Large language models (LLMs) have generated a lot of excitement—some of it justified, some of it over the top. The demos are flashy: code generation, real-time language translation, pitch decks spun up in seconds. It all feels impressive, but also a bit disconnected from what managers actually do day to day.
When you’re deep in a week of 1:1s, feedback docs, and tough interpersonal conversations, you’re not looking to automate your job—you’re looking for leverage. A way to reduce friction. A way to write the thing you know you need to say but haven’t found the words for yet.
So I’ve been experimenting: where can LLMs quietly support the rhythm of a regular manager’s week? Here are a few patterns that don’t feel like novelty—they feel useful.
1. Draft the first version, not the final one
I use LLMs as a "rough draft" engine. For example:
- Writing a performance review summary based on bullet points from a rubric.
- Generating a first pass of a promotion nomination after I dump a bunch of highlights into the chat.
- Drafting a follow-up email after a skip-level that surfaced some tricky feedback.
The trick here is: you don’t ask the LLM to think for you. You think, it writes. You shape. It saves you from the blank page, not the hard decisions.
For general-purpose drafting, I’ve found ChatGPT 4o to be fast and reliable.
2. Triage and summarize context
One underrated use: summarizing meeting transcripts, PRDs, or large documents. I’ve had good results with NotebookLM when feeding it transcripts or long-form content and asking:
"Give me a summary of this conversation, highlight risks, and extract any action items assigned to me or my team."
It’s not perfect, but it’s close enough to speed up my review loop. I’ll usually edit or validate key parts, but it's often 80% of the way there.
Bonus: summarizing PRs and technical design docs before asking follow-up questions. It's not just for managers—senior ICs love this too.
3. Unblock writing by reflecting back your intent
Sometimes, I find it hard to articulate why a reorg feels off or why a strategy doc isn't landing. In those moments, I’ll write a messy stream-of-consciousness paragraph about what’s bugging me, and then ask the LLM:
"Can you reflect this back more clearly?" or "What’s the core tension in this paragraph?"
It doesn’t always get it right, but it’s like pair writing with someone who’s never tired and doesn’t judge your sloppy first take. ChatGPT 4o is particularly good at this reflective style of writing support.
4. Make meetings and updates tighter
Every manager has a few “repeat” templates:
- Weekly team updates
- Project status reports
- Incident write-ups
Instead of starting from scratch each time, I’ve started templating these out and using an LLM to auto-fill them based on raw notes or Asana tickets. It’s especially useful when you’ve written 90% of the content in Slack, just scattered across five threads.
It’s not about automating the update, it’s about accelerating the packaging.
5. Simulate uncomfortable conversations
This one surprised me.
I’ve started “practicing” difficult conversations with an LLM:
- A direct report who’s not getting the message.
- A peer who keeps dropping work on your team.
- A project stakeholder who’s visibly disengaged.
You write what you want to say, and ask the model to respond as that person might. Then iterate until the words land better.
Is it weird? Yes. Is it useful? Also yes.
6. Build internal tools faster (especially for metrics)
If you’re a manager who still codes, LLMs can be a boost for those glue tools you always meant to write:
- A script to analyze team velocity from GitHub PRs.
- A dashboard for on-call incidents.
- A tool that tracks feedback themes from your 1:1 notes.
I’ve had particularly good luck with Claude Sonnet 3.5 for this—it handles large inputs and brainstorming sessions for internal tools with clarity and helpfulness. Sonnet 3.7 tries to do a bit too much.
The bar for useful internal tooling just dropped—if you’re willing to tolerate a bit of mess in the output.
7. Document your thinking as you go
Finally, the quietest and most valuable use for me: writing down what I think, when I think it, in a structured way.
I have a habit now where I log:
- What decisions I’m considering this week
- What I’m worried about
- Who I think is doing well, and why
Then I run it through an LLM to generate structured notes or shareable blurbs (for performance reviews, check-ins, or team-wide context).
It’s a journal with better formatting. ChatGPT 4o and Claude are both great for this kind of personal synthesis.
Closing Thoughts
LLMs won’t replace your judgment, your context, or your relationships. But they can be a useful part of your toolkit—especially if you treat them like an over-eager chief of staff who writes fast, doesn’t mind rewriting, and never complains about the meeting that should’ve been an email.
Just don’t tell them your skip-levels are using them too.