Creating my marketing agent was easy. Not having to explain my brand every time is what made it useful.
I use an agent called marksudo inside 9xChat, the AI workspace we’re building at 123sudo.
My ideas rarely arrive as clean briefs. They start as scattered thoughts, questions, or a feeling that something needs a different direction.
I wanted an agent that could help me structure those thoughts without restarting the conversation about my brand every time.
Here’s what makes that work.
A role, an approach, and access
My agent’s configuration has three distinct parts:
- Instructions: its marketing role, responsibilities, and guidance about how to work with me.
- Skills: reusable guidance for tasks like content strategy, research, and editing.
- Tools: capabilities such as web search, reading pages, searching previous conversations, and browser interaction.
The distinction matters.
Telling an agent to “research the market” doesn’t give it internet access. Giving it web access doesn’t explain what good research should look like.
Instructions define the job. Skills guide the approach. Tools make actions possible.
In 9xChat, I can add or create skills and configure the tools available to the agent. I can also choose which model powers it, rather than treating one model as the agent’s entire identity.
Memory is what I notice every day
The biggest benefit is not repeating:
- What our brand stands for.
- Which strategic direction we’ve chosen.
- What language feels like me.
- Which approaches we’ve already decided against.
When relevant preferences and decisions are saved, there’s something to build on.
During a longer strategy, the agent helps me reorient: where we are, what we’ve discussed, and what might come next.
That doesn’t mean it automatically remembers everything.
I separate three things:
| Need | Source |
|---|---|
| Brand preferences and saved decisions | Memory |
| Details from earlier discussions | Conversation search |
| Information from reference documents | Knowledge search |
For example, checking what we’ve already posted needs an accessible record. A remembered brand preference isn’t a publishing history.
That distinction keeps my expectations realistic.
Research happens where the work happens
When an idea needs research, I don’t have to leave the conversation, search manually, and paste everything back.
With the relevant tools enabled, the agent can search the web and read pages inside the workflow.
Browser interaction goes further: it can navigate pages, fill form fields, or work through booking steps. Those capabilities still depend on website access, permissions, and appropriate confirmation before consequential actions.
For my marketing work, the useful part is simpler: research and thinking stay connected.
My actual workflow is less polished than a demo
It usually starts with something like:
“I have a rough idea. Help me connect it to what we’re already doing.”
Then we:
- Connect the thought to our existing direction.
- Retrieve earlier context or research what’s missing.
- Turn it into an outline, plan, or clearer argument.
- Review what fits—and discard what doesn’t.
Sometimes I use read-aloud to listen to the response instead of reading another screen of text.
I still choose the direction. I check claims, review sources, and approve what gets published.
The agent helps structure my thinking. It doesn’t own it.
What I’d suggest if you’re building one
Start with one recurring job, not every possible capability.
Give the agent a clear role, add relevant skills, and grant only the tools it needs. Save durable preferences and decisions—not every passing thought.
Then test it on unfinished work. Notice what you keep repeating. Improve the setup around that friction.
For me, the biggest improvement wasn’t a more elaborate prompt.
It was continuity.
I don’t need an agent that owns my brand direction. I need one that helps me carry it forward.
Disclosure: I’m part of the team building 9xChat. This is my own workflow, not an independent review.
Configuration details: Agents · Tools and Skills · Memory
What do you find yourself explaining to your AI repeatedly?





