This morning, I was thinking about how many times I’ve started a new job. And not just a new full-time job, but a new client engagement. Then I thought about how many times I’ve hired, onboarded and trained a new employee, or a collaborator I’ve brought onto a client project.
Over the years, I’ve learned the hard way that the way that you set someone up for work is one of the most critical parts of the work itself.
You don't simply hand a new marketing hire a laptop and say "go rewrite the homepage and give me three landing pages for these three products."
First, you have to give them context. And you have to do it in way that is both thorough and structured. Walk them through who the customers are, what they care about, what they need, why they made a purchase, and why they purchased again.
You show them the brand voice, with examples for how brand tone shifts across channels and customer experiences.
You tell them the words you never use — the banned vocabulary — and the claims you can't make.
You introduce them to the competitors you watch, and the companies outside of your category doing something interesting you want to test for your own brand. You point them at the work that's already approved. The brand vocabulary. The brand claims. The customer testimonials. The CEO’s recent quote from the media interview that drove sales.
Nobody calls that overhead. It's how you get a new person briefed and ready to develop content that’s on-brand.
But not every team has the resources to do thorough onboarding and training. And many don’t have the resources for that new hire. So they turn to AI to fill in the gaps on their team. But in doing so, they often create holes in their branding and storytelling.
An Agent Doesn't Get a First Week
An AI model or agent starts from zero every time. It's fluent, fast and eager, and it knows nothing about your brand except what you remember to tell it in that conversation. So you do what everyone does: you paste in a paragraph, or you write a long prompt, or you hope the agent guesses well. Even a thorough conversation will provide confident answers from an AI model. But those answers and that output? It can still be confidently wrong.
I came to this the long way around. I'm a visual wordsmith who prefers color and formatting, and I resisted terminals for most of my career. When my Jinn co-founder Dan walked me through setting up Claude Code over a weekend, my first thought wasn't about what the model could do. It was about what it didn't know. It was the realization that once I had developed my Brand DNA, I could suddenly deliver more value for my clients. It meant I could create even more compelling branded content, campaigns and experiences, and help my clients compete with the larger players in their industry.
That realization? It was enough to hook me. I went from customer zero to co-founder at Jinn. You can read more about that experience here.
But what really changed is that, as a 20+ career marketer, I started to think about the big picture again. I stopped thinking in terms of what our team could deliver, and more importantly on what our team could create that would truly wow our clients, and drive even more measurable results.
And all of this was working in a terminal, using Jinn Agents via MCP.
What MCP Does, in Two Sentences
MCP, the Model Context Protocol, is a standard way for an AI tool to ask an outside source for information while it works. Instead of you pasting your brand into the conversation, the agent can ask for it.
Jinn Agents is our version of that source. It exposes your Brand DNA as an MCP surface, so any agent or app you run can read the same record. You can see the same idea on jinn.works, where the example call asks for a brand by name and the record comes back.
The Briefing List
Here is what a new hire needs, and where it lives in a Jinn brand record. The six groups are Identity, Palette, Voice, Evidence, Audience and Competitors.
| What You'd Tell a New Hire | Where an Agent Finds It | |
|---|---|---|
| Who we are and what we stand for | Identity | |
| How we look | Palette | |
| How we sound, and what we never say | Voice | |
| What we can prove, and what we can't claim | Evidence | |
| Who we're talking to | Audience | |
| Who else is in the room | Competitors |
If you've ever onboarded a marketer, you've made this list. The difference is that you only have to make it once.
Four Steps
Jinn Agents is in early access, and the door is an application. We’ve also opened access to Ghost, our Brand DNA Voice and writing tool, is also open, and we’re offering founding member pricing for the first customer cohort.
Jinn Agents is listed in the official MCP registry, and the docs live on jinn.works. Each person on your team can create their own access token for the brand, and everyone on the same subscription draws from one shared monthly pool so you never have to re-brief your team. Admins can see every token and revoke any of them, at any time.
This is the habit that matters. Start the task by telling the agent to read your brand record, then write. Don't assume it will.
Check the work early. Give feedback. The record gets you most of the way on voice and facts, and a person still decides what goes out.
A Prompt You Can Steal, No Really
The last line is the one that does the work. Asking the agent to show what it used turns a black box into something you can check.
Asking the agent to show what it used turns a black box into something you can check.
What This Doesn't Do
An agent that knows your brand isn't an agent that approves your work. The facts it reads came from your brand record. Whether this particular claim is right for this particular launch is still a human call. The tools that change anything in Jinn are few, and the ones that exist can't spend money.
I find that comforting. A new hire with great onboarding still gets edited.
A new hire with great onboarding still gets edited.
Next Steps, If You Want Them
If you work with AI agents every day and you're tired of re-explaining your brand to each one, apply for early access to Jinn Agents and founding member access to Ghost. The setup documentation to guide you are at jinn.works/docs/mcp. If you get stuck? Don’t worry. You can reach us at hello@jinn.works.
What does MCP do?
MCP, the Model Context Protocol, is a standard way for an AI tool to ask an outside source for information while it works. Instead of you pasting your brand into the conversation, the agent can ask for it.
How should I start a task with an agent?
Start the task by telling the agent to read your brand record, then write. Don't assume it will.
Does an agent that knows my brand approve my work?
An agent that knows your brand isn't an agent that approves your work. Whether this particular claim is right for this particular launch is still a human call.
How does a team share access?
Each person on your team can create their own access token for the brand, and everyone on the same subscription draws from one shared monthly pool so you never have to re-brief your team. Admins can see every token and revoke any of them, at any time.