AI can generate pages. But can it make them feel designed?

Been thinking about this a lot lately.

Code gen has gotten really good. You can ask an AI to make a landing page, a product site, even a full app flow, and it’ll give you something usable fast.

But most of the time, it still feels like AI made it.

Not broken.
Not even ugly.

Just... flat.

The spacing is off. Type feels default. Hierarchy is technically there, but nothing really lands. Everything is “fine,” but nothing feels intentional.

That gap is basically what we’re working on with Lokuma.

We’re building a design layer for AI workflows - something agents can call when the output needs actual layout, typography, structure, and visual judgment, not just more generation.

Curious if other people here feel the same:

When you use Cursor / Claude Code / Codex / whatever, where does the output usually fall apart for you?

For me, it’s rarely the raw code now. It’s the last 20% of design quality that still takes too much manual cleanup.

Would love to hear where others hit that wall.

Mu