There has been much debate following Cursor's moving away from using a CMS, and Sanity's equally strong response. The discussion is stimulating, as it questions a product layer that has been a backbone of the web for twenty years. In my view, it points to a much broader debate about how content should be produced, governed, and operated within large organizations.

On one side, the "CMS-free" promise feels like fresh air for editors: AI dramatically reduces friction. Interfaces like CMS can disappear, content can be generated, queried, refactored through natural language. Incidentally it nicely fits Cursor's own narrative as SaaS-killer.

On the other side, Sanity's response is equally valid: most "CMS-free" approaches eventually recreate a CMS. Content at scale needs structure, governance, reuse, localization, workflows, and consistency. Markdown plus Git plus agents does not magically erase that reality.

My conviction is that the real friction doesn't come from the CMS itself—even if some are better designed than others—but from content governance in large organizations.

What works in a 200-person tech startup, with a simple product narrative, does not translate to a 100,000-person industrial group, operating across dozens of countries, brands, business lines, and tens of thousands of SKUs. Especially when the website is only a tiny fraction of the marketing surface, as content also feeds e-retailers, print, customer support, sales, regulations, and more.

My take is that a CMS should not be seen as a finished product, but rather as a framework, that is now to put into the hands of AI—one could probably say the same about the website itself, but that's another topic.

At Michelin, we've been fortunate to work for many years with an exceptional partner. Over time, Apostrophe has been the framework on which we rebuilt over 500 websites and apps, across multiple brands, business lines and fifty-plus countries. All other prominent solutions failed to fit that complexity—this one didn't.

Why? Because we never treated it as a boxed CMS. We treated it as a highly plastic and robust framework: first standalone, then headless, layering abstractions where needed—Svelte, Astro—contributing back to open source, and integrating it deeply into our stack.

Today, the rise of AI in digital experiences and software delivery workflows is a powerful opportunity to extend that same mindset into a new technological paradigm, and rethink its role inside a broader content generation and orchestration ecosystem, in a world that has clearly shifted.

Let me add a final analogy from our own industry: inside a tyre, there is an invisible metallic and textile structure. You don't see it, you don't touch it—but it's absolutely critical. Yes, the rubber absorbs the friction of the road. But it's the structure that allows the car to hold the curve, at speed, safely. Sometimes, the most important things are precisely the ones designed to disappear.

In response to this post