How I Built an AI Chief of Staff for My Solo Business (So the Plan Stops Living in My Head)

The 4 layers, the planning chain, and the 4 commands that run my week: how an AI chief of staff holds my priorities so I don’t have to.
The short version
- More AI output doesn’t move your business forward. Deciding what deserves the output does.
- An AI chief of staff holds your priorities and flags drift. Most AI assistant demos just clear your inbox.
- The system is 4 layers: context, maintenance, commands, growth. The first 2 do the real work.
- Every task traces 1 straight line: today, this week, the sprint, the quarter, the revenue plan, the 3-year vision.
- Start with 3 files: what you said matters, what actually happened, and today.
In this post
- What is an AI chief of staff?
- What are the 4 layers under an AI chief of staff?
- How does the context layer answer “why am I doing this today?”
- What keeps the plan current when the week moves?
- What are the 4 commands that run the week?
- What should an AI chief of staff be allowed to touch?
- Can you make your chief of staff a full agent?
- How do you start with the smallest version?
- Your next step
- FAQ
A year ago most of us used AI the same way.
You opened a chat, you asked a question, you got an answer, and then you carried that answer into the rest of your work yourself.
Now AI executes. It drafts the caption. It pulls the keyword research. It builds the thing
So you get to the end of a week with more output than you’ve ever had. And it feels productive!
But here’s the question nobody asks: did any of it actually move your priorities or your goals?
Every tool is brilliant at the task in front of it.
But you’re still the one deciding what to work on….Still the one remembering how it all connects.
You are still the glue.
I felt the same way. I was tired of being the only place all of it lived… the ideas, the priorities, the wait, did I already decide that?
That’s why I decided to create an AI Chief of Staff Agent.
But not a typical agent that checks your calendar or emails.
This is an agent that holds your business context: your priorities, your plans, and a log of what actually happened.
It proposes what to work on, keeps those files current, and flags when your week drifts.
This post walks you through how I set this up.
I’ll also run through how this planning is centred on a cascade of documents from a 3-year vision down to a note for what actually happened today, and the agent that runs the chain so I’m not the one connecting it by hand.
What is an AI chief of staff?
There are tons of chief of staff demo videos, and most do the same few things: clear your inbox, manage your calendar, prep your meetings. Useful, sure.
But what I felt was missing is the harder part: something that holds your priorities and goals and tells you when you’ve drifted from them.
Because AI collapsed the doing.
Drafting, research, building: the execution layer got fast.
What didn’t change is that someone still has to hold what matters this quarter, notice that the work you shipped this week serves none of it, and say so. In a solo business, that someone is you.
And that’s the trap of a productive week.
It produces a lot but how much of it actually moves the needle is questionable.
The fix isn’t one more connected tool, either.
Plenty of tools can see your calendar and still not know which meeting actually needs your attention. Access alone doesn’t fix the scattered-context problem.
That’s where Piper, my Chief of Staff comes in,
She’s what I call an orchestrator agent: she sits above a set of specialists, each one built for a narrow job.
3 tiers, 1 chief of staff at the top, and a model matched to each one.
On the rules side, she inherits the same operating rules every agent in my setup reads (the root file I broke down in my Claude Cowork setup post), plus her own job description on top: strategic co-pilot first, router second.
She also has a small set of hard boundaries, and a routing map, so she knows exactly who owns which part of the business.
But the title “Chief of Staff” means nothing on its own. What makes her work is what sits underneath.
What are the 4 layers under an AI chief of staff?
4 layers, and the first 2 do the real work.
1. A context layer. My North Star: the priorities, what actually matters to me. A vision, a revenue plan, the current quarter.
2. A maintenance layer. The living documents: the sprint I’m working, the plan for this week, today’s log. These update constantly, and Piper maintains them.
3. A commands layer. The commands I actually run: plan my week, what’s today, wrap up, review my week. I say it, she runs it.
4. A growth layer. What happens without me asking at all, and which tools she actually has access to.
And the order matters.
It’s the first 2 layers that do the real work: the context, and what gets maintained on top of it. That’s where the rest of this post spends its time.
How does the context layer answer “why am I doing this today?”
The context layer is the piece most people skip, and it’s the one that decides whether the system gives you something specific or something generic.
Mine isn’t 1 file. It’s a chain, each link pointing straight at the one above it.
At the top: a 3-year vision, what I’m building toward.
It points down to a revenue plan.
The revenue plan points down to the current quarter.
I work in 90 days, so the quarter holds a small set of standing priorities, decided ahead of time. (If you’ve never sat down and decided those, start with the 7 questions I ask before planning a year.)
It’s also why a year is the wrong anchor for any of this. An annual plan gets written once, in January optimism, and quietly stops being true by April.
A 90-day cycle gets re-decided 4 times a year against what actually happened.
One of those priorities points down to a sprint, when I choose to run one.
I’m a huge fan of sprints. These are tightly scoped projects small enough to actually finish by a specific date.
For example, your first Amazon book in 14 days. Or your next offer in 7.
Otherwise, the priorities point down to this week’s plan: 3 to 5 priorities, and my top priority this week traces straight back to that sprint or that quarter. Not floating on its own.
If the structure sounds familiar, it’s the same method my CREATE Intensive Biz Planner walks through.
My actual task today traces in 1 straight line: today’s task, this week’s priority, the sprint, this quarter’s standing priority, the revenue plan, the 3-year vision.
Every link in the chain has an answer for “why am I doing this today.”
Ask me why am I doing this today and every single link in that chain has an answer.
Nothing’s out of place. Everything ties back to a plan.
What keeps the plan current when the week moves?
The sprint I’m working, the plan for this week, today’s log… those aren’t static files sitting there.
Piper writes them, updates them every time we talk, based on what actually happened. So a new session starts by reading where things stand, instead of asking me.
That’s the actual payoff of keeping these documents current. I’m not carrying the state of my business in my head between sessions anymore. She is, on paper, where I can actually see it.
And I’ve layered 1 habit of my own on top.
When a session ends mid-thought, or when I need to compact, I ask for a handoff: a note that records what’s decided and shouldn’t get reopened, what’s still open, and what to do next.
The next session reads that note before anything else. That’s what makes Claude remember me, session after session. I’ve tweaked it multiple times such that handoff is now a specific skill I call.
What are the 4 commands that run the week?
4 commands, 1 loop: each day ends by leaving a note to itself.
“Plan my week.” I run this on Friday, before the week ahead starts.
Piper reads where the business actually stands, checks last week’s carry-overs, and flags the plan if something’s drifted.
Then she proposes 3 to 5 priorities, each one required to trace back to something I actually decided this quarter. If a whole area of the business has no active sprint, she says that plainly instead of inventing a priority to fill the space.
She drafts the note, then asks 1 question: “What am I missing?” That’s the step that matters. It invites me to push back. She’s doing the cross-referencing I used to do by hand across 4 different places.
“What’s today.” Every morning, she pulls up today’s commitments, plus a scan of the running log for anything tagged #revisit… the things I said I’d come back to.
If something’s overdue and nobody’s looked at it, it resurfaces on its own.
I don’t have to remember it. And when the same priority has been named 3 mornings running without moving, the morning check says so out loud: shrink it, protect a block for it, move the date honestly, or drop it.
That 1 rule is where this beats every planner I’ve owned.
A planner takes ‘not now’ for an answer every single time. This system doesn’t.
“Wrap up.” Every evening, 3 things.
She writes today’s log: what got done, what I struggled with, anything worth noting.
She checks priority health against the week’s plan: what’s advancing, what’s quietly falling behind.
And she closes the entry by naming tomorrow’s priority, right there in the same log.
The next morning, “what’s today” reads that line before anything else, so tomorrow starts warm instead of cold. Each day ends by leaving a note to itself, and the next one starts by reading it. That’s the loop.
“Review my week.” When I want a fuller look back, this command reads the whole week’s logs against the plan: what got completed, what got skipped and why.
It looks for a pattern, not a tally.
And that review is exactly what “plan my week” reads first, next Friday. Zoomed out, it’s the same loop: the week ends by leaving a note to itself too.
What should an AI chief of staff be allowed to touch?
2 things people ask about constantly once they see this working: how Piper is actually built, and what I let her touch.
For the build, think org chart.
Your Chief of Staff is the role making cross-business judgment calls, so it gets the most capable model you’ve got.
The specialists beneath can run on the next best. The narrow, high-volume jobs get the cheapest.
| Role | Model tier | Why |
|---|---|---|
| Chief of staff | The most capable you’ve got | Cross-business judgment calls |
| Specialists (the desks) | The next best | Drafting, editing, narrow expert work |
| High-volume checks | The cheapest | Pulling keyword data, checking 100 rows against a rule |
And I want to be honest about why, because it wasn’t just “give the smartest thing the top job.”
I didn’t want to spend top-tier tokens on jobs a cheaper model handles fine.
Some models draft better, some are stronger elsewhere; it takes a little testing.
I deliberately started this system small, because what I wanted was to get rid of the scatteredness and keep to priorities.
Once you’re comfortable, you layer on more responsibilities: checking the calendar and blocking real deep-work time around what your agent already knows matters that week, reminders on Slack, a morning digest, filling up your Notion boards.
But whatever tool you connect, think about scopes and boundaries first.
This is where a Chief of Staff parts ways with the typical AI personal assistant for business: the question is never just what it can do. It’s what it should be allowed to do without asking you first, every time.
The first thing you’d think to do is give it access to everything, so it can just make your life easier, right?
What you actually want is the narrowest slice of access it needs to do the job. Start there. Expand only as you actually need to, not upfront.
Which of your AI roles is running your priciest model on work that never needed it? That’s what the free AI team org chart helps you with. In it, you’ll discover:
- The 4-tier chart: chief of staff, directors, specialists, scouts, and exactly what belongs on each rung
- The 1 question that tells you when a cheaper model does the job just as well
- The access ladder: what each role gets to touch, and where you stop and ask first
Can you make your Chief of Staff a full agent?
What I’ve described so far isn’t technically an agent yet. It’s a persona with a set of skills. Piper shows up when a command runs over the planning folder: she reads the files, reasons about drift, writes back with my yes. But I start every session. She doesn’t exist between them.
A full agent is different.
An agent is something you can call by name from anywhere in your workspace and hand work to. “Have Piper prep my week.” She goes off, does it in her own separate session while you keep working, and reports back with a short summary.
You stop going to her office. You call her into yours.
But start with the smallest version
Does this system take discipline? Yes.
And it has to fit how you actually work, not how you wish you worked. I’ve watched people build beautiful systems that die in a week.
Not because they lacked willpower, but because the thing was built for a version of their day that doesn’t exist, or for the type of person they aren’t. I’ve made those very same mistakes.
What made this one stick is that it solved the exact thing I named at the start. I’m not the only place it all lives anymore.
So here’s the smallest version. 3 files are enough:
- Something that holds what you said matters.
- Something that remembers what actually happened.
- And today.
That’s what catches whether your goals and priorities are drifting.
And don’t design the entire team up front either. Start with 1 hire, then watch what you keep doing by hand.
Every specialist I’ve added came from noticing a job I kept repeating, not from an org chart I drew on day 1.
Your AI Chief of Staff shouldn’t just manage the noise.
It should notice when the important work stops moving.
If this is a system you’d like to try, I’ve packaged the planning core as well as entire set of documents in the Your First Hire Pack:
- The 5 commands that run your mornings and evenings, start to wrap up
- A connected planning folder that carries your priorities across days
- I don’t teach systems I don’t run: this is the planning core of my own chief of staff, packaged so you can install it in about 15 minutes
You can check it out here ➡️ https://systems.meerakothand.com/your-first-hire/
Your next step
Back to the question from the start. Is your business actually moving toward its goals, or is AI just helping you produce more?
Holding the whole plan in your head doesn’t have to be the price of working with AI. You don’t have to be the glue.
Start with the boundaries before the build: grab the free AI team org chart, decide the narrowest slice of access, and set up your 3 files this week.
Then tell me what you’re still carrying in your head… and if you set up the 3 files, come back and tell me what the log caught by Friday. I want to know.
FAQ
Can AI act as a chief of staff for a solo business owner? Yes, if you give it the right job. An AI chief of staff holds your business context (your priorities, your plans, a log of what happened) and runs your planning routine on top of it: proposing the week’s 3 to 5 priorities, resurfacing what’s overdue and flagging when your work drifts from what you decided. It doesn’t make your decisions. Start with 3 files: what you said matters, what actually happened, and today. (More on what to hand off to AI and what to keep.)
What does an AI chief of staff actually do? It holds your business context (your priorities, your plans, a log of what happened) and runs your planning routine on top of it: proposing the week’s 3 to 5 priorities, resurfacing what’s overdue, and flagging when your work drifts from what you decided. It routes work to specialist agents; it doesn’t do the specialist work itself.
How is an AI chief of staff different from an AI personal assistant for business? Most AI assistant demos clear your inbox, manage your calendar, and prep your meetings. A chief of staff does the harder part: it holds your priorities and goals and tells you when you’ve drifted from them. The difference is judgment about what matters, not access to more tools.
What’s the smallest way to start? 3 files: 1 that holds what you said matters, 1 that remembers what actually happened, and 1 for today. That’s enough to catch whether your priorities are drifting. Add commands and specialists only after those files are real.
Which AI model should an AI chief of staff run on? The most capable model you have access to, because the role is cross-business judgment calls. Specialists can run on the next tier down, and narrow high-volume jobs (pulling keyword data, checking rows against a rule) on the cheapest. Match the model to the job, not the smartest model to everything.
Does an AI chief of staff need access to my calendar and email? No. Mine reads plain text files I own: my priorities, my plans, and a log of what actually happened. Calendar and email can plug in later, but the heart of the system is files. Which also settles the multiple-accounts worry: files don’t care how many inboxes you have.
Does an AI chief of staff need access to all my tools? No. Start with the narrowest slice of access it needs to do the job, and expand only as you actually need to. The question to answer first is not what it can do, but what it’s allowed to do without asking you, every time.


