From Broken Blogs to Polished Posts, How I Built an AI Agent to Fix my Messy Content

If you prefer to watch, I have recorded a video. - https://youtu.be/70WYukFGvt0

A year ago, I started my solopreneur journey and did what any sensible millennial would do: I created a shiny new website at data-nova.io. But there was a problem. I'd been blogging on my old site, prathy.com, for years, and there was content there I genuinely cared about. I wanted everything in one place.

Wix had this handy little tool that promised to seamlessly import all my WordPress blogs. It imported, but...

The Import Disaster

The tool migrated the posts, but it ignored tags and categories, stripped code formatting, and left me with a mess of unreadable content. Featured images? Gone. Proper formatting? Non-existent. My old posts looked worse than they did on WordPress.

I knew I needed to fix it, but the thought of manually updating hundreds of posts made me want to close my laptop and pretend it never happened. So I did what any procrastinator does: I ignored it for a year.

Then I Started Playing with Agentic AI

Recently, I've been exploring agentic development and MCP (Model Context Protocol) tools. For those who haven't come across it yet, MCP lets you build custom tools that AI agents can use to interact with external systems. Think of it as giving Claude or ChatGPT the ability to do more than chat with you.

That's when it hit me: what if I built an agent that could intelligently fix my blog posts for me?

What I Built

I created an MCP tool that connects to the Wix API and uses Claude Code in VS Code to process each blog post. Here's what the agent does:

• Reads through each post and identifies issues

• Formats code blocks properly (my old posts had raw, unformatted code everywhere)

• Assigns relevant categories and tags based on the content

• Fixes grammatical mistakes (English isn't my first language, so this is genuinely helpful)

• Generates custom featured images using background templates I've pre-loaded

• Creates SEO-optimised meta descriptions

• Updates everything as a draft (never auto-publishes)

The last point is crucial for me. I don't trust it enough yet to let it publish automatically. I want control over what goes live on my site.

In MY VS Project Structure looked something like this:

Project Structure

wix-blog-manager/

├── main_interactive.py # Interactive workflow (RECOMMENDED)

├── main.py # CLI orchestrator

├── core/ # API client, orchestrator

├── modules/ # 10 improvement modules

│ ├── code_formatter.py

│ ├── seo_optimizer.py

│ ├── image_generator.py

│ └── ...

├── scripts/ # Utility scripts

├── data/ # Audit results, change history

├── backgrounds/ # Custom background images

└── blog_headers/ # Generated header images

How It Actually Works

I run the tool from my VS Code terminal using Claude Code. It walks through each post step by step and asks for my approval at key stages.

Here's what a typical run looks like:

1. The agent identifies a blog post that needs work

2. It finds code blocks that need formatting and shows me the before/after

3. I approve the formatting changes

4. It generates a summary of all the updates it plans to make: grammatical fixes, meta description, categories, tags, and featured image recommendations

5. It asks which background image I want to use (I have a folder of templates numbered 1-15)

6. It creates the featured image and shows it to me for approval

7. Once I approve, it updates the Wix draft

Sample Output in Terminal

Sample Background Image before Text Layer:

The whole process takes a few minutes per post, but it's saving me hours of manual work. More importantly, it's actually improving the quality of my content in ways I wouldn't have had the patience to do manually.

The Results

Here's a before and after of one of the posts I updated:

Before

After

The difference is night and day. The post is now readable, properly categorised, and optimised for search. And I didn't have to spend an hour fixing it myself.

Why This Matters (Beyond My Blog)

Someone recently asked me: "I use ChatGPT and Copilot for work, but I don't understand how I could use agentic AI in my day-to-day. How do I even start learning this stuff?"

Honestly, I don't think there's a formal learning path for this yet. I'm sure AI experts have frameworks and courses, but from where I'm standing, the best way to learn is to find a real problem you have and build something to solve it.

For me, it was my messy blog. For you, it might be:

• Automating repetitive data tasks in Power BI or Fabric

• Generating reports from multiple sources

• Cleaning up documentation or codebases

• Managing content across platforms

The key is finding something that's actually useful to you. That way, you're motivated to finish it, and you're learning something practical in the process.

What I'd Do Differently

If I were starting this project again, I'd probably spend more time upfront defining the logic and approval workflow. I started by exploring, so I didn't spend any time writing a prompt. After a few random failures, I added an "algorithm" section to my tool that outlines exactly what steps the agent should follow, and that made a huge difference. Without it, Claude Code would occasionally go off track and make random decisions.

I'd also think harder about error handling. Right now, if something breaks mid-update, I have to check which items were updated and which weren't manually. It works for my use case, but it's not bulletproof.

This project isn't perfect, and I'm still iterating on it. But it's already saved me dozens of hours, and more importantly, it's made me think differently about how I can use AI in my work.

If you've been curious about agentic AI but haven't found a way to apply it yet, I'd encourage you to look at your own workflows. What's the tedious thing you keep putting off? What's the repetitive task you wish you could automate? Start there.

You don't need to be an AI genius to build something useful. You need a problem worth solving and the willingness to experiment.

Until Next Time,

Prathy:)