There is a question circulating in boardrooms, at investor conferences, and in the quieter moments between strategy sessions that few people are asking plainly. It goes something like this: if an AI agent can reason, execute, and close the loop on complex enterprise work without a human ever touching a screen — what, exactly, are we paying our SaaS vendors for?

It is a question worth sitting with. Not because the answer is simple, or because the outcome is predetermined — but because the forces now converging around agentic AI feel genuinely different from the waves of automation and digital transformation that came before. This time, what is at risk is not just efficiency. What is at risk is the fundamental reason many enterprise software products exist at all.

This is the first in a three-part series where I want to explore that question. I'm not here to declare the death of SaaS, nor to dismiss the concern as hype. I'm simply curious about what is actually defensible, and what might not be — in an era where AI agents are beginning to do the work that enterprise software was built to organize.

I want to start, in this first post, with a framework. Because before examining any individual company's position, it helps to be precise about what kind of value each company actually provides — and where that value lives in the stack.

The Question Beneath the Question

When people say that agentic AI will make SaaS redundant, they are usually gesturing at something real but imprecisely stated. The more useful version of the argument, to my mind, is this: agentic AI will make the user interface layer of SaaS largely redundant — and for a surprising number of SaaS companies, the user interface layer is the moat.

Think about what enterprise SaaS actually is, at its architectural core. It is, in most cases, a database with business logic sitting behind it, wrapped in an interface designed to help humans do structured work. The forms, the queues, the dashboards, the approval workflows, the ticket portals — these exist because humans need scaffolding to perform complex organizational tasks consistently and at scale.

Agentic AI doesn't need that scaffolding. It can call APIs directly, reason about state across systems, interpret natural language intent, and execute multi-step processes without a human ever navigating a menu or completing a form. The cognitive overhead that enterprise software was designed to absorb — organizing information, routing decisions, tracking status — is precisely what large language models and autonomous agent frameworks are becoming capable of managing on their own.

The question is not whether AI will change enterprise software. It already has. The more interesting question is: when AI can do the work without the interface, what is the interface actually worth?

This is not a fringe concern. It is a structural challenge that some of the world's largest and most profitable software companies are actively grappling with. The strategic responses — Salesforce's Agentforce, ServiceNow's Autonomous Workforce, Microsoft's Copilot embedded everywhere — are all, in different ways, attempts to answer the same underlying question: can we remain the platform on which agents run, even after the interface we built our business around becomes optional?

Three Types of Moat — and Why the Distinction Matters

Not all enterprise software moats are equally exposed to the agentic transition. To think about which companies face real pressure and which face genuine opportunity, I find it useful to group competitive advantages into three distinct types — each with a different risk profile in an agentic world.

Data & System of Record

Appears Strong — potentially enhanced by Agentic AI

This feels like the strongest moat in an agentic world. Companies with this moat own the canonical, authoritative data that agents must read from and write to in order to do their work — customer relationship history, financial and supply chain records, employee and org data, infrastructure configuration maps. Agents don't make these systems irrelevant. They make them more critical. You can't automate what you can't access. The system of record becomes the foundation on which agentic execution is built.

Workflow & Process Complexity

Appears Moderate — Defensible With Active Strategy

This moat seems moderately defensible, but likely requires active investment to maintain. Companies here own deeply embedded, cross-functional process logic that took years — sometimes decades — to configure, tune, and integrate. The complexity itself creates switching costs. But as agentic AI becomes more capable of traversing workflows via APIs and orchestration layers, the question worth asking is whether the workflow engine remains the home base for process execution, or whether it gradually gets abstracted away.

User Interface & Engagement

Appears Exposed — Worth Watching Closely

This is the moat I think about most. Companies whose primary competitive advantage is delivering a well-designed, intuitive interface through which humans perform work face a direct structural question: agents don't use interfaces. When the value proposition is essentially 'we make it easy for people to do this work,' and AI makes it possible for no one to do that work manually at all, the interface premium is worth revisiting. That's not an automatic death sentence — but it does invite some honest reflection.

Why the Lines Are Never Clean

In practice, most enterprise software companies have elements of all three moat types — which makes the analysis more nuanced than any clean taxonomy suggests. Salesforce, for example, is a system of record for customer relationships (Moat 1), owns complex workflow logic in its Sales Cloud and Service Cloud products (Moat 2), and has invested heavily in the CRM interface that millions of salespeople use daily (Moat 3). The question worth asking isn't which moat a company has — it's which moat actually drives retention, pricing power, and renewal rates.

That distinction matters because the strategies for defending each moat type are quite different. A company whose stickiness is genuinely driven by being the system of record might invest in making its data layer more accessible to agents — becoming the substrate on which the agentic economy runs. A company whose stickiness is primarily driven by interface engagement faces a harder choice: either build genuine agent-native capabilities, or accept a gradual compression of its value proposition.

There is a third path that some companies are attempting — positioning the platform itself as the place where agents are built, deployed, and governed, rather than merely the application that agents interact with. This is the most ambitious play, and it is the one that ServiceNow, Microsoft, and Salesforce are each pursuing in different ways. It's also the play with the highest execution risk and the longest time horizon.

The companies that navigate this well will likely be those that are honest about where their value actually lives in the stack — even when that honesty is uncomfortable.

What's Next? 

In my next post, I apply this framework to the world's ten largest SaaS companies by revenue, looking at each through a simple lens: how they make money, what their moat actually is, and what they appear to be doing about the agentic transition. I find some of the responses more reassuring than the headlines suggest — and others a little thinner than I'd expected.

And in the final one, I'll zoom out to the ecosystem level and ask the more forward-looking question: in a world increasingly organized around agentic AI, which types of players are structurally best positioned to capture value? Hyperscalers, LLM providers, vertical AI specialists, data infrastructure companies — each has a different claim on the agentic future, and I don't think all of those claims are equally well-founded.

The conversation around AI and enterprise software is often conducted in extremes — either dismissing the disruption as overhyped, or predicting the imminent collapse of SaaS as we know it. My honest read is that the truth is more interesting and more nuanced than either position. It depends enormously on which company, which moat, which customer base, and which window of time we're talking about.

That's what I'm hoping to think through together in this series, not to reach tidy conclusions, but to ask better questions.

This is the first post in a three-part series exploring the intersection of agentic AI and enterprise SaaS. The views expressed are my personal observations intended to stimulate thought and conversation, not investment advice or formal analysis.