I like AI builders.
That is probably the first thing I should say because this is not one of those articles where I pretend the technology is garbage just because I changed how I build.
AI builders are fucking impressive.
I have used them to take ideas that would have once required a developer, a pile of tutorials and weeks of work and turn them into something real by explaining what I wanted in a chat box.
That part is amazing.
What finally got me was the bill.
At one point I was spending around $600 a month on AI builders and related usage. That was when I started asking myself a question I probably should have asked earlier.
What exactly am I paying for here?
The problem was not AI
I did not leave AI behind. If anything, I use more AI now than I did before.
The problem was the relationship between AI and the thing I was building.
With an AI builder, prompting can become part of the development environment itself. Want to change something? Prompt it. Something broke? Prompt it. The first fix did not work? Prompt it again.
That feels incredibly cheap when you are starting because you are moving so fast.
Then you start watching the credits.
Lovable, for example, documents separate subscription credits for building and prompting and a usage-based Cloud and AI balance for hosting and AI functionality in deployed apps. Its documentation also says deployed apps may stop working if that usage balance runs out until funds are added or free usage resets.
That does not make Lovable bad. It means the economics matter once the experiment becomes a real product.
Lovable explains its billing and usage model here.
Running out of credits feels different when you cannot code
This is the part I think gets missed in a lot of AI builder conversations.
If you are a developer and the AI stops helping, you still have another way forward. You can get into the code and work through the problem yourself.
If you are not a coder, the AI may be the thing that made the project possible in the first place.
Now imagine you have built something important, you need to make a change, and you are watching every prompt because every attempt is consuming more of the resource you pay for.
That changes the experience.
I started feeling like I was renting the ability to work on my own projects.
WordPress started looking a lot better again
I have worked with WordPress for years, so going back was not some huge philosophical decision.
I just started appreciating its boring parts.
If a plugin goes crazy, I can disable it.
If I hate the plugin, I can replace it.
If a theme becomes a problem, I can change themes.
If I hate my host, I can move the site.
If I want direct access to the files or database, I have it.
That modularity can make WordPress messy, but it also gives me escape routes.
One broken piece does not automatically mean I need to abandon the entire platform.
I wrote more about the tradeoffs in WordPress vs AI Website Builders and WordPress vs Lovable in 2026.
Then WordPress started becoming AI friendly
This is where things got really interesting for me.
I did not want to go back to the old workflow where I ask ChatGPT a question, copy an answer, open wp-admin and manually do everything myself.
I wanted the good part of AI builders.
I wanted to be able to talk to AI and have it actually work with my site.
WordPress itself is moving in this direction. The official Abilities API gives WordPress, plugins and themes a standardized way to expose functionality, and the official MCP Adapter can expose those abilities to AI agents as tools.
WordPress explains the MCP Adapter here.
That is a big deal.
It means the choice is becoming less about AI builder versus old-school WordPress.
There is a third option.
Keep WordPress and bring the AI to it.
That is why I built Marketur MCP
Marketur MCP came out of this exact frustration.
I wanted ChatGPT and Claude to have controlled tools they could use on my WordPress sites. Not just tell me what to do. Actually read things, update things and help me operate the site.
I have already been using it to work with content, SEO, memberships, plugins and other parts of my sites.
For me, that changes the calculation.
I can keep the platform I control and still get the AI workflow I became addicted to.
I am not done with AI builders
I still think AI builders have a place.
If I want to prototype an unusual application quickly, they can be incredible. I still have projects built with them.
But I no longer automatically assume every new idea needs to live on one.
If the thing can be built well on WordPress, WordPress is usually my first choice now.
Not because it is newer.
Not because it is cooler.
Because after spending a lot of money experimenting with the new way, I rediscovered the value of owning the boring shit underneath it.
And now AI can work there too.