SaaS in an AI world

The Model That Defined SaaS

For years, the success of SaaS was built on a durable set of economic pillars: recurring revenues, high switching costs, scalable margins and predictable seat-based expansion.

Such characteristics made software one of the most attractive business models in modern markets. Customers paid subscriptions on a per-user basis, rarely switched once embedded and each additional user often generated disproportionately higher profits because the marginal cost to serve them was low. The result was a category of businesses that felt unusually stable, resilient and easy to underwrite.

Software also became mission critical for nearly every part of a modern enterprise. Finance, HR, legal, sales, marketing, operations - across almost every business function, software moved from being a helpful tool to becoming part of the operating fabric of the business.

Over time, many of the strongest vendors evolved beyond single-purpose applications and became broader platforms, with customers building multiple workflows and processes on top of them. That evolution mattered because it reinforced retention: once software became the system where data lived, workflows ran and decisions got made, it became way harder to rip out and replace.

What makes the current moment feel different is that emerging AI, particularly agentic systems like Claude Cowork and Claude Code (tip of the iceberg) - challenges each of these economic pillars at once.

The shift to an Agentic world & the trillion dollar elephant in the room

Over the past several months, high-quality SaaS companies have sharply repriced lower, with seemingly little regard for underlying product strength, customer relevance, or long-term positioning. And its been difficult to watch.

I read a lot of folks argue that the real shift is not merely using AI to improve existing workflows but redesigning workflows from the ground up for agents.

If AI agents continue to evolve the way we think - the important question is no longer how a human uses a software tool more efficiently but what happens when an agent can connect APIs, retrieve context, generate code, coordinate data across systems and execute tasks end to end.

SaaS quickly starts to look less like a fixed interface operated by people/teams and more like a flexible set of capabilities orchestrated by agents.

Its at this inflection point where a chokehold on the traditional SaaS model begins to come into view.

Seat-based pricing becomes more vulnerable if agents can perform work that previously required many human users. Switching costs become less durable if agents can move across applications, stitch together workflows and choose whichever tools are cheapest, fastest, or easiest to integrate. Competitive barriers may also weaken as AI lowers the cost of building software and enables new entrants to deliver similar functionality without recreating the full legacy stack.

That said, it is worth pausing here. Not all SaaS will be disrupted at the same pace or in the same way. In more technical and industry-specific verticals, AI may act less as a replacement engine and more as a force multiplier for productivity.

Historically, software innovation in areas like engineering and industrial design has driven higher output rather than displacement and early evidence of broad productivity gains from AI remains limited at this moment in time. So, in these environments, its the combination of trust, deep workflow embedding and governed data that may continue to favor incumbents more than many market voices assume.

Another caveat which I think is worth not overlooking is how
in our pre-AI SaaS world, many apps needed significant team coordination and admin standardization before they became useful for users. AI or agents may spread differently: users can derive meaningful value immediately, without waiting for company-wide deployment, which means adoption can begin with individual power users and then move into the enterprise later.

Put differently, this motion is like a “self-funded adoption flywheel” in which users buy high-end AI capability directly themselves and the enterprise subsequently formalizes usage with governance, security and compliance controls later.

An example is a finance enterprise, where many individuals purchase Claude Cowork themselves, start building LBO models out of the box, save themselves time, start building more interesting excel model artifacts and widgets. They then start "spreading the word", bringing other individuals onboard. Before you know it, the CIO catches on, and a new Claude Cowork pilot top-down is launched and enterprise scaling and governance occurs later.

I would look at this new type of "shadow AI" pheonomon more carefully and thoughtfully if I were an investor. Its likley, in a world full of hype and distrust, AI value may accrue to SaaS products that win bottom-up first, become habit-forming quickly and then layer in enterprise controls after adoption is already underway. A kind of "trojan horse" model as such.

However, investors must understand the shift is not frictionless. In many enterprise environments, especially in regulated industries, adoption still requires layers of governance, compliance and integration with existing systems. While individual power users may pull new tools into the organization, scaling that usage into mission-critical workflows often takes longer than expected.

A new "right to win" in the age of AI

Putting this all together. I think there are three common themes or new right to win(s) that stand out right now:

1. Right to Win: Own the Core Workflow and the Customer Context

Across industry conversations, X, BlueSky, Discord and investor circles, an increasingly common view is that value within SaaS is unlikely to accrue broadly or evenly. Instead, it is more likely to concentrate in a narrower group of companies: those that already sit at the center of mission-critical workflows, control proprietary data and distribute their products through unified platforms rather than fragmented single point solutions. In other words, the advantage increasingly belongs to software that does not merely support one task but anchors the full operating context around it.

That said, the pace of this transition will likely vary significantly by category/industry. In high-consequence or regulated workflows, becoming a system of action is not simply a product challenge but a trust and accountability challenge. The platform is not only expected to automate work, but to do so in a way that is auditable, reliable and compliant - which may slow the rate at which new entrants can displace incumbents.

2. Right to Win: Become the Orchestration Layer for Agents

A second and closely related idea is that the next wave of winners may be defined less by how many seats they sell and more by whether their products can become the orchestration layer for agents. As agentic systems become more capable, software that is easy to integrate, rich in workflow context and capable of coordinating actions across multiple systems should become more valuable than software that simply offers another interface for a human user. Products that help users generate recurring leverage - by automating work, compressing time exchaning data and improving overall output - may therefore prove more defensible than traditional seat-based applications whose value is tied mainly to headcount.

3. Right to Win: Evolve from a "System of Record" into a "System of Action"

Finally, the most advantaged SaaS companies may not be generic applications with large user bases but platforms that begin as systems of record and evolve into systems of action. The distinction matters. A system of record stores the data and sits inside the workflow; a system of action uses that position to trigger, coordinate and automate work across the organization. What changes in the agentic era is that merely storing the data and a few workflows is no longer enough. Agents can now execute end-to-end tasks either natively within incumbent platforms or by connecting externally via APIs which means the competitive question shifts from who owns the record to who controls the action. SaaS apps will have to innovate and improvise quickly to evolve and open-up their tech stack to gain competitive advantages.

Just an caveat to the system of action point here - it truly matters.

I think quite a lot of folks haven't got their heads round it yet.

Mainly because it requries a complete upgrade of the tech stack - whether thats APIs, more AI-native integrations and orchestration - it all matters so much! Leaders will need agent/AI readiness, proprietary and vertical-specific data, memory moats, scalable distribution and business models tied to value driven outcomes. That's a lot of change!

But I get it. It's a lot of change. Lots of SaaS companies will turn their head at this - as it represents a mountain of challenges for leadership to overcome.

Examples of early SaaS winners

ServiceNow has been making a lot of great architectural decisions over the past ~12 months or so. They own end-to-end service management space, including project management and ticketing through a system of action because they route incidents, enforce business and project controls, orchestrate ticket remediation and automate adjacent cross-functional processes all through agents which share data and context end-to-end.

Likewise, GitLab is owning the software development/DevSecOps lifecycle - as agents can now build, scan, merge, deploy, test and policy enforce code across the lifecycle - all while sharing data, connecting with other third-party systems autonomously.

Datadog is another stand out, as they continue to embed telemetry, alerting logic and automated remediation into operational workflows across a unified system of action architecture.

All three of these companies are already progressing beyond the systems of record. They already control structured actions, workflow logic and execution, not just the underlying data - thats differentiation.

They do not simply store data and wait for a user to decide what to do next. Each already sits inside a live operational workflow where decisions are made and actions are executed proactively.

That distinction matters because a true system of action is not created by merely layering AI features onto an incumbent product. They are not an incumbent who simply "adds AI". I see this a lot in the market and it makes total sense, of course. But often, they are not a system of action.

Large language models (LLM) may improve search, summarization, triage, or code suggestions but cannot replace the platform where the workflow itself requires orchestration, permissions, policy enforcement, remediation and deep integrations.

In other words, adding a copilot or chatbot to a traditional web app interface is not the same as becoming a system of action. A system of action is defined by ownership of the workflow itself — the ability to trigger, coordinate, govern and complete work across the stack.

But what about smaller SaaS players?

I agree. Up until now, I've articulated winning attributes in a way that naturally favors either scaled incumbents or very focused AI-native challengers.

My impression here is that most smaller SaaS vendors are unlikely to make a full transition into true systems of action within the next 18 months, at least not in a durable, defensible way.

Many can add AI features on top but far fewer can build the full workflow depth, integrations, trust and distribution needed to become the operational control layer for agents.

Putting my investment hat on, I would be looking at these smaller players carefully. They may be able to add useful AI but if it occupies only one narrow step in user workflow, it may struggle to become the place where agents actually coordinate and execute work.

I will follow up in a seperate blog post regarding this topic in general.

But looking at the near-term market, its most likely going to bifurcate between AI-native challengers vs incumbents defending the system-of-record position, leaving many mid-sized or smaller traditional SaaS vendors in an awkward middle ground.