I had about 100 blog posts stuck in draft on my old WordPress blog. Not outlines. Not notes. Actual drafts with titles, intros, and half-written arguments that I never finished.
I moved to a new blogging platform on March 1st when I got the itch to write again. I already have 50 drafts here. It is not the platform. It is clearly me. I have the ideas. I have the content. I never hit send.
I start things. I research obsessively. I outline carefully. Then I stall. The draft sits. I open it a week later, decide it needs more work, and close it. Repeat for months until the topic feels stale and the draft joins the pile.
This post is about whether AI can solve that problem.
Why solo blogs die
Most bloggers write alone. That is the whole point. Personal blog, personal voice, personal responsibility.
It is also why most blogs die.
Professional publications have editors, fact-checkers, reviewers, and a production schedule that forces work out the door. Solo bloggers have none of that. You are every role at once, and the role that always gets skipped is the one that finishes the draft.
I know because I skipped it a hundred times.
What I built
I built an agentic AI workflow. Not one assistant. A team of over 20 specialized agents organized into a pipeline.
I start off with an idea for a blog article. I write what I know and the agent interviews me 1:1 before every post. Targeted questions that force me to say what I actually think instead of what sounds safe. They organize my notes and past writing so the tone stays consistent. Sometimes they phrase something better than I would have. I keep those. Sometimes they produce something that sounds like nobody wrote it. I cut those. The judgment call is mine either way.
Here is the order they run, because the order matters:
Before I write anything:
- A signals scout scans 80+ tech blogs weekly and tells me what the industry is talking about. It also tracks what I published recently so I do not repeat the same topic back to back.
- A topic strategist takes my rough idea and sharpens it into an angle with a clear audience and outcome.
- A research analyst gathers evidence and sources so the post is grounded before I write a word.
Building the draft: 4. An outline architect builds the narrative structure from my raw ideas and the research. 5. A draft writer produces the first scaffold from my voice, notes, and outline.
Review gates (these run in parallel): 6. A fact-risk reviewer checks every claim and flags anything I cannot verify with public sources. 7. A security reviewer scans for confidentiality issues. I work at Microsoft, so this is not optional. 8. A contrarian reviewer pushes back on weak arguments before a reader does.
Polish: 9. A voice editor catches sentences that sound like AI filler and flags them for my rewrite. 10. A revision loop takes all the reviewer feedback and produces a corrected draft. It loops until the quality gates pass. This is the part that used to be entirely on me, and it is the part where most of my drafts died. 11. A SEO editor handles metadata, slugs, and distribution prep. 12. An image director creates image briefs and alt text.
Publishing: 13. A publisher agent sets the final frontmatter, runs pre-flight checks, and handles deployment on a set schedule once I approve. 14. A LinkedIn poster prepares distribution variants. I review and post manually. 15. An editorial scribe logs everything that happened during the workflow so the next post benefits from what this one learned.
Every post runs through the same gates in this order. If a post fails any gate, it does not advance.
That pipeline has already failed me once. A loophole in the quality gates let something through that should not have passed. I fixed it. I am sure I will find more loopholes as I write. That is part of the experiment.
If I cannot stand behind every paragraph in a live conversation, it does not ship.
You will be the judge of whether I get it right.
Does it work?
I am not sure yet.
The agents handle the pipeline. The pipeline forces me forward. When a fact-checker flags a weak claim, I have to fix it or cut it. When a voice editor highlights a generic paragraph, I have to rewrite it. The system does not let me coast the way I did with 100 drafts on WordPress.
Whether that is solving the problem or just building a better system around it, I will find out by publishing consistently for the next year.
The risk nobody talks about
The internet has a slop problem. AI-generated content published without a human edit, content that takes more effort to read than it took to produce. I do not want to add to that pile.
My agents once produced a paragraph about “leveraging cross-functional synergies.” I had to read it twice to confirm that yes, an AI had just written LinkedIn filler in my blog post. I deleted the whole paragraph.
That is one risk: AI that sounds confident and means nothing. If you are not reading closely, it ships. So I treat every agent output as a suggestion, not an answer. If it does not sound like me, it does not stay.
The other risk is more personal. I keep coming up with ideas and never publish. That was true before AI, and AI does not fix it by itself. The pipeline forces a schedule. The agents force quality gates. The question is whether structure can override the instinct to keep tweaking instead of shipping.
Some people will not like this approach, and that is fine. If I get one person writing again, that is enough.
This week
Ship something this week. Imperfect beats silent every time.
I spent years collecting drafts instead of publishing them. 100 on WordPress. 50 already on this platform. That pattern is the problem, and I am the cause.
I still have my voice. I still stand behind everything I post. The goal is for people to learn from it. If even one person reads this and decides to finally ship that draft they have been sitting on, that is enough.
Can AI solve the problem that is me? This blog is the experiment. I hope you join me for the ride.
Disclaimer: I work at Microsoft. The views expressed here are my own and based solely on publicly available information. This content is for educational purposes and does not represent official Microsoft guidance or commitments.