In Four Days, from Automated Publishing to Letting an AI Agent Manage an Entire WordPress Site
- From automated publishing to managing an entire WordPress site
- By the time I woke up, many website problems had been fixed
- AI agents feel like the steam engine revolution for information technology
- A new website project that once seemed distant is now getting started
- Now I can do many things at once
- Looking back at 2019: it has only been seven years
- This is the way we should be working today
Only four days ago, I tried using Codex to publish WordPress articles automatically, bringing my long-dormant personal blog back to life.
And now, things have taken off even further.
From automated publishing to managing an entire WordPress site
I quickly realized: if it can publish articles automatically, what if I give Codex access to the whole WordPress website folder and explain how to read and write MySQL? Couldn’t it then help me manage the entire site directly?
By management, I don’t just mean publishing articles.
I also mean adjusting the site’s theme and layout, modifying or adding plugins directly, checking articles for dead links, and identifying content that could be improved or revised.
Still, I was cautious about letting Codex work directly with the WordPress folder and MySQL.
So before starting this experiment, the first thing I did was back up the WordPress data, before asking it to modify the website.
First, confirm that the backup can actually be restored
I used to handle this manually myself, but now I leave it entirely to Codex.
Even so, before starting the work, I asked Codex to design a rigorous backup and restore testing procedure.
It had to confirm that the backup could be restored successfully and that the whole mechanism worked before making any changes to the site.
Codex handled the task very well.
Once the backup and restore checks passed, I had it thoroughly inspect another long-established information website that I run.
It quickly produced a detailed site audit, listing the problems that needed attention and how to address them.
Next, I asked Codex to manage those repair tasks in an SQLite database.
This also reflects something I now feel strongly about AI agents.
Codex is highly proficient with Python and can build almost any tool it needs.
Need to organize data? It can write Python.
Need a database to manage the work? It can use SQLite.
Need to scan a website? It can write its own program.
It can handle checking articles in batches too.
Within just a few hours—I even went to sleep in the meantime—it had fixed most of the article content issues.
By the time I woke up, many website problems had been fixed
After waking up, I continued repairing the old website and started adjusting my personal blog at the same time.
Yes, the productivity gains once Codex has access to the WordPress folder are remarkable.
I wasn’t happy with my blog’s layout, so I told Codex I wanted a look and feel resembling Apple’s dark theme.
It started making adjustments immediately.
I then pointed out a few more things to change. Within minutes, it handled them all using a WordPress child theme.
It feels like the steam engine revolution for information technology!
AI agents feel like the steam engine revolution for information technology
The steam engine brought enormous change as people gradually mechanized and automated actions that had previously required manual control, ultimately turning steam into a sustained source of power.
I get a very similar feeling when I look at AI agents today.
AI agents are undoubtedly a new wave of revolution.
And our world now depends heavily on computers and information processing.
Website management, data organization, content checks, code changes, reports, databases, and all kinds of administrative operations: almost all of these tasks can gradually be handed over to AI agents.
I would even say with confidence that most consultants offering website management and maintenance already struggle to match AI agents in expertise and execution speed across many tasks.
Codex’s efficiency is astonishing.
And it costs only around US$20 a month.
My goodness!
I used to spend NT$20,000–30,000 a month hiring part-time help to handle a large amount of tedious website work.
Auditing and revising the content of a single website could take months.
Now, many of those things can be handled in minutes.
A new website project that once seemed distant is now getting started
Next, I began reconsidering a new website project that had previously seemed far off.
I have now decided to build the new website entirely with Codex and Claude Code.
The process is simple.
As usual, I start with several rounds of conversation in ChatGPT on the web, gradually clarifying the website’s requirements, data structure, research direction, and other details until they form a complete work order.
Then I hand the work order to an AI agent to carry out.
In fact, I already have some proven website topics and advertising monetization approaches that still work today.
For example, a website I built for Japan was generating steady income within a year of launch.
But replicating that model in other countries used to require one or two months, or even longer, of data collection, research, and planning.
And that was only the preparation.
The actual work used to take as long as a year
Once the work began, I still had to write VB.NET and Python code myself, define SQL database tables and fields, and organize the source data using VBA, Python, or VB.NET.
A website could take a year to go from initial planning to actually launching.
During that entire year, I could hardly work on many other projects at the same time.
Now I can do many things at once
With help from AI agents, things are completely different now.

I can make games at the same time.
Repair old websites at the same time.
Develop new websites at the same time.
And keep adding articles and features to my personal blog.
Oh, and run an e-commerce business at the same time too.
Of course, there are also plenty of everyday chores.
Those everyday matters take up at least half of my time.
And I do all of this on my own.
There is no team.
The startup costs for these projects are also very low.
A new domain costs only around NT$300.
I use computers I already own.
Starting another project doesn’t mean I need another AI agent subscription.
Doing this many things at once used to be unimaginable.
Now it is actually happening.
Looking back at 2019: it has only been seven years
I found myself thinking about 2019 again.
That was when I first started learning WordPress and built my first website.
I stopped using that website long ago, but I have always kept it around.
After all, maintaining a website costs very little.
Looking back now, I am truly amazed.
It has only been seven years.
Seven years ago, running an independent media website meant dealing with many technical problems myself.
How to set up the site, use WordPress, modify themes, work with plugins, organize data, design databases, and write code—I had to solve all those problems, large and small, myself.
Beyond the technical problems, there was also a lot of tedious operational work.
Now, those problems have almost all disappeared.
At least, I no longer need to tackle each one myself.
AI agents have also virtually eliminated many computer tasks that used to be very tedious.
This is the way we should be working today
Clearly, most people have yet to truly integrate AI agents into their workflows.
Then again, our industry has always been a small group of people.
I believe people will soon start learning this way of working and quickly marvel at how powerful it is.
For me, AI agents are no longer something to perhaps explore in the future.
This is the way we should be working today。

