How the SaaSpocalypse Ended DXP as a Category

Per CMSWire, the DXP category is officially over. Called it!

I don’t expect anyone to credit my RIP DXP dot com microsite, which I can now retire. It’s not like I was the only one pointing out the category contradictions and incoherence. I also didn’t foretell exactly how or when the end would come, which would have been a neater trick, vs. just calling an overall industry trend. Still, in the words of TV character Harper Stern, “I was proven fsking right!”

Vindication feels good, but the real question is what happens next. That’s what this post will get into. However! Am I going to permit myself a short victory lap? Absolutely.

The Emperor Had No Clothes — But It Took AI To Point That Out

The DXP category was always an analyst projection, helping to prop up a number of big enterprise software brands for nearly a decade. Over the past year, the force of AI rushing onto the scene blew the whole house of cards to pieces. Here’s how it played out.

These “platforms” were Frankensoftware — sewn together piles of acquisitions. The integrations were often weak, sometimes non-existent. The seams were large. They were sold on a great vision of value, but buyers struggled to operationalize and implement. So much shelfware.

That was always a source of frustration, but as AI changed enterprise expectations when it comes to implementing and operating software, it became increasingly unpalatable. Nobody wants to make a big multi-year investment in a legacy way of doing things.

These companies had already stopped growing, but now they’re all staring down real decline. Many are burdened with heavy debt, both financial and technical. They’re not startups in danger of going splat, but they are stuck in a kind of zombie mode as their ability to innovate diminishes, and with it their relevance.

While the vendors may live on for years, the category itself is done. And some of the big names that first put it on the map might be more vulnerable than you would think. 

What’s Going On With SaaSpocalypse Now

This macro trend is bigger than the DXPs: it’s an endemic problem in the B2B SaaS universe. An underlying driver of the “SaaSpocalypse” narrative is how unwieldy enterprise software suites have become. You can see it with Salesforce “going headless” — AI is just a better UI for these massive, amalgamated products.

Before AI, the big DXP players were already facing headwinds from next-gen “composable” players. Net new implementations for legacy technologies had slowed to a trickle as of several years ago, but next-gen players were gaining ground very slowly. It felt like a war of attrition.

The composable alternatives actually had a similar burden to the complexity of the all-in-one DXP. Going with a multi-vendor microservice/API approach shifted the risk from something bought to something built, which was ultimately not much more appealing to the broader market.

This is why prior to AI you saw headless players start buying heads, composable MACH darlings quietly assembling suites, and so forth. It was going to be a long, slow process to fight through another cycle of enterprise software development. 

But then AI upended everyone’s assumptions. The sudden abundance of agentic helpers changed the conversation in the market. What felt like slow enterprise decline shifted into something more like an existential crisis. This was the killing blow for the category. 

The DXP dream remains a good one: consistent, integrated, fast-moving digital experiences, same as it ever was. But in a world where you can just ask Claude to make your website — even if that might not be the best idea long-term — the old paradigms literally stopped making sense.

The Race is On for Real AI Value 

AI changes the unit economics of creating for the web. A broader segment of people and agents can produce more, in parallel, at greater speed. But the hard part is making sure what gets created hits the mark at scale: on brand, secure, accessible, performant, and ready for production.

The category that matters isn’t a laundry list of features; it’s an operating model that turns this new abundance into dependable outcomes. How this should work is a live question in the market, with real stakes.

Rubbing AI on aging frankensoft platforms is going to fail. It doesn’t matter if they pivot away from DXP as a label; I still think it’s RIP for most of them. It might take a few more years for the zombies to stop shambling, but eventually they will, and even Adobe’s future prospects are in doubt given the market for their core creative suite is suddenly in freefall.

I also don’t think that AI will make “composable” stacks sustainable. It certainly alters the cost of initial implementation, but when what’s delivered is still a bespoke in-house solution, you’re one staff departure away from that becoming a burning platform.

Finally, there’s loads of chatter in the market about people eschewing any purpose-built tools at all, moving directly to vibe-coding. “Claude is my new CMS” is a bold statement, certainly the flashy thing of the moment, but it’s already got a backlash brewing.

In the AI era, everything is up for grabs. At Pantheon, we have an opinion about the winning formula.

Move Fast, Stay In Control

While LLMs have made code abundant, the fact that they’re unpredictable makes them capricious. AI can produce different results from the same request; on a high-stakes site, consistency matters. What pops as a prototype might prove much more problematic in production.

When it comes to the web, if you’re a small business trying to get your first online presence established, or even a corporate customer in need of a fast (but also disposable) microsite, you’re probably pretty flexible on details. Claude might do as well as or better than a low-budget freelancer. Great.

But if you’re putting out assets that carry a known brand across a larger surface area, or if that microsite could still open up a liability concern, it turns out the fine print matters. 

Anyone who has worked with AI on creative projects, sweating the details, is familiar with the frustration of fine-tuning by prompt. Because what LLMs do is inhale and exhale blobs of text, absent a larger governing structure, they are going to take in the entirety of the project for even a simple copy edit. As the complexity of the artifact grows, the accuracy of these operations goes down while the costs increase.

The other thing about using vibe-code to make a site is that what’s in that code itself might surprise you, and not in a good way. IT organizations are playing catch up with the fact that rolling out AI meant every employee suddenly has a junior developer on demand. The security and safety practices necessary to make this tenable for high-visibility, public-facing projects are largely lacking within most orgs.

The best software engineering organizations in the world are just beginning to figure out how to run effective “software factories” using LLMs to safely and scalably generate code. We believe many of these practices will translate to the broader use cases of the web and digital experiences.
We’re going to see a rise of increasingly opinionated harnesses for LLMs with approved access to enterprise data that bring agents and people together using the following:

• Shared context: business goals, brand guidelines, and other grounding truths.
• Pre-approved inputs and ingredients: design systems, maintainable frameworks, approved enterprise data sources.
• Standards for output: security, accessibility, performance.
• Accountability: every change attributed, including what came from a human vs. an agent.
• A reversible path to production: full preview and the ability to undo become industry standard.

The answer is not to slow down creation. It is to make the right way the easy way: give people and agents shared context, clear guardrails, and a visible route from idea to production. That is how web teams move fast and stay in control—and how the sites that matter keep evolving as the pace of change increases.

If you haven’t already guessed it, that’s what we’re building. Sign up for the P1 Beta now and be among the first to see it firsthand.