5. How I Built My Own AI Marketing Agent as a Senior (Non-Techy) Marketer
- Samara H. Johansson
- Aug 4
- 7 min read
Updated: 4 days ago
The best way to understand something that scares you is to build it yourself.
I am not an engineer. I have never written production code. A recent colleague questioned whether I could adapt quickly enough to new systems (citing my learning curve with a website publishing tool as evidence that I might need significant hand-holding with anything technical.)
That stung. And it stuck.
So when I found myself passed over for a consulting role — partly because I lacked agent-building experience — I made a decision. I wasn't going to read another article about AI agents. I wasn't going to attend another webinar about the future of marketing technology. I was going to build one.
This is the honest account of what happened next.
The Question Every Senior Marketer Is Quietly Asking
Before I built anything, I had to answer a question I suspect many of you are sitting with right now:
Is this actually something I can do? Or is agent-building genuinely the domain of engineers and developers?
My first instinct was to look at Copilot Studio. This is Microsoft's dedicated agent-building platform. It's powerful, well-designed, and purpose-built for creating and deploying AI agents at scale across an organization.
But I quickly learned it wasn't what I needed. Copilot Studio is designed for wide-scale, company-wide deployment. For personal use (for a solo consultant wanting to build and test an agent) my existing Microsoft subscriptions were sufficient. I had Office Agent and Microsoft 365 Copilot. That would do.
The barrier I had assumed existed — the one that said "this is only for technical people with enterprise budgets" — turned out to be much lower than I thought.
What I Decided to Build and Why
My goal was deliberately strategic, not just technical. I didn't want to build an agent for the sake of building one. I wanted to build something that:
Proved I could do it: to myself as much as anyone else
Produced real, usable output: not a demo, not a prototype
Served an actual business purpose: promoting SamaraGlobal.com and my consulting practice
Reflected my point of view: written in my voice, advancing my argument about AI and marketing
The result was a Content Agent consisting of a five-module system that takes a topic and produces a complete content package: research, strategy, a full SEO-optimized blog post, three LinkedIn post variations, and an editorial review with a quality score.
The topic I chose for the first run? "Why firing your marketing department for AI agents is a strategic mistake."
Deliberately provocative. Entirely intentional.
What I Had to Set Up Before Writing a Single Line of Content
This is the part nobody tells you about. Before the agent could produce anything, I had to build the environment it would run in. For someone who had never done this before, this was the most intimidating part. And also, in retrospect, the most empowering.
Here is everything I set up from scratch:
Python: The programming language the agent runs on. I had never used it before.
Visual Studio Code (VS Code): The coding environment where the agent lives and runs.
pip: Python's package installer, used to add the libraries the agent needs.
OpenAI API access: The AI brain that powers the agent's thinking and writing.
python-dotenv.: A security layer to store my API key privately.
A .env file: To keep my credentials off GitHub and out of public view.
A GitHub account and repository: To store the agent's code and showcase it publicly as part of my professional portfolio.
None of this was on my computer two weeks ago. I set all of it up, step by step, with guidance from Office Agent; asking questions, making mistakes, and figuring it out as I went.
The Five Modules Inside the Agent
Once the environment was ready, the agent itself was built around five modules. Each one handled a distinct stage of the content creation process:
01. Researcher: Finds relevant data, statistics, market context, and source material on the given topic. This is the equivalent of asking a junior team member to spend a morning doing background research before a brief.
02. Strategist: Takes the research and builds a content framework (angle, structure, key messages, audience focus.) This is where my own brand positioning and voice profile are applied.
03. Writer: Drafts the full blog post using the research and strategy as its brief. I fed this module my own writing samples so it could learn my tone, rhythm, and style.
04. Repurposer: Takes the finished blog and creates three LinkedIn post variations; different hooks, different lengths, different angles on the same core content.
05. Reviewer: Runs an editorial review of the blog and LinkedIn posts. It checks for voice consistency, SEO alignment, clarity, and overall quality. It produces a score and specific recommendations.
The entire pipeline runs sequentially. One topic in. A complete content package out.
What Happened When I Hit Run
It didn't work.
The conversation timed out. The code had bugs. And for a brief moment I thought: maybe they were right about me.
But instead of stopping, I did something that felt surprisingly natural: I turned to Microsoft 365 Copilot and asked it to debug the code. I described what had happened, shared the error, and asked for help.
It fixed it.
Second attempt: the agent ran all five modules without interruption, saved the output to a timestamped file, and printed a completion summary to the screen.
I sat there and read it. A complete blog post. A content strategy. Three LinkedIn variations. An editorial review. All generated in minutes from a single topic prompt.
The feeling wasn't triumph exactly. It was something quieter than that. More like: oh. I can do this.
What the Agent Does Well. And What It Doesn't
I want to be honest here, because the point of this series has never been to oversell AI.
What the agent does brilliantly:
Produces well-structured, readable first drafts at speed.
Maintains consistent formatting and SEO structure across every post.
Never gets tired, never misses a deadline, never needs a brief repeated.
Scales infinitely, changing one line of code generates an entirely new content package.
What the agent cannot do:
Decide which topics matter and why. That strategic judgment is entirely mine.
Know that four blogs should form a coherent series with a narrative arc. I designed that.
Understand the nuance of a market moment. Why this argument, right now, for this audience.
Replace the experience of having sat in many many stakeholder meetings and knowing what actually moves people.
The agent executes my strategy. It does not create it.
What I Know Now That I Didn't Know Three Weeks Ago
Building this agent taught me three things I didn't expect:
1. The technical barrier is lower than you think. If you are already using AI tools daily for research, writing, analysis; you are closer to building an agent than you realise. The gap between "using AI" and "building an agent" is smaller than the industry wants you to believe.
2. The strategic barrier is higher than most people acknowledge. An agent without a clear brief, a defined voice, and a coherent strategy produces generic content at scale. The technical build took days. The strategic foundation (my messaging framework, my audience definition, my point of view) took two decades.
3. This changes what "senior marketer" means. The most valuable marketing leaders going forward will not be the ones who can do everything themselves. They will be the ones who can direct both human teams and AI agents toward a coherent strategic outcome and know when to trust the output and when to override it.
That is a deeply human skill. And it is not going away.
The Bigger Point
I built this agent to prove something to myself. What I proved was both more and less than I expected.
More because the technical side was genuinely achievable for a non-engineer with curiosity and patience.
Less because the agent, impressive as it is, only works because of everything I brought to it. The strategy. The voice. The point of view. The two decades of knowing what good marketing actually looks like.
A company that fires its marketing department and replaces it with agents will get fast, cheap, scalable content. What it will not get is a brand. It will not get a strategy. It will not get the judgment that turns content into customers and customers into advocates.
Those things still require a human. A senior one. And now that human can also build the agents.
--
You can follow the build and see the full code on GitHub. And if you are curious about what this kind of capability could mean for your own content production, I would be glad to talk.
A note on accessibility. In its current form, the Content Agent requires Python, a code editor, and API keys to run. It is built for technically curious users who are comfortable in a development environment. If that is not you, that is completely fine. The concepts apply regardless, and the outputs speak for themselves. A version that anyone can run from a browser, without any setup, is on the roadmap. That is the direction this is heading.
Follow the build: github.com/SamaraHJohansson/agent-portfolio
--
The complete series:
--
Samara H. Johansson is a senior B2B marketing consultant specializing in developing and then activating global brand positioning, messaging frameworks, and AI-augmented marketing strategy to generate leads. She works with companies navigating growth, repositioning, and international market expansion. Learn more at SamaraGlobal.com and subscribe to The AI-Augmented Marketer on LinkedIn.
Comments