Is AI Going to Kill CMS? We Accidentally Ran the Experiment

Here’s the thing about marketplaces: they don’t sit still. Requirements shift. Logic changes. What looks like a simple content portal today starts asking awkward questions about multi-vendor flows, custom taxonomies, and region-specific rules by next quarter. Most off-the-shelf CMS platforms are built for stability — which is great, until your product is built for change.

So we made an early call. Instead of inheriting someone else’s architecture, we’d define exactly what we needed and build to that. No adapting. No workarounds. No “we can almost do that with a plugin.”

And that decision shaped everything about how we used AI in the process.

One day. Core CMS. Done

Using Claude Opus, we built the core of the system in a single day. Not a prototype. Not a proof of concept. A working CMS, structured around the actual product logic, not borrowed from someone else’s assumptions.

The rest of the week went into client review, feedback rounds, and iteration. Without AI, the same cycle would have taken two to three weeks minimum. Probably more.

But speed wasn’t the interesting part.

What actually made this work was structured, deliberate AI-assisted development — knowing exactly what to define before the first prompt, which decisions to feed to the model, and which ones to keep firmly in human hands. AI is an exceptional executor. But execution without engineering judgment is just fast mistakes. The model doesn’t know your client’s market, their compliance constraints, or what this product needs to look like in eighteen months. Your team does.

That’s not a limitation of AI. That’s just the division of labor that actually works.

Where AI Went Beyond the Code

The system didn’t just use AI to build itself — it runs on AI too.

Content interaction inside the CMS is powered by a smart RAG solution, meaning the system doesn’t just store content, it understands it and responds to it contextually. Multi-language support across 10+ languages was handled through AI-generated translations from a master version, cutting content preparation time significantly. Day-to-day operational workflows around content management were also AI-assisted.

Looking for help to start building efficient AI solutions?

Contact our AI Center of Excellence

None of these is dramatic in isolation. Together, they quietly remove a surprising amount of manual work from the process — and make the whole system faster to iterate on.

So — is AI Killing CMS?

Probably not killing. But it is making the “build vs buy” conversation considerably more awkward for the “buy” side.

CMS platforms exist largely because building custom systems is expensive and slow. When that cost drops to a day, the default logic starts to wobble. Teams that would never have considered a custom build now have a genuinely viable alternative — one shaped around their product instead of the other way around.

What this doesn’t mean is that engineers become optional. If anything, the opposite.

AI accelerates what you put in front of it. Put vague in, get fast vague back. Put clear, well-structured thinking in — backed by people who understand systems, constraints, and trade-offs — and you get something that actually works. And more importantly, something that keeps working.

What We Actually Learned

The project worked not because AI is magic, but because the combination was right. Deliberate, judgment-led use of AI. A team that knew what to build and, just as importantly, what not to build. A tool that could execute quickly once those decisions were made.

The CMS we ended up with is simple, flexible, and built to evolve alongside the product. No inherited limitations. No platform ceilings to hit later. Just a system that does exactly what this product needs — and nothing it doesn’t.

That might be the most underrated outcome of AI in development. Not speed. Not cost. The ability to build precisely — and stop there.

Have you run into a similar build-vs-buy decision recently? Curious how teams are thinking about this now.

• #AI/ML