Can Claude by my CMS?
What is happening to marketing websites right now is about removing friction. Not cost, not novelty. The distance between having an idea and having it live is collapsing, and the requests marketers make have started to sound like this.
"Make me a landing page for our charity event tomorrow and draft all the content."
"Tell me which pages on the site are out of date based on this recent news article."
"Tell me the ten worst performing pages on my website." - "...now tell me what i can do to fix them" - "...fix them for me"
None of those are unreasonable things to want. All of them used to cost a meeting, a ticket, a developer, a sprint, a deployment and a person with credentials to three different systems.
Now they cost a prompt in Claude.
The build step stopped being the bottleneck
For twenty years the marketing website followed a fixed sequence. Someone had an idea. A designer made a picture of the idea. An agency or an internal team rebuilt that picture inside whatever platform the company had already bought. Then everyone spent two weeks arguing about why the live page did not match the picture.
That rebuild step stayed invisible because it was universal. You cannot see a tax you have always paid.
What changed is that the picture became functional. When a marketer describes a landing page to Claude and gets back a page that actually works, the second build stops being production and starts being duplication. Teams look at their own process and realize they have been building every idea twice.
What used to consume a development cycle, an authoring cycle, and a publishing cycle now takes a few minutes of prompting to produce a live page that is on brand, accessible, and compliant with whatever rules you gave the agent to work inside. A campaign idea on Tuesday morning is a live page Tuesday afternoon. No ticket, no queue, no waiting for a sprint that starts in eleven days.
The page is not the prize. The prize is the collapsed distance between having a hypothesis and having evidence about whether it was any good.
The AI first stack is genuinely good now
This is the part that surprises people. The tooling is not a compromise. It is better than what most enterprise teams were running two years ago.
A typical build looks like this:
Claude writes the components and the pages.
A modern framework like Astro or Next.js renders them. (invisible)
Git holds the history. (invisible)
Vercel or Cloudflare deploys on merge, usually in under a minute. (invisible)
Sanity holds the content so that copy lives in a structured document rather than inside markup. (invisible)
Every piece of that is fast, well documented, and cheap compared to the license you are replacing. Sites come out lighter because nothing is dragging along a page builder's runtime. Deploys are quick enough that shipping stops feeling like an event.
The freedom matters as much as the speed. Interactive tools are among the highest converting assets B2B marketing produces, and building one inside a template driven page builder usually ends with the tool crammed into an embed field that has a character limit. When the site is code your team owns, the tool is simply part of the page. The same goes for every small detail that used to die because it could not justify a line item on an invoice. A hover state that responds. A counter that animates. A calculator that gives a real answer instead of a gated PDF.
The connectors are the best part
Most teams assume integration is the obstacle. It stopped being one. Google Analytics, HubSpot, and Sanity all have MCP servers available today. Connecting one is a configuration file and an authorization step, and it takes an afternoon rather than a procurement cycle.
What matters is what happens once they are connected. An agent holding your analytics, your CRM, and your content layer at the same time no longer has to be told where to look. Ask which pages are underperforming and it queries GA4 for the traffic, checks Sanity for what the page actually says and when it last changed, looks at what converted in HubSpot, and comes back with a ranked list and a reason for each entry. Ask it to fix the top three and it already has everything it needs.
It connects the dots and then acts on your behalf. Every step where a person used to carry information from one system into another was friction, and most of those steps are now gone. That includes the chores nobody wakes up excited to do. Broken links, crawl errors, orphan pages, pages that used to rank and quietly stopped. An agent works that list overnight and hands you a summary in the morning, which beats the quarterly cleanup sprint that never gets scheduled.
What you keep
The headline compresses two different things into one word, and the distinction is worth ten seconds.
Claude is replacing the page builder. It is not replacing the content layer.
The page builder is the thing you were fighting. Drag handles, a class naming convention, a component model set up by an agency two agencies ago that nobody has fully understood since. That layer is genuinely being eaten, and good riddance.
The content layer is different. It is where a headline lives so that a person can change it at four in the afternoon without a deploy. If every copy tweak becomes a pull request, you did not give marketing speed. You moved the agency into your git history. This is why Sanity keeps coming up in these conversations. Its content is structured data rather than assembled pages, and it now exposes Agent Actions and an MCP server, so the same agent building your components can read the schema and write documents against it.
The other thing to keep is a rules file. Type scale, spacing, color, component behavior, and voice, written down once and read by the agent on every task. It takes an hour and it is the difference between forty consistent pages and forty dialects of your brand.
Practical takeaways
- Write your design system and brand rules into a file the agent reads on every task, before the first page gets built.
- Keep a structured content layer so a copy change is an edit and not a deploy.
- Connect your analytics, CRM, and CMS through MCP early. The compounding value is in what the agent can do once it sees all three.
- Build reusable sections and add each new one back to the library so pages become assembly rather than generation.
- Keep pull requests and previews. Speed with a rollback path is a capability.
Gartner expects sixty percent of brands to be using agentic AI for customer interactions by 2028. For marketing websites that future arrived early, and the teams already working this way are not going back. The build step is no longer the constraint. What you do with the time you just got back is the interesting question.
