Context Is Gravity in an Expanding, Not Dying, Software Universe

My last issue on SaaS moats crumbling and the opportunity in context-as-a-service clearly hit a nerve. Lots of great comments, a mix of agreement, questions, and pushback. Thank you. This is exactly the kind of conversation our industry needs to be having. Bring on more ideas and rigorously stress-test the heck out of them.

Several questions and challenges came up repeatedly in the feedback. So let’s talk through this a bit more deeply…

First, let’s address the elephant wearing a “SaaS Is Dead” t-shirt: SaaS is not dead.

Zylo just released their 2026 SaaS Management Index, built on data from over 40 million SaaS licenses and $75 billion in spend. The average organization now manages 305 applications and spends $55.7 million annually on SaaS — up 8% year over year. Large enterprises are running portfolios north of 1,000 apps. Global SaaS spending is on track to blow past $300 billion.

But here’s what makes the data really interesting. The total portfolio count barely budged. What did change, dramatically, was the velocity underneath. Organizations are bringing on more than eight new tools every month, representing an annualized growth rate above 34%. They’re also churning out tools they’ve stopped using. The stack is in constant motion — a river of apps flowing in and out, not a stagnant lake.

AI is accelerating that motion. Spend on AI-native applications jumped 108% overall. For large enterprises? 393% in a single year.

The stack is growing, churning, and getting more expensive. That’s not the profile of a dying industry. But it is the profile of an industry in flux — which, for those of us thinking about platforms, is arguably more consequential than simple growth. A stack that turns over a third of its apps every year screams for a coordinating platform.

Several people noted that, despite Wall Street anxiety and LinkedIn doomsaying, many SaaS companies are performing quite well. Revenue is growing. Customers are renewing. The sky is not, in fact, falling.

They’re right, but I want to be clear about the timeline here.

In the Martech for 2026 report that Frans Riemersma and I published in December, we were explicit: we expect the majority of the existing martech stack to stay in place through 2026. AI is infiltrating the stack, but mostly as a complement to existing tools. Our research shows that 85.4% of marketing teams were using AI to enhance existing functionality. 42.7% were implementing entirely new capabilities with AI. Only 30.1% said they were replacing existing SaaS functionality with AI.

AI is operating in and above existing products right now. Honestly, it’s an early sketch of what context-as-a-service can look like, with AI tapping into the data, logic, and domain models embedded in these platforms. It’s CaaS in embryonic form, even if nobody’s calling it that yet.

But — and it’s a big but — things are changing. Investors aren’t wrong to push existing SaaS companies to articulate a vision for what will carry them into the next decade. The question isn’t “are you profitable in 2026?” It’s “show me how AI accelerates your business rather than commoditizes it.”

Instead of fighting the thundering hordes of a million new AI apps and agents, how can you harness that kinetic energy as a source of growth?

That’s ecosystem judo. It’s a very different strategy than feature-shipping your way to the next quarter.

Steven Sinofsky’s recent piece, “Death of Software. Nah.” walks through the historical transitions that should give any “software is over” pundit pause: in every major technology transition, the thing everyone predicted would die ended up becoming vastly larger than anyone imagined.

Sinofsky makes five predictions about AI that are worth internalizing. I’ll focus on two that directly reinforce the CaaS thesis.

First: there will be more software than ever before. Not just because AI makes software faster to build, but because we are nowhere near meeting the demand for what software can do. Every analog process, every domain that hasn’t yet been fully digitized, every interaction that could be smarter — it’s all runway for new software creation. More software being created means more moving parts to coordinate. More moving parts means more value in platforms that can bring coherence to the chaos.

Second: domain expertise will be wildly more important than it is today, because many domains will become vastly more sophisticated with AI. As domains become more sophisticated, the value of deep, structured, domain-specific context grows with it. General-purpose AI can power the intelligence. But someone has to provide the context: the governance rules, the taxonomies, the operational patterns, the institutional knowledge, etc. that makes AI outputs actually reliable in a specific professional domain. That’s what a CaaS platform can do masterfully.

Tomasz Tunguz wrote a provocative piece, “Can You Fly That Thing?,” built around The Matrix scene where Trinity downloads a helicopter pilot program directly into her brain. His core thesis: in the AI era, enterprises will provision capabilities rather than just applications. AI agent “skills” — programs written in English that tell an agent how to accomplish a task — are a pure embodiment of such capabilities.

It’s an elegant reframing. But Tunguz also raises a critical concern of trust and distribution risk. A recent analysis of nearly 5,000 AI repositories found malware embedded in them. When capabilities are distributed as code that agents execute autonomously, who’s vetting them? Who’s the trusted operator?

This is exactly where CaaS platforms can play a role. A domain-specific CaaS platform can serve as the trusted ecosystem of skills for its area of expertise — the “Tank” in Tunguz’s Matrix analogy, uploading verified capabilities into the agents that operate within its domain. It can also be the context provider that those skills tap into for the right data and the right tools at the right moment.

Something familiar is happening inside the enterprise tech stack, and Ben Thompson of Stratechery gave us the language for it a decade ago.

Thompson’s Aggregation Theory identified a fundamental shift created by the internet: when distribution costs drop to zero, power flows to aggregators who can mediate between abundant supply and consumer demand. Google aggregates web content. Meta aggregates social content. The supply is effectively infinite; the scarce resource is the curation, discovery, and context that connects supply to demand.

I think a parallel dynamic is emerging inside the enterprise.

The explosion of apps, agents, skills, automations, and data sources is creating an abundance of capability inside every company’s tech stack. The number of things a business can do with software is growing exponentially. But the ability to discover the right capability, connect it to the right data, apply it in the right context, and govern it responsibly? That remains scarce and is getting scarcer as the stack proliferates.

A CaaS platform can be the enterprise aggregator for its domain. The more apps, agents, and skills that connect to it, the more valuable it becomes to each of them and to the business overall. Every additional connection adds value to every existing connection — classic network effects. Not because the platform built all those capabilities itself (it never could keep up), but because it provides the contextual substrate that makes all of them coherent and trustworthy.

This is not about the aggregator hoarding features. It’s about harnessing the creative energy of hundreds or thousands of other builders — the ecosystem around the platform and within a company’s broader tech stack — in a way that makes their creations more valuable, helps the business get more value from them, and accrues durable value to the platform at the center.

If a CaaS platform is genuinely open with its data, tools, and context to accommodate a kaleidoscope of apps and agents, where’s its revenue model?

You could charge the apps and agents for API access. But I think that’s the wrong model. It creates friction with the very ecosystem you need to attract and grow. Every toll booth on the highway into your platform discourages a builder from choosing your road over the empty, toll-free alternative next door.

The real value of a CaaS platform accrues to the business running it — the company whose operations are orchestrated through that platform, whose data flows through it, whose agents and apps are coordinated by it. That’s the customer you should charge, and you should charge them based on the usage generated by all the services running through the platform.

This is how the data cloud companies operate. Databricks and Snowflake don’t charge the partner applications that run on their platforms. They charge the businesses using those applications based on compute and platform consumption. The more things you connect, the greater the consumption. But also, implicitly, the more value the customer is getting from them (or they would simply stop using them).

The incentives align beautifully: the platform, the ecosystem developers, and the customer all win when usage goes up.

There may still be seat-based revenue from the humans who directly interact with a CaaS platform. But given the explosion of agents and automations leveraging the context layer, seat-based pricing alone would undercount the value being delivered. You need a consumption dimension that scales with the growth of all usage.

Multiple people pushed back on my point about the UI layer being decoupled from the platform’s core value. “UI still matters!” (Thank you, designer subscribers.)

It does. And nothing I’ve said suggests that CaaS platforms won’t have a UI. Most will. Probably a good one. But here’s the shift: the platform’s UI won’t be the only interface through which people and agents engage with that context layer. That’s a meaningful change from how most SaaS platforms have operated, where the UI was the sole conduit to the data and functionality underneath.

And, hey, let’s be honest. Most SaaS UIs are not exactly what you’d call compelling differentiators. Sorry. People use them because they’ve had no other way to access the context inside the platform. Not because the screens sparked joy. The UI has been a conduit to value, rarely the value itself.

Will there be exceptional UIs that users genuinely choose over the many other options they’ll have? Absolutely. But they’ll be exceptional — as in, they’ll be the exceptions. The bar goes up considerably when your UI is competing not just with other SaaS products’ screens, but with conversational agents, AR overlays, and whatever weird-but-wonderful interfaces creative developers and motivated vibe coders dream up on top of your context layer.

That’s not a threat to the value of the platform. It’s a threat to platforms whose value proposition was “you have to look at our screens to get your work done.”

Not every software product needs to become a context-as-a-service platform. In fact, the vast majority won’t and shouldn’t.

Platform dynamics create strong forces of consolidation. Any given domain, depending on how you scope it, likely supports only a few winners at scale. Platforms succeed through virtuous cycles of network effects: every additional connection adds value to the platform and to everything else connected to it.

This is why I am so adamant that companies who want to win as a CaaS platform must openly embrace the ecosystem as a first-class priority of their engineering and go-to-market teams. You’re not going to win at scale by cranking out features. You’re going to win at scale by being the coordinating platform for thousands of apps and agents that are cranking out features in their areas of deep passion and expertise, all of which make your platform more valuable to customers and to those ecosystem developers.

For the winners, the prize is a large and durable business — potentially the kind that defines a category for a decade or more.

But what about that wonderfully diverse long tail of specialized apps and agents that aren’t trying to be platforms?

There will be very good, profitable businesses there. The cost of creating software is dropping every year, and the demand for specialized, domain-deep tools is only growing. But it’s going to be harder to build a very large business around a single specialized tool.

The more narrow your focus, the more depth of capability you can deliver and the more defensible that depth becomes against the ever-growing army of competitors that AI-accelerated development is unleashing. Go narrow and go deep. Find a platform ecosystem where your expertise is valued, and build there.

The future of software isn’t a monoculture. It’s an ecosystem. A few large CaaS platforms providing the context layer for their domains, surrounded by a thriving, specialized, long-tail economy of apps and agents that make those platforms (and their customers) more capable. Both the platform and the long tail can win. They just need to be honest about which game they’re playing.

I fully believe that collectively, software is going to grow and thrive like never before. More apps, more AI, more spend, more sophistication, more demand. The TAM for software is expanding faster than at any point in its history.

But individual software companies are going to face substantially more competition. The barrier to creating software has dropped to its lowest point ever, and it’s still falling. Existing SaaS platforms relying on their previous moats — features, data lock-in, you-need-our-UI-to-do-your-job — are going to find those moats are becoming puddles.

I truly believe the transformation from SaaS to CaaS is within reach for many of these companies. The domain expertise is already embedded in their platforms. The data and operational patterns are already there. The context that AI needs to be reliable and useful in their domains? They’ve been accumulating it for years. It’s a question of strategic choice — embracing openness, investing in ecosystem, rethinking the revenue model — more than technical feasibility.

The domain expertise is already embedded. The context is already there. What’s missing for most companies isn’t capability — it’s conviction.

The theories are interesting. But what’s actually happening in practice?

Frans Riemersma and I are running a survey for our next State of Martech 2026 report that asks about a full spectrum of marketing use cases: where marketers are using AI features within an existing SaaS tool, where they’ve brought on a new AI-native tool, where they’ve built their own AI solution, and where they aren’t doing anything with AI at all. This is the empirical counterpart to my futurecasting above. Real data on how the SaaS-to-AI transition is unfolding, use case by use case.

If you work in marketing or marketing technology, I’d love your input. The results will be shared freely and ungated with everyone in our State of Martech 2026 report.

[Take the survey →]

The more voices in the data, the clearer the picture. And if this whole CaaS thesis turns out to be the fevered ramblings of one crazy martech analyst — well, better to find that out with data than with vibes.

Never a dull moment in martech,

Scott