The biggest challenges after a website launch rarely stem from the design or the technology—they arise afterward, in day-to-day operations. A new blog post, a new showcase project, an updated team profile: Someone opens the CMS, finds the right template, fills in the fields, checks the references, and publishes. For us, this step now works very differently.
A link and a few lines
We have recently started using our own agent infrastructure, which is based on the Model Context Protocol (MCP). Through the Gridonic Engine MCP – our internal orchestration layer – and DatoMCP Or Statamic MCP—our integration with various CMS platforms—allows us to create new content not just through the traditional CMS interface, but via conversation, just as we integrate with AI tools today. I type, “Create a new project, make it like this one /project-x, and publish.” The agent recognizes the content model, structure, and reference logic of the template—and creates the new record, complete with text content and images, neatly linked to the correct categories, authors, or related projects.
That sounds simple, but it changes something fundamental: Editing websites becomes much easier and faster. Plus, I don't have to know the entire project—what's linked to what and where each image is located. The agent knows the project inside and out and takes care of these tasks.
Why this works—and why it's no coincidence
MCP standardizes how AI models interact with real, live systems—not just through text, but through APIs, databases, and CMS structures. This is the same idea we’ve already implemented in "AI in the Web Development Workflow" As described above: AI speeds up execution, while architecture, content modeling, and quality assurance remain the responsibility of humans.
This is only possible because the foundation is right. In "Why We Rely on DatoCMS at Gridonic" We’ve explained why we rely on a structured, API-first headless CMS: clear content models, clean references between entries, and reliable multilingual support. It is precisely this structure that enables an agent to reliably “understand” what a new entry “like this one” means. Without a well-thought-out model in the background, any AI automation remains piecemeal—but with a clean data model, it becomes a true efficiency gain.
Our perspective on Design Systems and Component-Driven Design Here's the key: The more consistently components, tokens, and content structures are built, the more reliably AI systems can derive new, accurate instances from them—whether in the UI or the CMS.
My Own Medicine, Prescribed by Myself
This approach isn’t just something we recommend—we use it ourselves every day. Instead of relying on a single, all-purpose AI that’s supposed to do everything at once, we delegate tasks to specialized agents with clearly defined responsibilities. One agent specifically reviews new and existing content for SEO structure and discoverability. Another handles proofreading—checking texts for spelling, grammar, and style before a post goes live. A third continuously monitors a website’s performance and suggests specific optimizations, from image sizes to load times. Each agent works within their area of expertise, with clear responsibilities—much like a good team of specialists rather than a single person who tries to cover everything inadequately. This reduces errors, keeps responsibilities clear, and makes the system transparent rather than opaque.
What this means for our customers
For editorial teams, this means less time spent learning the CMS, fewer click paths, and fewer sources of error for recurring content types—team profiles, case studies, blog posts, and product pages. At the same time, control over content and structure remains where it belongs: with the people who approve content and ensure quality. This aligns with what we’ve seen in "The Agent-Ready Web" As we have already described in relation to the public image of websites: Structured, clearly organized content is not only easier for editorial teams to maintain, but is also reliably readable by AI agents—whether internally during publication or externally when read by ChatGPT, Gemini, or Perplexity.
For us, design, technology, and now operational AI workflows aren’t separate entities—they’re part of a system that builds on one another. If you want a website that not only looks good but is also easy to manage on a day-to-day basis, these three levels need to be understood together from the very beginning. Our Gridonic Engine is LLM-agnostic, meaning it works with all models.