Upgrading My Marketing OS | Rowan Builder Marketing

I’ve been working a bunch on upgrading my marketing OS. I’m trying to embody “AI native” for marketing departments. This means AI has context and capabilities to deeply improve all channels and roles, and people can use it.

I haven’t done an overhaul since February, right before the International Builders Show. Bad timing on my part. On that theme, I decided to do another overhaul during Q3 planning and onboarding a large client. At least my team can help this time.

It is just baffling how quickly everything is improving. I expect most of the capabilities that I’m building will be sold in a SaaS within a year or two in an astronomical subscription + usage plan.

Anyways, I wanted to document some of what I’ve built, so I’m having AI write some of the capabilities below.

Here are a couple TLDRs:

Start with a loop: Research better skill building and research skills, then rebuild the research and skill building skills
Use paid services including exa.ai
Check out Gbrain and the Karpathy Wiki - fantastic structure and capabilities
Use Google’s Open Knowledge Format for frontmatter
Have AI remind you how to use the system
So much is just about context management

Enjoy!

Mitch

AI wrote everything below:

What This System Actually Does

It’s an operating system that gives AI the two things it’s always missing: your company’s full context and the right tools, organized the way a marketing department is organized.

In practice, that shows up as work getting done differently. A project that used to take a week of back and forth now gets a solid first draft by morning. Every channel, the newsletter, the ads, the social posts, stays on brand and on strategy because the same knowledge feeds all of it, instead of each piece getting reinvented from scratch. What the company knows keeps building on itself instead of living only in one person’s head, which means it survives a busy week, a vacation, or a new hire. And there’s a lot less scrambling to find a freelancer for the routine stuff, because the routine stuff already has a home.

Most people use AI like a temp with amnesia. You open a chat, explain your situation from zero, get something generic back, and start over next time. This runs AI more like a department with real people in real seats, each one already briefed on the company and already good at their specific job.

The sections below each stand on their own, so skip ahead to whatever you’re curious about.

It’s Organized Like an Org Chart, Not a Chatbot

Instead of one all-purpose chatbot, the work is split into seats, the same way a real marketing department is split into roles.

There’s a seat for writing content, one for email, one for ads, one for reporting, one for planning. Each seat has an actual job description. That’s different from opening a single chat window and asking it to do everything, which is why that approach tends to feel generic. A general chatbot has no role to play and no fixed idea of what “good” looks like for the task in front of it, so it guesses at a reasonable middle ground and hands you something forgettable.

A seat fixes that by combining three things: clear instructions for that one job, the slice of company knowledge that job actually needs, and its own set of tools built for that work. The next two sections dig into the last two pieces.

Ask the newsletter seat and the ad-copy seat to work from the exact same facts about a project, and you’ll get two pieces that are shaped correctly for where they’re going: one reads like an update to someone who already trusts you, the other reads like a pitch to someone scrolling past. Same facts, right shape, because each seat knows its job.

Every Seat Gets Its Own Box of Crayons

Each seat keeps a box of crayons: written, repeatable ways of doing one task well that it reaches for instead of improvising.

A skill, in this system, is just a documented method. The steps to follow, what good looks like when it’s done, and usually a real example to work from. When a seat needs to write a newsletter or pull together a weekly ad report, it doesn’t start from a blank page and hope. It reaches for the crayon that already knows how that job gets done well.

The part that matters most is what happens after you build one. Fix a process once, turn it into a crayon, and it’s done for good. Every future newsletter, every future report, starts already knowing the method instead of relearning it. That’s how the work compounds instead of resetting to zero each time.

It also means quality stops depending on who happened to ask or how clearly they explained it that day. The newsletter crayon might lay out the structure that’s worked for months: what goes in the opening, how long each section runs, what tone fits. The ad-report crayon might define exactly which numbers matter and how to talk about them plainly. Either way, the standard lives in the crayon, not in someone’s memory.

The Secret Is Context: A Filing Cabinet Everyone Actually Reads

Most AI output feels generic because the AI doesn’t actually know your business, and fixing that is most of the game.

Ask a general AI to write about your company and it’ll guess. It doesn’t know your brand voice, who your customers actually are, or the decisions you already made and why. So it fills the gaps with something plausible and forgettable. This system’s whole job is closing that gap. It keeps one organized, living record of the brand voice, the strategy, the customer facts, the campaign history, and the decisions that came out of real meetings.

None of that is new information for a business to have. Most companies already have some version of it, scattered across a shared drive, a pile of old emails, and whatever a longtime employee happens to remember. The difference here is what actually gets read.

A person, no matter how good, isn’t rereading the full history of the brand before every task. The AI does. Every single time, it reads the whole filing cabinet before it writes a word. That’s institutional memory that never quits, never forgets, and never needs to be walked through the backstory again.

Why It Lives in Plain Files, Not Google Docs

The files themselves have to be something the AI can actually read, not just glance at.

A Word doc or a Google Doc looks like text on your screen, but underneath it’s really a bundle of formatting and layout instructions with your actual words buried somewhere inside. The AI can’t sit inside Google Docs the way you do. It can only see the file from the outside. A plain text file is different. Nothing is hidden in it, so the AI reads it exactly the way you would, and it can open the file, change a line, and save it back, the same way you’d edit a note.

That’s the idea behind the workshop I’ve built. Everything sits together in a plain code editor (I use one called VS Code), where the files and the AI share the same room. The AI isn’t just chatting with me about the files from a distance. It can go in, reorganize them, fix something, and leave the rest untouched.

Each file also carries a small index card up top, a few lines noting what the file is and what it covers. That’s what turns search into finding instead of guessing. Google actually built an open standard for exactly this, called the Open Knowledge Format, and its whole pitch is that if you can open a text file, you can read it. No special software required.

And the files link to each other, the way Andrej Karpathy has described building a kind of living wiki for an AI to work from. Instead of a pile of documents, you get a connected set of notes that keeps growing and holds together.

Matching the Brainpower to the Task

Not every task needs your smartest, most expensive thinker.

You wouldn’t send your CMO to resize an image or format a spreadsheet. That’s not what they’re for, and it would be a waste of their time and your money. The same logic applies once AI is doing real work across a marketing department. Some tasks genuinely need heavyweight judgment: setting strategy, weighing a tricky decision, catching something subtle. Other tasks are just production, the kind of work an intern could handle with clear instructions.

So I wrote it down. A simple rubric tells the system which kind of task it’s looking at, and routes it accordingly. Strategy and judgment calls go to the strongest thinking available. Routine, repeatable work gets handed to something cheaper and faster that’s plenty capable of doing it well.

The payoff shows up in two places at once. You get senior-level judgment exactly where it matters, and you get intern economics on everything else. Most setups either overpay for every task or underpower the important ones. Splitting the work by seniority gets you better speed and lower cost without giving up quality where it counts.

What About Privacy and Security?

Every client’s information stays in its own locked room, with its own key.

Nothing bleeds between clients or companies. Each one is walled off by design, not by good intentions. On top of that, I keep written rules for what the AI is allowed to read, what it’s allowed to send anywhere, and what never leaves the building under any circumstance. Anything headed out to the public also gets scanned automatically before it goes anywhere. It’s the same instinct behind office keys and an HR handbook, just applied to how the work actually gets done.

Where This Is Headed (and Why I’m Telling You Now)

None of this is magic. It’s good filing, clear roles, and the right tools for each job.

I want to be straight about that, because it’s easy to hear “AI marketing department” and picture something exotic. It isn’t. It’s a filing system that actually gets read, a set of defined jobs instead of one genie trying to do everything, and a simple rule for matching effort to task. Every piece of it exists today.

My guess is that within a year or two, someone packages this up and sells it as expensive software, charged per seat, with a polished dashboard on top. That’s fine. But the ingredients are already sitting out in the open right now for anyone willing to get organized about how they use these tools.

The part I keep coming back to is this. The software was never really the hard part. The hard part has always been knowing what a marketing department should actually be doing, what good work looks like, and how to tell the difference between busy and effective. That’s the piece I’ve spent my career on, long before any of this existed. The tools just changed what I can do with it.

Some Prompts to Get You Started

If you want to build your own version of this, don’t start by building all the pieces. Start by building the thing that builds the pieces, and then let it improve itself. The first prompt below is the one that matters most. It sets up a loop where the AI researches how to build good skills, builds you a skill builder, uses that to build a research skill, and then uses the better research skill to improve the skill builder again. Each pass makes both a little sharper. Everything else gets easier once that loop is running.

These are written in plain language. You can paste them into whatever AI tool you already use and adjust the parts in brackets.

1. Start the loop (do this one first).

Research how to write excellent, reusable AI skills: what makes an instruction
set clear, repeatable, and hard to misread. Summarize the best practices into a
short guide. Then use that guide to write me a "skill builder," a repeatable
procedure for turning any task I describe into a clean, reusable skill. Next, use
the skill builder to create a "research skill," a repeatable procedure for doing
thorough, well-sourced research. Then use that new research skill to re-research
how to build skills, and rewrite the skill builder to be better. Keep looping.
Each pass should sharpen both the research skill and the skill builder. Tell me
what changed each round.

2. Start your filing cabinet.

Interview me about my business one topic at a time: our brand voice, who we sell
to, how we win work, our offers, and the big decisions we've already made. After
each answer, save it as its own short plain-text file with a label at the top
saying what the file is. Keep everything organized so you can reread it before any
future task.

3. Give a role its own seat.

Act as my [email marketing] lead. Write a job description for this seat: what good
work looks like, what it should never do, which of my context files it should read
first, and which skills it should use. Save it as a file I can reuse to brief you
every time.

4. Turn a task you repeat into a skill.

Here is a task I do often: [describe it]. Ask me how I like it done, then write it
up as a reusable skill: the steps, the standards, one good example, and the common
mistakes to avoid. Save it so future requests can reuse it instead of starting from
scratch.

5. Have it teach you the system.

Look at how my files and skills are organized and write me a short "how to use
this" guide in plain language: what lives where, how to ask for things, and the
three moves that make the biggest difference. Remind me of it whenever I seem to
be doing something the slow way.

None of this requires you to be technical. It requires you to be organized, and to be willing to write things down once so you never have to explain them again.