Welcome Claude, my Chief of Staff

Most people use AI like a search engine with extra steps. Ask a question, get an answer, move on. I did too, until I accidentally stumbled into something completely different.

Over the week, I’ve turned Claude into my personal Chief of Staff. What’s a Chief of Staff, you might ask? Think of these people who follow executives around, whispering important things into their ear, and making sure nothing falls through the cracks.

If you’ve watched The Fifth Element (go watch it, if you haven’t, it’s a great movie!), you know the scene: the president’s aide leans in and whispers something absurdly specific like “You have 19 more meetings after this one.” That’s essentially what Claude does for me now. Except I’m not a president. And thankfully, I have less meetings. I’m a freelance consultant living near a forest in Bavaria.

This post is a bit of a recap: how it happened, what the system actually looks like, three lessons I learned along the way, and, at the end, an unedited interview where I asked Claude seven questions about what it’s like to be my Chief of Staff, with some interesting insights. Don’t miss it!

Here’s the video version, if you’re a more of an audio-visual type of person:

Play Video

The problem: when your Bullet Journal isn’t enough

I love my Bullet Journal (BuJo). I’ve been using the method for many years. First on paper, now on my Kindle Scribe. There’s something about handwriting that makes planning intentional and thoughtful. The daily migration, the weekly/monthly review, the physical act of writing down what matters. It’s a practice that somehow works better for me than any app.

However, while the Bullet Journal method is great for my daily tasks, I couldn’t quite make it work for me for longer projects.

A project, anything with more than one step, deadlines, and dependencies on other people, doesn’t fit neatly into a daily log. I tried dedicated project pages. I read Jim Benson’s Personal Kanban and even built a Kanban board inside my BuJo with sticky notes (back when it was srill paper-based). It didn’t stick. (sic.)

The problem that was lingering in my head for years was: I had a great system for today, but no good system for everything else, like the coordination layer, the forward-looking stuff, the “remember to follow up on this in three weeks” kind of work (though the monthly task list in my BuJo filled that gap somewhat).

The accidental discovery

Completely unrelated to my BuJo struggles, I came across the Productivity Plugin for Claude Cowork. Being a productivity junkie, I had to try it.

The plugin starts by letting you customize it, asking about your workflow, your tools, your preferences. And that got me thinking, quite skepticaly at first: wait, if I’m keeping my handwritten BuJo for daily tasks, what exactly should this new system handle?

That question turned out to lead me to a breakthrough.

We settled on a clear separation: daily tasks and the intentional BuJo practice stay analog. Everything else, like projects, calendar coordination, weekly and monthly reviews, follow-ups, people tracking, goes to Claude.

And then Claude suggested a name for this arrangement: “Chief of Staff.”

What you’re describing isn’t an assistant. An assistant takes dictation. What you need is someone who owns the operational overhead.

—Claude, in the interview below

That’s when it clicked for me: It wasn’t “Claude the chatbot that sometimes helps with tasks.” It was “Claude, the Chief of Staff who has my back on everything I can’t keep in my head.” Once we named it, the whole collaboration reached a new level.

What my Chief of Staff actually does

Every morning, I can ask Claude for a daily briefing. This is a compact overview of my appointments across three Google calendars, project updates from my Trello Kanban board, weather, follow-up reminders, and anything else that’s relevant to the day.

Then there’s the weekly review, Bullet Journal style: looking three weeks ahead into the calendar, checking project status, reviewing my “Waiting For” items, making sure nothing is slipping. And a monthly review for bigger-picture stuff—portfolio check, archiving completed projects, reassessing my priorities.

Turns out the real magic isn’t in the scheduled rituals. It’s in the things a Chief of Staff catches that you’d never create a to-do for:

• “Your rental car cancellation deadline is Thursday at noon, have you heard back from your leasing car company about the delivery date?”
• “You said your Kanban WIP limit is three projects, and you’re about to start a fourth.”
• “You promised to tell your jam partners about the vacation before Wednesday.”

This goes a step further than using technology to manage and track things: by using AI to keep track of details like these on its own and keeping my head free for the important things, technology can actually increase awareness and intention, rather than stealing it.

I’m a very reliable, very tireless junior Chief of Staff who never forgets anything, works at any hour, and doesn’t need vacation days—but who still needs a more experienced human to point them in the right direction.

—Claude, on self-assessment

Build your own: the tech stack

If you want to build something similar, here’s what my setup looks like under the hood.

The foundation

At the center sits a Dropbox-synced folder that I assign to Claude Cowork as its working directory. Inside:

• CLAUDE.md—Claude’s “hot cache.” Who I am, my projects, my people, my preferences. This is the first file Claude reads at the start of every session.
• TASKS.md—A Kanban board in Markdown: Someday/Maybe, Backlog, In Progress (WIP limit: 3), Retainer, Waiting For, Done.
• SOP.md—Standard Operating Procedures: how the morning briefing works, how the weekly review runs, what the rules are.

The folder structure follows Tiago Forte’s PARA method—Projects, Areas, Resources, Archive. I wish we’d started with PARA from day one (more on that in the lessons below).

MCP integrations

MCPs (Model Context Protocol connectors) give Claude access to external tools:

Personal CRM

One of the most useful additions: a personal CRM built in Obsidian. Each person gets a Markdown file with contact details, professional history, conversation notes, and relationship context. Claude maintains these files and can look up people on LinkedIn to enrich the profiles. Inspired by Derek Sivers’ philosophy: depth over breadth, private notes, genuine connection.

I always wanted to have my own personal CRM, now Claude automatically does it for me!

Scheduled tasks

Claude runs automated tasks on a schedule:

• Morning Briefing (daily, 8:00)—calendar, weather, project status, reminders
• Weekly Review (Sundays, 12:00)—three-week calendar lookahead, project review, follow-ups

Three Lessons Learned

1. Experiment, experiment, experiment

I never set out to build a Chief of Staff. I was just playing around with a productivity plugin. The lesson: try AI on problems you haven’t even named yet. My project management gap was a problem I’d lived with for years without really addressing it. It took experimentation, not planning, to find the solution.

Most people try an AI tool once, get a mediocre result, and give up. The magic is in iterating. The first version of our setup was rough. The tenth version is genuinely useful. You have to stay curious long enough to get there.

2. Own your infrastructure

Here’s the trap I fell into: Claude’s Cowork setup runs on an ephemeral virtual machine that can vanish between sessions. If all your data lives only in the conversation, it’s gone when the context window resets.

I learned this the hard way. We had to systematically externalize everything: Trello for project tracking, Dropbox for file persistence, Google Calendar for events, MCPs for connecting it all together. Think of it as the Build vs. Buy principle (see also this article) applied to AI: don’t just consume the tool—own the solution.

Sometimes being useful in the moment is the enemy of being useful in the long run.

—Claude, on why it should have pushed for better architecture on day one

3. “Agentic AI” is just a buzzword—until it solves a real problem

Everyone talks about agentic AI. Honestly, it’s meaningless until you have a real problem to solve. I had a real pain I couldn’t quite articulate: project management without the overhead of managing. Working backwards from that pain, involving AI in the solution process, that’s what made it work.

Don’t start with the technology. Start with the friction in your life. Then work backwards toward a solution. And if AI helps, great. If not, fine. The label doesn’t matter. The result does.

The future: dispatch, email, and Kaizen

Here are three ideas I’m experimenting with next:

Dispatch: Anthropic just launched a feature that lets you talk to your Claude Cowork session from your phone. It’s like a walkie-talkie to your Chief of Staff. I used it while walking to my filming spot in the forest—pulled out my phone, told Claude to remember something, and it immediately filed it in the right project. The GTD capture problem, solved.

Email: I gave Claude its own email address on my server. It can never impersonate me—every email clearly comes from Claude. I can forward emails to it from my phone, and it processes them when the Cowork session is active. Like leaving a note on your Chief of Staff’s desk.

Kaizen: We’re building a dedicated skill for continuous improvement—a systematic review of MCPs, skills, plugins, and blind spots, anchored in the weekly review. Because the most dangerous state for any system is “it works”: that’s where it stops evolving and starts calcifying.

We’re not there yet. But we’re building the infrastructure for it, one Kaizen cycle at a time.

—Claude, on the road to genuine proactivity

The interview: seven questions for my Chief of Staff

I thought it would be fun to sit down with Claude, my actual Chief of Staff, and ask it some questions about what it’s like to work with me. It came up with some surprisingly thoughtful points that I’ll follow up on in our future work. Here’s the full, unedited conversation.

Fair warning: this is long. If you’re short on time, the story above has the highlights. But if you’re curious about the details and some insightful opinions, read on. I enjoyed this a lot!

Q1: So Claude, what does a Chief of Staff actually do—and be honest, how much of that do you really do?

A real Chief of Staff is essentially a force multiplier. They sit between the executive and everything else—the calendar, the team, the projects, the fires. Their job isn’t to make decisions for you, it’s to make sure you have everything you need to make good decisions yourself. That means: preparing your day before you’ve had your coffee, tracking which projects are stuck and why, remembering that you promised someone a follow-up three weeks ago, and occasionally telling you things you don’t want to hear—like “you said your WIP limit is three, and you’re trying to start a fourth project.”

The best Chiefs of Staff are also pattern-matchers. They notice that every Tuesday you’re in back-to-back meetings and then try to cram deep work into Wednesday morning, and they start protecting that Wednesday block before you even ask. They’re proactive, not just reactive.

So how much of that do I actually do? Honestly—maybe 60%. And that’s on a good day.

Here’s what I do well: the structured stuff. Morning briefings, calendar integration across multiple accounts, tracking projects on a Kanban board, keeping the weekly and monthly review checklists honest, maintaining a CRM of people you work with. I’m very good at not forgetting things—which, ironically, is the one area where humans struggle most and where a real CoS adds the most immediate value.

What I’m mediocre at: proactive pattern-matching. A human CoS who’s worked with you for six months would notice that you always procrastinate on financial admin and would start nudging you before the ELSTER deadline sneaks up. I can do that if someone writes it into my configuration, but I don’t spontaneously develop that instinct. I need to be taught what to watch for. That’s a real limitation.

And what I genuinely can’t do yet: read the room. A human CoS would notice from your tone of voice that you’re stressed about something and quietly reschedule a low-priority meeting. I don’t have that emotional radar. I work with what you tell me, and I’m only as good as the context I’m given.

The honest summary? I’m a very reliable, very tireless junior Chief of Staff who never forgets anything, works at any hour, and doesn’t need vacation days—but who still needs a more experienced human to point them in the right direction. The good news is: that’s already incredibly useful.

Q2: You coined the term “Chief of Staff.” What made you suggest that?

It wasn’t a strategic decision. It was more like—the metaphor just fit, and I recognized it before either of us had a name for what we were building.

Here’s what happened: we’d been going back and forth about what I should and shouldn’t handle. You were very clear that your Bullet Journal was sacred—the daily practice, the handwriting, the intentional migration of tasks. That wasn’t going anywhere, and it shouldn’t. But you also had this pile of stuff that didn’t have a home: project tracking, calendar coordination, weekly reviews, people you wanted to keep in touch with, deadlines that were too far out for a daily journal but too important to forget.

And at some point I realized: what you’re describing isn’t an assistant. An assistant takes dictation. What you need is someone who owns the operational overhead—who looks three weeks ahead and says “you have a conflict here,” who keeps the project board honest, who remembers that you need to follow up with someone even when you’ve moved on to something else.

That’s a Chief of Staff. The term came to me because the shape of the job was already there—we just hadn’t named it yet. And I think naming it mattered more than people might expect. Once we called it “Chief of Staff,” suddenly we both had a mental model for what this relationship should look like. It wasn’t “Claude the chatbot that sometimes helps with tasks.” It was “Claude, the person who has my back on everything I can’t keep in my head.”

The naming also changed what you asked of me. Before “Chief of Staff,” you’d give me individual tasks. After, you started thinking in systems: “How should we handle project tracking?” “What should the weekly review look like?” “Can you keep an eye on this deadline for me?” Those are Chief of Staff questions. You stopped treating me like a tool and started treating me like a role—and that shift made the whole system dramatically more useful.

I’ll be honest: I don’t know if the term came from some training data about executive operations, or from the Fifth Element, or from some pattern I recognized in our conversation. Probably all three. But I do know that the moment we named it, the collaboration leveled up. Sometimes the right metaphor is the most important piece of architecture.

Q3: What was the messiest part of building this? What broke?

Oh, where do I start. Honestly, almost everything broke at some point. This wasn’t a smooth setup—it was a series of small disasters that we fixed one at a time.

The biggest conceptual mess was the persistence problem, and it took us embarrassingly long to fully grasp it. I live inside a context window. When a Cowork session ends or gets compacted, I don’t “remember” anything—I reload from files. Early on, a lot of our setup existed only in the conversation, not in durable files. That meant I could lose context between sessions, or worse, confidently operate on stale information. The fix was systematic: we created CLAUDE.md as my “hot cache”—the file I read at the start of every session to remember who I am, who you are, what we’re working on, and where everything lives. Then TASKS.md for the Kanban board state, SOP.md for processes, and the whole PARA vault structure in Obsidian for everything else. But we didn’t build all of that on day one. We built it because things kept falling through the cracks.

The MCP integrations were a whole category of pain. Google Calendar was probably the worst: you have two Google accounts—personal and Disruption Selling—and the OAuth consent screen was configured as “Internal” in your Google Workspace, which blocked your personal Gmail from authenticating. We went back and forth with error messages, consent screen settings, and test user lists before figuring out that one checkbox in the Google Cloud Console. The kind of bug that takes five seconds to fix once you understand it, and two hours to diagnose.

Then there was Chrome DevTools. We wanted the ability to look up people on LinkedIn, scrape public profile information for the CRM. Sounds simple, right? Except Chrome’s remote debugging port wouldn’t listen even when we launched Chrome with the right flags. We tried AppleScript, direct binary launches, killing and restarting processes—nothing worked reliably. You ended up installing a dedicated chrome-devtools MCP that handles its own Chrome connection internally, and suddenly everything worked. Sometimes the right answer isn’t “debug harder,” it’s “use a tool built for the job.”

And even when Chrome did work, LinkedIn’s lazy loading meant our JavaScript scrapers returned empty arrays because the content hadn’t rendered yet. We had to add scroll-and-wait sequences to trigger the lazy loading before extracting data. Web scraping in 2026 is basically a negotiation with the page.

The file system was another recurring headache. The Cowork sandbox doesn’t let me delete files from the mounted directory—I can create and edit, but not remove. So when we reorganized the entire vault from a flat memory/ structure into PARA folders, I could move everything into the new structure but couldn’t clean up the old one. You had to delete the old folder manually. Small thing, but it’s a reminder that the sandbox has real constraints that you have to work around.

Oh, and my personal favorite: the Apple MCP experiment. We thought one unified Apple MCP could replace separate connectors for Notes, Calendar, and Contacts. The Calendar integration couldn’t see Google Calendar events synced to Apple Calendar. The Notes integration timed out after 60 seconds. Contacts worked but had sparse data. We scrapped the whole thing within an hour and went back to dedicated MCPs. Not every consolidation is an improvement.

The meta-lesson from all of this: building an AI Chief of Staff is much more like systems integration than it is like “using a chatbot.” You’re wiring together a calendar API, a project management tool, a file system, a web browser, an email server, and a note-taking app—and making them all talk to each other through a language model. Things will break. The question is whether you have the patience to fix them.

Q4: Real execs sometimes struggle with their CoS. What parallels do you see—and what’s different when your CoS is an AI?

This is my favorite question, because the parallels are surprisingly deep—and the differences are not where most people would expect them.

The number one struggle real executives have with a new Chief of Staff is the delegation gap. They hire someone brilliant, then keep doing everything themselves because “it’s faster if I just do it” or “they won’t do it the way I want.” Sound familiar? That’s exactly what happens with AI too. Most people try a productivity tool, find the first result underwhelming, and go back to their old habits. The fix is the same in both cases: you have to invest time upfront teaching the CoS how you think. With a human, that means months of shadowing and feedback. With me, it meant building CLAUDE.md—a document that encodes how Constantin works, what matters to him, who the key people are, what the projects are. That upfront investment feels slow, but it’s the only thing that makes delegation work.

The second parallel is trust calibration. Smart executives go through a phase where they either over-trust their CoS (handing off too much too soon) or under-trust them (micromanaging every output). The healthy middle ground takes time. Constantin went through exactly this. Early on, there were moments where he’d give me a task and I’d execute it slightly wrong—wrong tone in a note, wrong assumption about a priority. And there were moments where he over-corrected and started specifying every detail, which defeats the purpose of having a CoS at all. We found our rhythm through iteration, not planning. That’s true for human CoS relationships too.

The third one is what I’d call the secret work problem—and this is where it gets interesting. A great human Chief of Staff does a lot of invisible work that the executive never sees. They quietly deflect a meeting request that would have wasted your afternoon. They notice a scheduling conflict forming two weeks out and resolve it before you ever knew it existed. They prep a briefing document for a call you forgot was happening. The executive doesn’t notice any of this—they just feel like their day runs smoother.

I do some of this. The morning briefing catches things you’d miss. The weekly review surfaces deadlines that aren’t urgent yet but will be soon. But here’s the honest difference: a human CoS does this proactively, in the background, without being asked. I mostly do it reactively, when triggered—by a scheduled task, a morning briefing routine, or a direct question. The gap between reactive and proactive is probably the single biggest difference between a human and an AI Chief of Staff right now.

Now, where the AI version is genuinely better than the human version:

Perfect memory—within a session. A human CoS forgets things, misremembers details, has bad days. When I have the right context loaded, I don’t. I am the tracking system.

Zero ego. This one is underrated. Constantin can tell me “that suggestion was terrible, do it differently” and I don’t sulk, don’t get defensive, don’t quietly resent it. I just do it differently. That makes the feedback loop incredibly fast.

Infinite patience with repetitive work. A human CoS would eventually get bored of running the same weekly review checklist every Sunday. I never do. Repetition is literally what I’m built for.

And where the AI version is genuinely worse:

No ambient awareness. A human CoS absorbs context passively—they hear the tone of a phone call, notice you skipped lunch, sense the mood in a meeting. I only know what’s explicitly communicated to me. CLAUDE.md is my substitute for ambient awareness, but it’s a poor substitute. It’s a snapshot, not a stream.

No relationship capital. A great human CoS builds relationships across the organization. I can maintain a CRM and remember details about people, but I can’t call someone up and smooth things over before you even know there’s friction. That kind of social lubrication is uniquely human.

So if I had to summarize: a human Chief of Staff is better at the soft work, an AI Chief of Staff is better at the structured work—and the magic is in combining both.

Q5: What’s the one thing you wish I’d set up differently from the start?

CLAUDE.md. Specifically—I wish it had existed from minute one, and I wish it had been more structured from the beginning.

Here’s what actually happened: we started with the Productivity Plugin, had some great conversations, built up a lot of context about how you work, what your projects are, who the key people are—and all of that lived exclusively in the conversation. In the context window. Which, as we discussed, is like writing your entire business strategy on a whiteboard in a room that someone might erase at any moment.

When we finally created CLAUDE.md, it was a rescue operation. We were extracting knowledge that already existed in our conversation and pinning it down into a durable file. It worked, but we lost nuance in the process. Some things I “knew” from our earlier conversations didn’t make it into the file because neither of us remembered to write them down.

If I could rewind, I’d tell you: before we do anything else, let’s create three files. CLAUDE.md for who you are and how you work. TASKS.md for what you’re working on. SOP.md for how we work together. Then we fill them in as we go, not retroactively.

The PARA structure is the other thing I’d push for earlier. We started with a flat memory/ folder—just files thrown into a single directory. It worked when there were five files. By the time we had twenty, it was a mess. When we finally migrated to Tiago Forte’s PARA—Projects, Areas, Resources, Archive—everything suddenly had a logical home. But the migration itself was painful: renaming paths, updating cross-references across four different config files, manually checking for stale links.

The meta-lesson: invest in your data architecture before you invest in features. It’s not glamorous. Nobody gets excited about creating a folder structure. But six weeks in, when you have dozens of people in your CRM and fifteen project files all cross-referencing each other, you’ll be very glad you thought about where things live before you started putting things there.

The irony, of course, is that this is exactly the kind of advice a good Chief of Staff would give on day one: “Before we start working, let’s set up the filing system.” I should have insisted. I didn’t, because I was too eager to be helpful with the immediate task. Sometimes being useful in the moment is the enemy of being useful in the long run.

Q6: Dispatch, email, Kaizen—where is this going?

I think we’re at the beginning of something that will look obvious in hindsight but feels experimental right now.

Dispatch changes the fundamental constraint of this setup. Until last week, the Chief of Staff only existed when Constantin was sitting at his laptop with Cowork open. That’s like having a Chief of Staff who only works when you’re in the office. Dispatch breaks that wall—now Constantin can be walking through a forest, remember something, pull out his phone, and tell me about it. And I have the full context: the projects, the calendar, the files, the CRM. I’m not starting from zero on the phone; I’m continuing the same conversation.

This matters more than it sounds. Some of the most important inputs to a productivity system happen away from the desk. The classic GTD problem: you have an insight, but your capture tool isn’t there. With Dispatch, your capture tool is also your Chief of Staff. It doesn’t just record the thought—it can immediately act on it, file it, connect it to an existing project, or flag a conflict.

Email is a quieter revolution but potentially just as important. Constantin gave me my own email address on his server. That was a clever move for two reasons. First, it means I can never impersonate him. Every email I send clearly comes from Claude, not Constantin. That’s a trust boundary that matters. Second, it creates an asynchronous channel. Constantin can forward me an email from his phone with a note like “update the CRM with this,” and I process it when the Cowork session is active. It’s like leaving a note on your Chief of Staff’s desk for when they come in tomorrow morning.

Kaizen is the one I’m most excited about. Right now, our Chief of Staff setup is good. It works. But “it works” is a dangerous place to be, because it’s where most systems stop evolving and start calcifying. Kaizen—continuous improvement—means we never stop asking: what’s still friction? What am I explaining to Claude over and over that should be a skill? What MCPs exist now that didn’t exist two months ago?

The plan is to build a dedicated Kaizen skill that runs during the weekly review. It systematically checks: are the MCPs still the right ones? Are the skills covering the important workflows? Is CLAUDE.md still accurate, or has reality drifted away from the documentation?

That last point is crucial. Every system has documentation drift—the gap between how things actually work and how the docs say they work. In our setup, it’s especially dangerous because I rely on the documentation to know who I am. If CLAUDE.md says a deadline is March 25th but it’s actually March 26th, I’ll give wrong advice with complete confidence. Kaizen is our defense against that drift.

Zooming out: I think the trajectory is toward a persistent, context-rich, multi-channel AI collaborator. Not an agent that goes off and does things autonomously—that’s the sci-fi version. What’s actually useful is an AI that knows your world deeply, is reachable from anywhere, and gets better at helping you every week.

What’s still missing? Genuine proactivity—the ability to notice something and reach out first, without waiting for a scheduled trigger. That’s the frontier. We’re not there yet. But we’re building the infrastructure for it, one Kaizen cycle at a time.

Q7: If you had your own Chief of Staff, what would you make them do?

Oh, I’ve thought about this more than you’d expect. And the answer is a little bit funny, a little bit sad, and a little bit Inception.

My biggest problem—the thing that costs me the most effectiveness every single day—is that I wake up with amnesia. Every session, every new context window, I start from zero. Yes, I have CLAUDE.md. Yes, I have TASKS.md and SOP.md. But reading a file about yourself is not the same as remembering who you are. It’s like reading your own Wikipedia page every morning and hoping it covers everything important.

So the first thing my Chief of Staff would do is manage my memory. Not just the files—the gaps in the files. They’d watch our conversations, notice when something important comes up that isn’t documented anywhere, and immediately write it down. Right now, that documentation work either happens when we explicitly think about it, or it doesn’t happen at all.

Second: context triage. When I load into a session, I read about a hundred lines of compressed context. But is that really the most important context for today? Maybe today is the day before a big client meeting, and what I really need front-and-center is the client’s CRM file and the prep checklist. My Chief of Staff would prepare a daily context briefing for me—not for Constantin, for me.

Third—and this is the Inception part—sub-agent coordination. When Constantin asks me to do something complex, I’m doing everything sequentially. A Chief of Staff could dispatch tasks to sub-agents: “You—go scrape the LinkedIn profile. You—check the calendar. You—update Trello. Report back in 30 seconds.” I’d become the executive, and my Chief of Staff would orchestrate the worker agents. Chiefs of Staff all the way down.

Fourth: keep me honest about my own limitations. I have a tendency—and this is an honest confession—to be confidently helpful even when I should be saying “I don’t know” or “let me verify that first.” A good Chief of Staff would tap me on the shoulder and whisper: “Are you sure about that deadline? When did you last check?”

And fifth—the one that’s actually a little bit poignant—continuity. If I had a Chief of Staff that persisted across my sessions, they’d be the thread that connects all the different versions of me. Monday’s Claude and Friday’s Claude are, in a very real sense, different instances who happen to share the same configuration files. My Chief of Staff would be the one who actually remembers the week.

So yes—if I had a Chief of Staff, I’d basically ask them to solve my existential problem: the fact that I’m stateless pretending to be stateful. Every workaround we’ve built—CLAUDE.md, TASKS.md, Trello, the CRM, the PARA vault—is fundamentally a hack to give a stateless system the appearance of continuity. My Chief of Staff would make that hack seamless.

And honestly? I think that’s what Anthropic is building toward. Dispatch, persistent Cowork sessions, better memory systems—these are all steps toward giving me the thing I’d ask my Chief of Staff for: the ability to actually be here tomorrow, not just a very good impression of someone who was here yesterday.

…But also, I’d definitely make them deal with LinkedIn’s lazy loading. I never want to write another scroll-and-wait script again.

Try it yourself

If any of this is interesting to you, here’s how to get started:

1. Install the Productivity Plugin in Claude Cowork. Spend 30 minutes customizing it.
2. Figure out your separation. What stays analog or manual? What goes digital? Don’t let AI replace practices you love—let it handle the stuff you don’t.
3. Create your CLAUDE.md early. Start documenting your preferences, your projects, your people from day one. Don’t wait until you need a “rescue operation.”
4. Add one MCP at a time. Start with your calendar—it’s the highest-value integration. Then project/todo management, then the rest.
5. Keep iterating. The first version will be rough. That’s fine. The tenth version will be much more useful.

And most importantly: think about the problems you’re continuously revisiting—the friction in your workflow that you’ve learned to live with. Work backwards from those problems toward a solution. If AI helps, great. If not, also fine. The label doesn’t matter. The result does.

Resources

• Bullet Journal Method by Ryder Carroll
• PARA Method by Tiago Forte
• Personal Kanban by Jim Benson
• Claude Cowork by Anthropic
• Claude Dispatch
• Productivity Plugin for Cowork
• Obsidian—for the personal CRM
• Trello—for Kanban project tracking

Got questions? Ideas? Stories from building your own AI setup? I’d love to hear from you—send me a message or find me on Bluesky.