There are tasks in every project that you know you have to do but that you dread doing. Not because they are hard. Because they are painfully repetitive.
For me, on a recent client project, that task was uploading 100 posts to a WordPress website. Each post had its own fields: title, URL, date, writer, content. All of it had to be entered manually, field by field, post by post. No way around it.
Or so I thought.
The setup
The content lived in a Word document. Everything was there, just not in a format that was easy to work with at scale. Before doing anything, I did a quick planning step that made the whole workflow possible.
I use Claude. I gave it the document and asked it to read through everything and organize it into a table, one column per field, one row per post. It did exactly that. Suddenly I could see all 100 posts at a glance, every field mapped out cleanly, instead of scrolling endlessly through a document trying to find where one post ended and the next began.
That table became the source of truth for everything that followed.
Teaching the tool
Before handing anything off, I made sure WordPress was ready. I created all the custom fields the posts would need: title, URL, date, writer, content. Once the environment was set up, I gave Claude the URL and told it to open the site.
This is where it gets interesting. Because Cowork can drive your browser, the tool literally opened a tab, read the page, and started figuring out how to map each field in the table to the corresponding field on the site.
I did not just let it run loose from the start. I had it do one post first while I watched. That trial run was where I caught a small issue: it was trying to use the date picker UI to enter dates, which was slow and clunky. I told it to just type the date directly into the text field in the right format. Faster, simpler, done.
After a couple of rounds of tweaks like that, it was working smoothly. I told it to do the next 20 posts.
The result
100 posts uploaded in about two hours. Done in the background while I kept building other parts of the site on another browser window.
Would it have taken me longer? Yes. Four to six hours, at least. That is not a guess, that is just the math of copy, paste, click, repeat, 100 times.
The last step was a verification pass. Instead of just asking Claude to confirm the table was accurate, I had it go back to the original source document and compare every field on every post. It found four mistakes it had made along the way. I asked it to fix them. It did, immediately.
What I took from this
This is not about AI replacing the work. It is about being honest with yourself about which parts of your work actually need you.
Data entry does not need me. Reviewing the result does. Building the site does. Thinking through the structure and the client relationship does. So that is what I did with the time.
The other thing worth noting: I did not install a single plugin. No import tool, no custom script, no third-party integration. The tool just used the browser the same way a person would. That was the cleanest part of the whole thing.
If you are a web designer or anyone who deals with repetitive content work, start looking at your workflow differently. Not “what can AI do” but “what in my process should I not be doing at all.” That is where the leverage is.
This post is based on the episode 05 of Pixelados, my podcast about design, web, and working smarter. Listen wherever you get your podcasts.