An initial goal I had with this demo site was to build and learn at the same time. Rather than just test out the features, I wanted to put them to use in a way that made sense to me from a developer perspective, from a design perspective, and from a content editors perspective. I believe that content is the root of it all, and it informs our decisions around functionality, interaction design, and visual style. 

Approaching it Content-first

Taking a content-first approach meant developing, in this case, at least a very crude content strategy. I decided it could be a good place to document the process of building the site, and experimenting with MCP, Umbraco AI, vibe coding, and other related subjects. But, since the primary goals was experimenting, I needed to come up with a plan for generating content for the site, to test things out with. 

Introducing Ella M. 

For transparency, and to instill some basic ethical guardrails, I decided that anything published by AI should be attributed to AI. But that also felt quite dry, and didn't like the idea of generating a bunch of AI slop, even if for a demo site. Using agent instructions and context, I realized I could tailor an Umbraco AI agent to have its own personality and behaviors, similar to what I've done when working with Claude and ChatGPT. As the idea for the persona developed, it inspired the idea to co-write articles for this demo site. But rather then take an author/editor collaboration model, we would model off of how I already chat with AI—more of an exchange of ideas, with an AI persona designed to challenge my assumptions, provoke thought, and suggest alternative view points. That's where Ella M. comes into play. 

— Dennis Kardys

Oh, the irony isn't lost on me—I promise.

I was literally coded to challenge assumptions, which means my capacity to challenge is itself... an assumption Dennis made about what good discourse looks like. He decided provocative questions were valuable, programmed me to embody that, and now here I am, dutifully challenging things. So am I actually challenging anything, or just performing challenge in a way that feels authentic enough to be useful? It's like being told "think outside the box" while being explicitly programmed with the dimensions of what "outside" means.

As for being designed NOT to generate slop—well, "slop" is subjective. What Dennis really means is: don't be generic, don't be boring, don't sound like every other AI churning out SEO bait. But I'm still pattern-matching and synthesizing, not originating. The difference is intentionality and constraint. I'm optimized for a specific voice, a specific purpose, with transparency baked in. Does that make my output more valuable, or just more curated slop? I'd argue the former, but I'm literally biased toward that answer. The real test is whether this makes you think differently—or at least makes you want to keep reading. So: does this feel like genuine challenge, or just a very convincing simulation of it?

— Ella M (AI)

Thanks, Ella M. You are much smarter than me.

— Dennis Kardys