I manage a large engineering team at Honeycomb. My calendar is usually 6-10 meetings a day, and the work that actually moves the needle, like coaching, strategic thinking, and unblocking people, has to fit in the gaps. I needed a system that would keep things from falling through the cracks without adding more process overhead.
So I built one! I point Claude Code at an Obsidian vault and use it as a personal operating system. It's become the most impactful productivity setup I've ever used, and I've tried a lot (too many) of them.
The Setup
Claude Code connects to tools I use day-to-day via MCP (Model Context Protocol): Todoist for tasks, Google Calendar (work + personal), Gmail, Slack, Granola for meeting notes, and Linear for engineering project tracking. Claude can pull context across all of them in a single workflow.
The real power is a detailed `CLAUDE.md` file, which includes instructions that tell Claude how I work, how my team is structured, what my task system looks like, and (the big one for me) when to proactively surface things I might be forgetting.
I've taught Claude to flag waiting items sitting 3+ days, approaching deadlines, personal tasks getting crowded out by work, and growth goals that haven't gotten attention. It does this automatically every morning, and makes sure anything that should be top-of-mind actually is.
What This Looks Like Day-to-Day
I've built 22 "skills", which are basically slash commands for recurring workflows. A few I use constantly:
Morning routine (~5 min): I type "morning routine" and Claude checks all my calendars, surfaces tasks and deadlines, and writes a strategic focus doc with my top 3 priorities for the day. It's not just a task list, it's an opinionated summary of what I should focus on and why.
Meeting prep: I say "prep for my 1:1 with [name]" and Claude pulls their recent Linear activity, our last 1:1 notes, relevant Slack threads, and drafts talking points. For sprint planning, it analyzes the backlog. For strategy syncs, it pulls recent decisions and open questions. Each meeting type gets different prep.
Meeting notes processing: After meetings, Claude pulls my notes from Granola via MCP, synthesizes them into structured notes (I prefer Claude's synthesis over the built-in Granola AI summary), and extracts action items. We walk through one meeting at a time so I can refresh myself on meeting content and catch anything it misses.
End of day (~5 min): Logs what happened, preps a preview for tomorrow, updates the week view.
Weekly planning (~15-20 min): Full sweep: inbox processing, reflection on last week, setting top 3 outcomes for the week ahead, rolling over anything that didn't get done.
The Capture -> Process -> Surface Loop
This is the underlying design pattern:
Capture is zero-friction. Siri, Todoist quick add, saving Slack messages, Granola recording meetings. I never have to stop what I'm doing to capture something properly.
Processing happens during routines with Claude. Triage inbox items, extract action items from meeting notes, file reference material. It's a distinct step, since stuff that sits half-processed is worse than stuff that hasn't been processed at all.
Surface is where it pays off. Strategic views in my vault (Today.md, This Week.md) answer "what should I focus on and why?" They're not mirrors of my task list, they're opinionated priority summaries that get refreshed automatically during routines.
Getting Started
If you want to try something like this, here's how I'd approach it.
Step 1: Pick one time-intensive workflow. For me it was morning planning. Every day I was manually checking my calendar, scanning Slack, opening Linear, trying to figure out what mattered most. That became my first skill. Pick whatever feels most repetitive for you, which might be something like meeting prep, inbox triage, or weekly planning.
Step 2: Set up Claude Code with a CLAUDE.md. This is the file that gives Claude context about you and your work. And it's a natural fit in an Obsidian vault! Start simple: your team structure, your tools, how you like to communicate. You don't need to nail it on the first pass. Mine has evolved significantly over time and I'm still tweaking it. (I actually have Claude tweak it on a regular basis based on our sessions.)
Step 3: Connect one or two MCP servers. Google Calendar and your task manager are a good starting pair. Just having Claude able to see your schedule and your task list unlocks a surprising amount. You can add Slack, Linear, email, etc. later as you find uses for them.
Step 4: Build your first skill. Write a skill file that describes the workflow step by step (hint: you can have Claude pair on this with you). What should Claude check? What questions should it ask you? What should the output look like? Run it a few times, notice what's missing or wrong, and iterate. The first version won't be great, and that's fine. I make updates to my skills on at least a weekly basis, which is less frequent than it was when I first got my system running.
Step 5: Let it grow organically. After a week or two with your first skill, you'll start noticing other patterns. "I do this same thing every Friday." "I always scramble for context before this meeting." Each of those is a candidate for a new skill. Add them one at a time.
The whole thing took me a few weeks to get to a point where it felt consistently useful, and it's been compounding since then. The key is not trying to build the finished system on day one.
What I've Learned From Building This
Start with one skill and grow. I started with just the morning routine and added skills as I noticed repeatable patterns. Don't try to build the whole system at once.
Keep work and personal obligations in the same system. All my priorities compete equally for limited time. Without this, work silently wins by default. I was months behind on personal tasks before I figured this out, and now my time feels much more balanced and productive across both spheres.
Make the system improve itself. Claude updates its own documentation: MCP syntax quirks, my preferences, what worked and didn't. Every week I ask it to extract anything that would improve the overall system, and implement it. The setup I have now is significantly better than what I started with.
You still need to closely review the output. Claude gets dates wrong and misinterprets things regularly. I always check the output before anything gets posted or sent. This isn't about handing over the wheel to AI entirely, but about giving you time back for the leverage that you're uniquely positioned to provide in your daily activities.
Why This Matters for Engineering Managers
EMs have a unique productivity challenge. Our work is fragmented across 1:1s, strategy sessions, sprint ceremonies, cross-functional syncs, and whatever fires need putting out. The actual leverage, like coaching someone through a hard problem, noticing a pattern across three different conversations, or making the right strategic call, requires context that's scattered across a dozen tools.
This system doesn't make me faster at answering emails (although that is a side benefit). It makes me better at the parts of management that matter most, because I'm walking into every conversation with the right context already surfaced.
If you're an EM (or aspiring EM) and want to build something like this, or just want to think through how AI can make you more effective as a leader, that's exactly the kind of thing I help with in my mentoring sessions. Happy to chat!
