This is an agent I built for myself. Writing and publishing content consistently is something every consultant, influencer, aspiring thought leader knows they should do and most struggle to actually keep up with. This is how I solved that problem for my own business, and it’s an example of the kind of solution I can build for clients.
The problem
As a solo practitioner, I don’t have hours to spend writing blog posts and LinkedIn content. But I also won’t publish ideas that aren’t mine, or content that doesn’t represent how I actually think. Posting generic AI-generated content that reads like it could have been written by anyone, or AI slop that reads like a bot wrote it is worse than not posting at all. My perspective, my takes on what’s happening in AI, my way of explaining things has to come through in what I post.
Writing a post fully by hand could take a few hours from first draft to published. That’s time I don’t consistently have, which means content either doesn’t get written or gets deprioritized. It’s taken me several months, knowing I should be posting regularly, to actually start doing it because I just couldn’t find the time.
What I built
I built a blog writing agent using Claude Projects. I make sure what I post is my thoughts and analysis by designing a workflow that requires me to provide a topic and outline for an article, complete with research and data points where needed. The agent is not asked to fill in gaps in information or provide its own analysis while writing the draft. I have provided it a set of reference files including guidance on SEO and AEO best practices, style and tone instructions, and a pointer to all of my previously published posts so that the draft it produces not only uses guidance on how to write an effective article but is in my voice. I sourced the SEO and writing guidance from online references and had Claude synthesize them into structured markdown files. My previously published posts serve as writing samples that train the agent on how I write.
The workflow is step-by-step and I stay in the loop at every stage:
- I write a topic and outline. The ideas and main points are mine before anything goes to the agent.
- The agent drafts a headline and post using the reference files and my outline as inputs.
- We edit together. If the draft is substantially off, I work with the agent to rewrite sections until it says what I want to say. Once it’s close, I edit by hand to get it fully into my voice.
- The agent saves the post to my local storage via Desktop Commander, a Connector in Claude that allows it to take action on my local computer. This preserves a backup of my articles in case of a website disaster.
- Once I’m done with the article, the agent writes three LinkedIn posts to promote the article. Each is highlights a point or sub-topic in the article and focuses on one of several goals for my posts: some posts are informational, some share my perspective on recent AI developments, some more directly promote my business.
- I edit those the same way: working with the agent where needed, then finishing by hand. When I’m satisfied, these get saved to my local computer as well.
- I publish the article to my website and use one of the LinkedIn posts to promote it.
- The agent logs the post to an Airtable data store: headline, publication date, and the LinkedIn posts. Over the following week I’ll re-up the article with the other LinkedIn posts and I have it mark those publication dates in Airtable as well.
On voice and AI assistance
Using AI to help write content that’s supposed to represent your thinking requires being deliberate about where the AI is doing useful work and where it isn’t.
In this system, the ideas come from me. The outline comes from me. The agent’s job is to help structure and draft the delivery, working from guidance that reflects how I write and what makes content effective. The draft is just a starting point; I do substantial rewrites when the draft is off, and I always do a final pass by hand (usually more than one) before anything gets published. Using this system gets me maybe 60% of the way to my final article – structuring the argument, applying SEO and AEO principles like separating the content into short sections with headers – but my manual edits are often still substantial.
Because I don’t yet have a large library of my own writing for the agent to learn from, getting the drafts fully into my voice takes more effort than it will over time. That’s expected, and it’s part of why the system is designed the way it is — as I publish more, the agent has more to work from. This might mean the Claude draft gets me closer but I will still want to reword things, shape the content, put in the personal content required to make the article fully my thoughts.
The net is: what goes out under my name reflects what I actually think. The agent just helps me say it more efficiently.
Where it stands
I’ve published three article using this system, so it’s still early. A post that would have taken a few hours by hand now takes more like one hour, maybe more if it takes a lot of research first. The bigger gain is consistency — having a structured process means I’m more likely to actually do it.
What’s next
The Airtable logging is the foundation for a reporting layer I haven’t built yet. Next up is pulling engagement data from LinkedIn and my website into that same base, which will give me visibility into how each post performs and that’s where the real magic will happen. Automating the data gathering and analysis will allow me to learn from what I post – what resonates most with people, what helps me find clients, what activity is and isn’t impactful to building my business.
That analysis will feed back into the agent’s guidance. It will also become a set of KPIs for my own activity, keeping me accountable to the marketing work the business needs.
If I had to do this by hand it would be hours of manual work each time I wanted to check my metrics which means I’ll procrastinate doing it, I’ll learn slower, and be less effective. With the automation I’ll be able to take a quick peek weekly to keep on top of things and adjust quickly based on the data.
Why this is relevant to client work
Content creation is a common bottleneck for people in many industries.
- Marketing requires generating content that resonates with people. Unless a company’s marketing is super dialed-in and the audience never changes, their message will need to be adjusted over time. Data makes that easier.
- The people with the most useful things to say are usually too busy to say them consistently, and off-the-shelf AI writing tools often produce output that doesn’t sound like anyone in particular.
- “Blank piece of paper syndrome” is real. Even great writers with lots of time on their hands occasional suffer from writer’s block or struggle to get started.
This system is an example of an approach that uses the best of AI plus the unique human judgement that can never be replaced: building an agent that’s trained on a specific voice, integrated into a specific workflow, and designed to keep a human meaningfully in the loop rather than just generating content on autopilot.