting to compare. The AI working from the data finds patterns I'd missed, and I bring context the data can't see.

It can also help you surface what you could have done better as a manager. If you're reviewing your own interactions with a team member, Claude can point out where you could have communicated more clearly, followed up sooner, or given more support.

I also use Claude to help me accept constructive criticism better. I can run the feedback through it and ask "Based on what you have noticed, does this ring true? What are the examples?"

Some decisions you can't take to your team because they're about the team. Restructuring. Promotions. How to handle a performance issue. How to think about org design when something isn't working. The kind of question where the obvious sounding boards are themselves part of the picture.

A board member or coach can help here, but they aren't always available, and the decision has to be made now. AI fills that gap well, as long as you give it real context and ask it to push back rather than agree.

When I was thinking through a restructuring question, I asked Claude what a good management consultant would say. It pointed out the thing I was avoiding: that the question I was framing wasn't the question I needed to answer. The real question was sitting underneath, and I'd been circling it without naming it. That reframe was worth more than an hour of going round in circles.

The trick is to instruct it to disagree with you. Otherwise you get sycophancy dressed up as analysis.

This is the use case I'd most want every manager to try. The learning curve is steep in a way that nothing else I've found matches.

After a coaching call, a 1:1, a difficult conversation, or a board meeting, I'll pull the transcript and ask Claude to rate it out of 10 and tell me what I did well and what I could improve. Specific things. Did I create a space for people to come up with their own solutions? Did I pivot too fast from problem to solution? Did I ask leading questions? Did I name the thing or dance around it? This could be particularly useful if you upload your professional development goals to Claude as a reference.

The feedback is usually accurate enough to act on by the next call. That tight loop — do the thing, get specific feedback, do it slightly better next time — is transformative for skills that normally take years to develop because no one ever tells you what you're doing wrong.

A few things I now know about my own coaching style that I didn't before doing this: I sometimes close down a thread with a summary when the person was about to go deeper. I underuse silence. I redirect from emotional content to practical content faster than I should.

I can also use this to spot patterns in other people. When I sense that something is off about how a team member shows up but I can't articulate it, I ask Claude to look across meeting transcripts with this person from a given time period.

A practical note: when the AI gives you an example, go back and check the transcript. I caught myself once accepting a vivid example that turned out to be slightly off. The patterns were real but the supporting case wasn't quite right. Always verify before using anything in feedback to someone else.

Once you have transcripts, the second use is treating them as raw material for everything that comes next.

• Default-on transcripts. Set transcript creation as the default for internal meetings. The tax of remembering to turn it on every time is too high. Default-on means you never have to wonder later whether you have one.
• Draft follow-ups. Use transcripts to draft follow-up emails, summaries, and updates. Drafting from a transcript is much easier than drafting from memory. The AI does most of the work, and you edit for tone.
• Extract action items. After a meeting, ask Claude to extract every assignment from the transcript with the owner and rough timing. This catches things that would otherwise slip, especially the small commitments people make in passing that no one writes down.
• Surface writing ideas. If you've discussed something interesting in a meeting, ask Claude to flag what might be blog-worthy. The work of turning it into a polished piece is still yours, but having a draft to react to lowers the activation energy.

The general principle: a transcript is a free input into half a dozen other deliverables. Most teams generate them and then never touch them again. Use them.

Before a conversation I'm dreading, I'll ask Claude to draft a full script: opening line, the pivot to the real topic, what to say if the other person responds in different ways. Then I do a rough role play, with Claude playing the other person.

You can also use Claude's Voice chat for the role-play, which makes the rehearsal feel closer to a real conversation.

This isn't about reading from a script in the actual meeting. It's about having walked through the shape of the conversation before I'm in it. By the time I'm in the room, I've already had the version where the person gets defensive, the version where they're relieved, the version where they want to negotiate.

The biggest unlock here is knowing what my fallback position is. I used to walk into hard conversations hoping to figure things out in real time. Now I walk in knowing what I want to leave with, and what I'll accept as the next-best outcome.

This is the use case I value most and the one that's hardest to describe.

When I'm about to send a Slack message, make a decision, or react to a team dynamic, I'll often paste my draft thinking into Claude and ask: "What am I missing? What's the read on this that I'm not seeing? Is there an emotional driver in here I should look at before I act?"

Once, I was about to send a fairly direct message asking a team member to clarify a decision they'd made. Claude pointed out that the phrasing was framed as a question but read as a rhetorical accusation, and that if I sent it as written I'd be undermining the collaborative tone I usually try to set. It was right. I rewrote it in two minutes and the conversation that followed was much better than the one I'd nearly started.

A coach would catch this. So would a good friend. But neither is available when the message is half-drafted.

It's worth giving Claude as much context as you can. You can say things like "Scan the transcripts for our last 10 1:1 meetings" so it has a richer picture of the relationship and communication patterns before giving you advice.

Once a quarter, I'll let Claude pull from my calendar, Asana, Slack, Gmail and Drive (and maybe even DoneThat!) and tell me what my time went to. Compared to what I tell myself I'm prioritising. Compared to my role description.

The output is uncomfortable in useful ways. The first time I did this it surfaced that a meaningful share of my week was going to operational tasks that shouldn't be a part of my role, and that the strategic and writing work I claim to value most was the work I was deprioritising under load.

This kind of analysis used to require a coach with access to your calendar and a lot of patience. Now it takes a prompt and twenty minutes.

I'm often the last person looking at grant applications, donor emails, and external comms before they go out. That's appropriate for the role but it makes me a bottleneck, and it means everything sits in my queue until I have the focus to give it a proper read.

So I built custom AIs trained on what I look for. One for grant applications, one for fundraising comms, one for general external writing. Each has my voice, my standards, the things I always flag, the questions I always ask. With enough context and instruction, the output is a thorough first review, much more complete than a quick pass from someone who hasn't internalised what I care about.

This isn't auto-approval. The team still sends me the final version. But by the time it reaches me, the obvious things are caught and the draft is closer to the bar I'd want anyway. I'm reviewing a stronger document, which means my review adds more value at the margin.

The setup is worth doing properly. I gave each AI examples of the kind of feedback I give, the kinds of things I tend to push back on, my voice on the page, and the standards I'm holding to. The current versions are useful enough that I only need to read it over once or twice.

Synthesising feedback from others: When we run a 360 review for someone on the team, we get responses from multiple people across multiple questions, often with overlapping themes and varying levels of specificity. Claude can pull this into themes, count how many respondents raised each issue, separate the sensitive items that should be handled carefully, and flag where the picture is consistent versus contested.

The output isn't the final review. But it cuts the time from raw responses to draft synthesis from a full day to about an hour, and it catches patterns I'd miss reading through linearly.

Giving yourself a 360: This is the use case I'd most recommend to any manager who's been in their role for more than a year.

The output is uncomfortable in useful ways. Claude has access to artifacts of your work that no single colleague sees in full. It catches things like: I'm clearer in writing than in meetings. I delegate operational work but hold onto strategic work too long. Some of what comes out you'd already know if you had a coach who'd watched you for a year. Some of it surprises you.

Triaging emails with Cora (or another email triage tool): I've been using Cora for a while and it has drastically cut the time I spend in my inbox. It sorts incoming email into what needs a response and what's just an FYI, then sends me a digest twice a day. That alone means I can work distraction-free. The feature I value most: for emails that need a reply, Cora drafts one. Most of the time I keep the draft and make small edits.

Running a recurring task to triage all your comms: Work reaches me through too many channels to track manually. Claude's Cowork mode lets you set up a scheduled task that checks all your comms (for me that's Asana, Slack, Gmail, and Drive) and produces a summary of what's due and high priority. You can also ask Claude to draft responses for your approval.