AI builder credits do not seem like a big deal when you first start using them.
You type something. The AI builds it. You type something else. It changes it.
It feels almost like magic.
Then you get deep enough into a project that the prompts stop being fun experiments and start becoming the way you maintain the damn thing.
That is when credits start feeling very different.
There is more than one kind of AI builder cost
This is worth understanding because “I ran out of credits” can mean different things depending on the platform.
Lovable is a good example because its documentation separates subscription credits used for building and prompting from a usage-based Cloud and AI balance used for hosting and in-app AI features.
Those are two different meters.
Lovable also says that if the Cloud and AI balance runs out, deployed apps may stop working until funds are added or the free usage allowance resets.
You can read Lovable’s current billing documentation here.
I actually like that they document this clearly. My issue is not that there is a secret charge. My issue is what this model feels like once you build something you depend on.
Every fix can become another prompt
Imagine you have a button that is not working correctly.
You ask the AI to fix it.
It changes something else.
You explain what happened.
It tries again.
Now the original button works but something on mobile looks wrong.
Another prompt.
This is normal software development. Developers have always spent time fixing one thing and discovering another.
The difference is that if you do not code, your ability to participate in that process may be tied directly to the AI interaction.
That is the part people should think about before choosing a platform.
Source code access does not magically make you a developer
Some people answer this problem by saying you can export the code or connect the project to GitHub.
That is important. Portability is good.
But access to source code and the ability to confidently maintain source code are not the same thing.
If somebody chose an AI builder specifically because they could not build the application themselves, handing them the code when there is a problem does not suddenly solve that original problem.
This is why I think non-developers need to consider maintenance before they get hypnotized by how quickly the first version appears.
This was one reason I went back to WordPress
I wrote about my own numbers in why I stopped spending around $600 a month on AI builders.
WordPress has costs too.
Hosting costs money. Premium plugins cost money. Developers cost money when you need them.
I am not pretending WordPress runs on fairy dust.
The difference for me is how many options I have when something goes wrong.
I can deactivate a plugin. Replace it. Restore a backup. Edit the files. Move the site. Hire somebody. Ask AI for help. Use a different AI.
There are multiple ways out.
I would ask three questions before using an AI builder
First, what happens when I stop paying?
Second, what happens when the AI cannot fix the problem?
Third, can I realistically maintain what I am building outside the original builder?
If the answers are acceptable, build the thing.
I am not trying to scare anybody away from these platforms. I still use AI builders where they make sense.
I just think the cost of creating version one is the wrong number to obsess over.
The interesting number is what it costs you to keep version 37 running six months later.
AI does not have to own the platform
This is what pushed me toward Marketur MCP.
I wanted the AI to operate WordPress instead of needing the entire website to live inside an AI builder.
WordPress now has its own official Abilities API and MCP Adapter work designed around giving external systems and AI agents controlled access to WordPress functionality.
The official WordPress Abilities API documentation is here.
That direction makes a lot of sense to me.
I still get to talk to an AI.
I still get the speed.
But the website underneath the conversation is WordPress.
That is a trade I am much more comfortable making.