AI-enabled CMSs: the content team’s holy grail?

Our development team has had its workflows disrupted multiple times in the past two years. Each time the change felt like a notable leap, until the next one arrived which dwarfed it.

The marketing teams we build for haven't (yet) had the same ride. Many of them use AI daily to draft copy, brainstorm campaigns and summarize research. But when it's time to turn that copy into a page on their website, they log into a CMS that looks and works about the same as it did five years ago. 

That gap is what I've been thinking about most lately, and I believe closing it will define the coming years of further CMS evolutions.

What AI changed for our team

Agentic coding. We’ve long since moved past AI as autocomplete. We've built systems that use agents to build sites the way our team does and follow our best practices across five different CMS approaches (so far). Those systems build out atomic components, scaffold CMS structure and populate and migrate content. The reduction in effort hasn't been incremental. Each iteration has felt like a generational jump, and the quality of the output has climbed as well.

Design to code. We've developed repeatable flows for turning finished designs into usable components and deployable CMS platforms. The time between design and specification handoff to site launch used to be where the bulk of project effort and timeline went. Now it's often one of the shorter parts of a project.

Quality at scale. We’ve built tools that quantify content and measure its relative quality, evaluate accessibility and review code proactively. Our internal and external documentation has leveled up too, which, if you read my last post, you know I care about.

Marketers deserve the same leap

So why haven't marketers felt it yet? Because AI-driven content management is an emerging space, while development revolutions have already been realized.

The bottleneck for marketers—beyond internal alignment—is the friction between defining copy and imagery, and a finished on-brand page that checks the laundry list of checkboxes needed for a modern site that surfaces and performs well both with end-users and AI agents. Assembling blocks by hand, deciding what blocks make sense in what order and why, porting content from a static doc to individual modules and seeking dev help if you run into a question or a blocker—all of these are notable friction. Then there’s always long-term brand drift: a slightly different button needed here, a hero image cropped differently there, a one-off section built for a launch that never went away and multiple content editors over time. Three years later, the site looks like it was designed by committee, because in a way it was.

That's the problem we want to solve with the same seriousness we brought to our own dev processes.

What I do on my personal sites

Outside of Alloy, I've taken this AI-enabled content creation idea to every site I maintain. All of my personal websites are now statically generated, and I don't log into a CMS to change any of them. I effectively have moved to a “no CMS” approach for everything I personally manage. Instead of editing content in a CMS, I just describe the change in a Claude chat, review what it proposes in a browser and let an automated deployment process build a preview and ship it. I wrote about that workflow, including how much of it I can run from my phone, in an earlier post.

The reasons are simple:

• It's easier. A plain-language request replaces hunting through admin screens. "Update my homepage to include a link to the new side project I launched on example.com. Pull the project icon from the other site and note that the project was built using Astro and Claude." That's the whole starting instruction. From there: read, rework, review, repeat.

• Content and code come from the same conversation. AI drafts the words, builds the layout needed and wires it all up in one pass.

• There's less to maintain. No database, no plugins to patch, no admin login to secure. Static pages are fast and cheap to host.

• Everything is versioned. Every change is reviewable, and every change is reversible.

For me, it's the best website workflow I've ever had, but I wouldn't hand it to a marketing team.

Why that doesn't work for an editorial team

My setup works because I'm the entire team. I'm the designer, the developer, the editor and the approver. I can read a diff, I know which components exist and why, and if the design drifts, I'm the only one who has to live with it.

An editorial team doesn't work that way, and it shouldn't have to:

• Many hands, one site. Multiple editors working at once, across departments, with roles and permissions that define who can change what.

• Workflow, not branches. Drafts, approvals, reviews and scheduled publishing, in the terms marketers use rather than pull requests and merge conflicts.

• Content that lives beyond a single page. Reusable content, global elements, forms tied to a CRM, localized content and a media library, all managed without touching code.

• Guardrails that don't depend on expertise. When the person making a change can't read the code, the system has to catch what they can't.

AI site builders like Lovable and Claude Design run into the same wall. They can produce something striking in an afternoon, we use similar tools ourselves. But when content lives in generated code, every edit goes back through the generator or back to a developer. And a generated site still needs everything underneath the pixels that makes a site perform:

• SEO instrumentation. Metadata, canonical tags, XML sitemaps, redirects and structured data, all managed consistently as the site grows.

• AEO. Answer engine optimization is how a brand shows up in, and gets cited by, AI tools like ChatGPT, Gemini, Claude and Perplexity. It depends on clean semantic markup, schema and content structured so it can be quoted accurately.

• Measurement and standards. Analytics and tagging, accessibility and performance that work under heavy scale, both in traffic and content surface are—something built for long-term reliable use.

We watch the stakes play out daily. We’ve built an internal platform that monitors how our clients show up across a wide set of AI providers, paired with social listening across a wide set of online services, to understand which channels are shaping each client's narrative. It has made one thing very clear to us: how a site is built shapes how it gets described, but it’s just one part of a much larger brand story. A site that looks right but isn't instrumented can cause big visibility problems in exactly the places people are increasingly asking about your brand.

Generating a site is the easy part, but making it real, enabling content teams and iterating based on results over time is the real brand need.

The gap everyone wants closed

Put those two stories side by side and the gap becomes obvious. Marketers want the ease I have on my personal sites: describe the change, review it, publish. And they need everything a real CMS provides: editorial workflow, permissions, reusable content, instrumentation and design components that are governed rather than improvised.

AI-based editing, with full CMS capabilities and governed design components, is a goal many are trying to achieve from varying directions. Sophisticated clients are already using AI to explore concepts and generate prototypes. Many already see the potential but can't tell whether it will ever be practical for their real-world site, or how they'd get there and feel confident in the solution. More and more eyes will be pointing at this experience gap, and it's the one I expect our industry to spend a majority of time closing in the coming months and years.

Defining the holy grail

What does closing that gap look like? I believe it passes four tests.

1. It starts from what already works. A marketer should be able to say, "Build a landing page for our fall event, like the event page we launched in the spring, but with the following changes" and get a new page that follows the same proven structure, adapted to new content. Being able to use existing references beats a blank canvas every time.

2. It knows a consistent vocabulary. The site has a defined set of components (heroes, feature grids, testimonials, stat bands, forms) and the AI builds only with those—it also knows the alternative names a module might be called. There’s no inventing a new card style because the prompt was vague, just a governed atomic component system that underpins every visual element. If a page truly needs something new, that's an intentional component design conversation, not an AI improvisation.

3. It has judgment, not just access. Knowing which components exist isn't enough. The AI has to know when each one is and isn't appropriate: one hero per page, a stat band only when there are real numbers to show, testimonials only from approved quotes, a comparison table only when there's something to compare, CTAs that use intentional action words of the brand. Those rules have always lived in designers' heads and brand guidelines. The holy grail incorporates those details into the context of the AI assistant.

4. It handles publishing workflows. The platform has to enable human reviews before anything is published, and remove deployment friction in the same way done for content editing—with chat based flows and definable hard guardrails.

A site that passes all four gives marketers the leap our developers got: faster pages that are also better pages, because the system protects the design instead of the design depending on everyone's discipline.

Where we are today

We're working hard to crack this nut, and we've now completed our first two projects that let marketing teams create and edit content with AI, directly inside their CMS. One early lesson is the one we expected: the AI is only as good as the system it works within. Well-defined components and clear usage rules shape the output more than any prompt will, and ensure better first-pass results.

We're not alone in seeing the opportunity. CMS platforms are racing into this space, and I expect the options to continue to multiply. You should expect us at Alloy to be testing as many as we can get our hands on. I am eager to see the delight on our client’s faces as they have their capabilities open up and their content management friction drop off.

The search continues

We haven't found the holy grail yet, no one has. But the gap between where marketers are, and where they could be, is closing faster than many of us expected, and there's real value available today for teams who are nimble and are ready to jump into this new world of opportunity. 

If you're wondering what AI-enabled content management can look like, or what it would take to get there, let's talk.