Technical background: Each agent loop begins with a real Astro project template pre-copied into the workspace. Rather than generating a file tree from scratch, Claude extends and modifies this known-good baseline. This design decision avoids the inefficiency and inconsistency of model-generated scaffolding. By working within a prebuilt, compilable structure, the model can focus on high-value tasks such as copywriting, layout, and visual hierarchy. Full architecture explanation on agent loop, Postgres-as-queue, edit-via-resume, and more can be found here: https://www.adamhsn.com/blog/conversational-website-builder-...

a couple of questions: how does it generate a preview of new changes - think homepage updated, blog post preview when user wants to publish content?

How is it different from chatting with Lovable or Claude directly?