Olly

I used Claude Sonnet 4 to help a customer import their blog archive in 46 minutes

I'm sure people are starting to get a little tired of hearing about how AI solved this coding problem or that coding problem. If you look at X (urgh, I know - don't) you'll see countless posts about agents 10x-ing productivity alongside tales of people building clones of Big SaaS in a weekend (followed by a cease and desist!). Some of it is totally legit, some of it is probably BS. Either way, it still blows my mind on occasion. 

Today I received an email from a new Pagecord customer asking about a feature that would help them import 300 (!) blog posts:

Uploaded image

My initial thought was to add a quick feature to show these hidden fields by default, but that would still be a pain for the customer. I asked them whether they had an export of their blog and they sent over a zip file with HTML and images in it (it was a Dreamwidth export). It looked rather fiddly but there was enough structure there to potentially extract the data and import it into Pagecord. 
 
The time was 7:55pm.

Writing an importer would have easily taken me a day of frustrating head scratching and endless bugs. Instead I prompted the Claude Sonnet 4 agent in VS Code to do it for me:

You are to help me import posts to a Pagecord blog.

* The data is in a folder with the following structure:

./entries
./images
./userpics

* entries consists of many HTML files
* the html files reference images and other entries using "..", e.g. "../entries/" and "../userpics".

Your job is to write a script that:
    

  1. takes a root folder of the export
    2. 

  2. parse each of the HTML files in the "./entries" folder
    3. 

For each post:
    

  1. extract the post title from <h3 class="entry-title">
    2. 

  2. extract the post date from a <span class="datetime"> that follows the <h3> (both have a parebt <div class="header">)
    3. 

  3. extract the content of the post from <div class="entry-content">
    4. 

  4. if the content contains images that are relative, load the image and save it in the content as an embedded actiontext image attachment
    5. 

  5. extracts a tag list from <div class="tag"><ul><li>. Each li is a tag. Sometimes the tags are hierarchical, e.g. "hobbies: gaming". In this situation, just use the last part (e. g "gaming").
    6. 

  6. set post.tag_list to an array of tag strings if tags are present
    7. 

  7. save the post
    8. 

Write the script in Ruby such that it can be run with the rails runner and save it in the /scripts folder.

Claude was outstanding! It took it around 3 minutes to write a script, then it wrote some tests to verify what it had done, then it ran the script against the test data that I supplied. It pretty much worked first time, just one bug with importing images. A few more prompts to fix those and I had fully imported the blog to my local environment. One of the posts didn't import which turned out to be a bug in Pagecord, which I fixed myself (Claude made a meal of this), then I did another run through and it worked perfectly.

I deployed it to production, ran the script which worked perfectly, then I emailed the customer to tell them the good news.

It was 8:41pm. The whole thing took 46 minutes. I didn't even look at the script it produced 😅

Of course this isn't some magic, scalable customer support solution. It's just a real-world example of how effective these AI coding agents can be in boring, painful, burdensome situations like this. It really is remarkable!