A Website Built with AI Still Has to Make Sense to the Next AI Session
A website built with AI still needs to make sense to the next AI session.
This is something I keep coming back to when thinking about how we build and hand over WordPress sites. I’ve worked with WordPress for many years, building websites, fixing projects left by other developers and working with different website builders. Now AI is becoming part of that work, and the questions go beyond how quickly it can create a page.
Key takeaways
- Every block, field and script an AI adds is a small decision. After hundreds of them, those decisions are the architecture of the site.
- The real test comes after the handover, when the client keeps working through AI in new chats that know nothing about how the site was built.
- The next session needs to know where content belongs, which blocks exist, what can be reused and which decisions were already made.
- A builder designed for AI can give every session the same understanding of the site, while still letting the agent write code.
Small decisions become architecture
Suppose you give AI access to your project and ask it to build a new section.
It creates a Gutenberg block. The page looks good. But can the client change the text and images through WordPress, or are they hardcoded in the template?
Tomorrow, you need a similar section with a different layout. Will AI reuse the existing block, or create another one that does almost the same thing?
Then you ask for a carousel. Will it notice the library already used in the project? And will it load the scripts only where the carousel appears?
Each decision seems small. But after hundreds of them, you have the architecture of your website.
The real test is the handover
You can review those decisions while building it. The harder question is what happens when you hand it over.
The client may want to keep working through AI: change images, publish content, add pages or manage data without opening the WordPress admin every time.
That means another AI session will need to understand the project. It needs to know:
- where content belongs;
- which blocks already exist;
- what can be reused;
- which decisions have already been made.
Otherwise, a simple request can introduce another way of doing something the site already handles.
In my opinion, architecture alone isn’t enough to keep a project consistent through many rounds of AI changes. The AI also has to keep the context of earlier changes: to the blocks, the structure, SEO and marketing. That lets the work go on as one long run, and reach big goals faster and better than many small runs that start from zero every time.
Maksym Verentsov, Head of Fabrement
Rules that stay with the site
This is where I see a useful role for a builder designed for AI. It can give the agent a consistent way to understand and work with the site, while still letting it write code. I wrote about why a builder’s real value for AI is its rules in Website Builders Were Made for People; the handover is where those rules matter most.
The client shouldn’t have to explain the architecture every time they open a new chat. Asking to add a page should be enough for the agent to find the existing blocks and follow the project’s rules.
Maksym is right that structure needs memory too, and we’ve already started building that into our blocks: every block version in Fabrement is saved with a short summary of what changed, so the next session can read the history. How to keep the same context for marketing and SEO is still an open question we’re thinking about.
Andrian Domashovets, in reply to Maksym
Andrian Domashovets is the tech lead at 7Fridays and Fabrement, where he builds the MCP server that lets AI agents build and run WordPress sites. Andrian on LinkedIn
Related articles
Website Builders Were Made for People. What Happens When AI Is the User?
Builders hide code behind buttons so people don't have to write it. AI can write the code, but not the judgment. Why…
Agents Took the Controls. The Interface Got a New Job.
When AI agents do the work, the interface stops being the controls and becomes the dashboard: it shows what is possible, shows…
WordPress MCP Adapter vs Fabrement
The official MCP Adapter is the bridge that lets an AI agent call what WordPress plugins register. Fabrement is a plugin with…