Lee Robinson, who works at Vercel, spent $260 in AI coding agent fees to migrate Cursor’s marketing site away from Sanity, their headless CMS, to Markdown files. That number should unsettle anyone who builds or sells content management systems for a living. His reasoning: "With AI and coding agents, the cost of an abstraction has never been higher". He argued that a CMS gets in the way of AI coding agents.

Knut Melvær, who works at Sanity, the very CMS Lee abandoned, wrote a rebuttal worth reading. He pointed out that Lee hadn’t actually escaped the complexity of a CMS. Lee still ended up with content models, version control, and user permissions. He just moved them out of the CMS and distributed them across GitHub, Vercel, and custom scripts. That reframing was hard to unsee.

Meanwhile, the broader momentum is undeniable. Lovable, the AI-first website builder, went from zero to $200 million in annual recurring revenue in twelve months. Users prompt what they want and Lovable generates complete, production-ready applications.

Ask me again in two years, but today’s Lovable is not a CMS replacement. So the real question isn’t whether CMSes are becoming obsolete. It’s who they’re for.

Historically, the visible layer of a CMS, the page builders and content creation workflows, is where most people spend their time. But the invisible layer is what makes organizations trust the system: structured content models, permission systems, audit trails, web service APIs, caching layers, translation workflows, design systems, component libraries, regulatory compliance and more. A useful way to think about a CMS is that roughly 30 percent is visible layer and 70 percent is invisible layer.

For more than twenty years, the visible layer was where the work started. You started from a blank state – a page builder or a content form – then wrote the headline, picked an image, and arranged the layout. The visible layer was like the production floor.

AI changes this dynamic fundamentally. You can now prompt a landing page into existence in under a minute, or generate ten variations and pick the best one. The heavy lifting of content creation is moving to AI.

But AI gets you most of the way, not all the way. The headline is close but not quite right, or there is a claim buried in paragraph three that is technically wrong. Someone still needs to review, adjust, and approve the result.

So the visible layer still matters, but it serves a different purpose. It’s where humans set direction at the start and refine the result at the end. AI handles everything in between.

A page builder becomes a refinement tool rather than a production tool. You still need the full UI because someone has to review, adjust, and approve what AI generates. You can try to prompt all the way to the finish line, but for the last mile, many people will still prefer a UI.

What happens to the invisible layer? Its job shifts from "content management" to "context management". It provides what AI needs to do the job right: brand rules, compliance constraints, content relationships, approval workflows. The system becomes more powerful while requiring less manual configuration.

So my base case for the future of CMS is simple: AI handles eighty percent of the work. Humans handle the remaining twenty by setting direction at the start, and refining, approving, and taking responsibility at the end.

This is why Drupal is not standing still. We recently launched Drupal Canvas 1.0 and one of its top priorities for 2026 is maturing AI-driven page generation. As that work progresses, Canvas could become AI-first by default. Watching it come together has been one of the most exciting things I’ve worked on in years. We’re far from done, but the direction feels right.

Lee proved that a skilled developer with AI coding agents can rebuild a marketing site in a weekend for $260. That is genuinely remarkable. But it doesn’t prove that every organization will abandon their CMS.

CMSes have to evolve. They have to become a reliable foundation that both humans and AI agents can build on together. The visible layer shifts from where you create to where you refine. The invisible layer does more work but doesn’t disappear. Someone still has to direct the system and answer for it when things go wrong.

That is not a smaller role. It’s a different one.