We Tested the Native AI in Bricks 2.4: What It Does Well, Where It Falls Short

Bricks 2.4 is a big release. For the first time, Bricks Builder ships its own AI integration: around 145 "abilities" that let an AI assistant read and edit your pages, plus support for MCP, the protocol that connects AI tools like Claude to other software. If that sentence sounded like alphabet soup, read our plain-English guide to MCP first and come back.

We build a competing AI product for Bricks, so you should read this test with that in mind. That is exactly why we are showing you the actual output instead of asking you to trust adjectives. Here is what happened.

The test: one hero section, one honest look

The setup: Bricks 2.4 beta on a real WordPress install, connected to Claude Desktop through the official MCP adapter. The prompt: a hero section for a premium travel agency, dark theme. Nothing tricky. This is the bread and butter of AI section generation.

Claude talked to Bricks, Bricks created the elements, and a few seconds later there was a finished hero on the page. Then we opened the JSON to see what it had actually built.

What it did well (credit where due)

• Clean, reusable classes. It created proper global classes with consistent BEM-style naming instead of dumping every style onto element IDs. That is the habit most human beginners take months to learn.
• Real responsive behavior. The headline scaled across four breakpoints, from 56px on desktop down to 32px on the smallest screens. Nothing overflowed, nothing broke.
• Correct overlay technique. The dark gradient was applied as an overlay on the section, which is the right pattern rather than a hacked background image.
• Readable contrast. Text over the dark background stayed legible. The hierarchy of headline, subtext and buttons was clean.

Let's be clear: for a one-off section on a fresh site, this output is usable in production. If anyone tells you the native Bricks AI produces garbage, they are selling you something.

Where it fell short (and why it matters)

1. Every single color was hardcoded

The generated section used colors like #2563eb, #0f172a, #93c5fd and #cbd5e1. If you recognize those, it's because they are Tailwind's default blue and slate palette: the statistically most common colors in the AI's training data, with no relation to the site whatsoever.

More importantly: there was not a single CSS variable in the entire output. No var(--primary), nothing connected to the site's palette. Why should you care? Because when the brand color changes (and it always changes), this section will not follow. You will hunt down hardcoded hex values by hand, in every section the AI ever generated. On one page that's annoying. Across a whole site, it's the maintenance nightmare that design systems exist to prevent.

2. The buttons went nowhere

Both call-to-action buttons were link elements with no href. They looked perfect and did nothing when clicked. Easy to fix, easy to miss, and the kind of detail that quietly costs conversions when nobody notices for two weeks.

3. Zero framework awareness

If your site uses a CSS framework like Automatic.css, Core Framework, or Advanced Themer, the native AI does not know and does not ask. It writes raw values where your framework expects its own variables and spacing scale. The section works, but it lives outside your system: a one-off island you now maintain separately. (If frameworks are new to you, here's our plain-English comparison.)

4. No memory between sessions

The native flow has no concept of a project brief. Your business, your tone, your palette, your spacing preferences: you re-explain all of it in every conversation, and any detail you forget gets invented for you. The output quality depends heavily on how much context you re-type each time.

What this means if you're not a developer

The short version: Bricks 2.4 gives the AI a very capable pair of hands. It can create, edit, and delete almost anything in Bricks, and it even saves a restorable revision when it edits pages and elements, which is a genuinely great safety net. What it does not have is a brain that knows your site: no design system awareness, no framework knowledge, no memory, no quality checks on what the AI produces.

That gap is exactly where we decided to keep building. Bricksfusion Studio generates sections that use your variables and your framework's spacing, runs every generation through code-level quality checks (including WCAG contrast verification), and remembers your project brief between sessions. We wrote a side-by-side comparison of both approaches, including the places where the native MCP is simply better.

Bottom line

Should you try the native Bricks 2.4 AI? If you are comfortable with technical setup, yes, absolutely. It is free with your license, impressive for one-off sections, and it will only get better. Just go in knowing what it does not do: respect your design system, know your framework, or remember anything about your project. Check every color, every link, and every value it writes. Or put a tool in the middle whose whole job is doing that for you.