How to Build an AI Command Center: Your Whole Business on One Screen
What is a command center, and how is it different from a dashboard?
A dashboard reports. It shows you numbers, charts and a status colour, and it leaves the interpretation to you. You look at it, think "okay," and close the tab.
A command center decides. It takes that same information and tells you three things: what's on fire, whether you're on pace, and the single most important thing to do today. The difference isn't visual. It's whether the screen ends with a number or with an instruction.
That distinction is the whole reason most people's dashboards die. A screen that reports is a screen you have to interpret every morning, forever. A screen that decides earns its place in about four days.
Why do most business dashboards get abandoned?
Three reasons, and none of them are design.
It shows the wrong things. Somebody built what looked impressive instead of asking what the owner actually needed to see. Six charts, no answer.
It has to be maintained by hand. Every manually-updated dashboard has a half-life of roughly nine days. You update it, then you update it late, then you stop, then it's lying to you.
It's full of invented numbers. If a section has no real data behind it and the tool filled it in anyway, the screen is worse than nothing — because now you're making decisions on fiction.
A command center that survives fixes all three: it's built from an interview, it's regenerated on a schedule, and it says "no data yet" instead of guessing.
What should be on a command center?
Whatever you actually open every morning, in the order you open it. That's not a cop-out — it's the design rule. If you open email before your calendar, email goes higher, even if that's not how a dashboard "should" be arranged. The screen you'll still be opening in March is the one shaped like your morning.
Here's what's on mine, as a starting template:
What's on fire — and who's waiting on me
The one move today — a single named action, at the top, alone
Invoices out — paid and still waiting, with an age on each
Goals — long-term and short-term in separate sections, showing pace not just progress
Projects — every one, and what it's blocked on
Content — what I owe and when
Calendar + the to-do list I actually use
Steal that list. Then cut half of it. A section you've never once acted on is a section that's costing you attention every morning.
The nine questions that build your screen
The interview is the product. The design is just what falls out of it. Answer these before anything gets built:
What do you open first every morning — and in what order?
What are you most afraid of forgetting?
What are the long-term and short-term goals you want to track?
How will you know you're on track — what has to be true this week?
What are the important things you want to track?
Do you want to track invoices you've sent — what's been paid and what hasn't?
Do you have a to-do list? Where does it live, and do you want it on here?
What apps do you want scanned every day so you know what's going on?
If this screen could tell you the one most important thing to do today, what would it be?
Question 5 is what makes it yours instead of a template — projects, clients, shoots, submissions, training, whatever your work actually is. Question 9 is what everything else is in service of, which is why it goes last.
How does it get your real data?
This is the step people skip, and it's the reason their version ends up half empty.
An AI model cannot magically know what's in your calendar, your inbox, your CRM or your invoicing system. It can only build from what it has been given access to. There are four ways in:
Files already in the folder. The simplest, and it never breaks. Notes, client files, project files, a CSV you exported this morning.
Connectors. The built-in integrations — calendar, email, cloud documents. Best for the things that update constantly.
MCP servers. For tools that publish one: a CRM, your books, a task manager. The screen queries them live on every rebuild.
Exports. The escape hatch, and there's no shame in it. Export a CSV, drop it in the folder. Half of a working command center is built this way in week one. Just put the date in the filename so the screen can tell you how stale it is.
Anything you don't connect won't appear. That's a decision, not a bug — as long as you make it on purpose.
How do you make it not look AI-generated?
Two free skills, and they do different jobs.
Frontend Design handles the overall look, layout and visual system.
impeccable reviews what came out and fixes spacing, hierarchy, readability and dark mode.
Install both, then build. Skip them and you get the grey card-grid every AI produces by default — technically fine, and completely forgettable.
Then pick a look on purpose. A spaceship console, a newspaper, a plain operator screen. It genuinely doesn't matter which, as long as you can read it in ten seconds. Describe the feeling or paste a screenshot of something you like.
What if the first version is wrong?
It probably will be, and that's normal. Two rules make revision fast:
Revise one problem at a time. "The money section is buried" gets fixed. "Change everything" produces a second draft worse than the first.
Name the problem, not the fix. You know what's wrong with your screen. You don't necessarily know the right layout solution — that's what the design skills are for.
And a useful test: if you have to scroll past six charts to find the thing you came for, the design failed. Not the data — the design.
How do you keep it current?
You never update it by hand. That's not a preference, it's the entire mechanism.
The page gets regenerated from source files on a schedule — every morning before you wake up is the sensible default. If something on the screen is wrong, you fix the source file and let it rebuild. You open the command center to find out, not to update.
Maintain one manually and it's dead inside two weeks. Automate the rebuild and it's still alive next year.
Where should it live?
Two front doors, and most people end up using both.
As a link — publish it so it works on your phone, on any laptop, and can be handed to a business partner without giving them access to everything else.
In your notes — pinned at the top of your vault, so it's the first thing you see in the app you're already in.
One caution before you publish anything publicly: read the page first. Client names, revenue figures and outstanding balances travel with it.
Frequently asked questions
How long does it take to build? An afternoon, if your information already lives somewhere readable. Longer if you're also connecting apps for the first time.
Do I need to know how to code? No. You answer nine questions and describe what you want it to look like. The build happens in Claude Code, but you're writing sentences, not syntax.
Do I need a second brain or vault first? It helps a lot, because the command center reads from it. If you don't have one, build that first — the free Creative OS template is part one of this series.
Will it make things up if it can't find data? Not if you build it with this rule in place: an empty section must say "no data yet — connect X" rather than showing a plausible number. It's the single most important constraint in the whole build.
Can I use a different AI model? Yes. Your knowledge lives in plain files you own. The model is a brain you plug in, and you can swap it whenever a better one ships.
Is the template really free? Yes, MIT licensed: github.com/MitchelDoomz/chcommandcenter. Take it apart.
The point of the screen
It doesn't make the decisions. It deletes the twenty minutes every morning spent opening six tabs and assembling a picture of your own business, so the only thing left is the call.
That's the version of AI at work that actually holds up — not replacing the judgment, just removing everything standing in front of it
Get the free template:github.com/MitchelDoomz/chcommandcenter
Want this built for your business? Book a call below
