The end of SaaS? How agentic AI will reshape your tech stack

Originally published in CustomerThink

Marketers love to call time on big, behemoth technologies. Disruption makes a great slogan (we all remember “email is dead,” which was catchy but ultimately incorrect). 

Now agentic AI has stepped into the spotlight, and people are wondering if SaaS is on the chopping block. AI agents can act independently, not governed by strict if-then logic and pre-set scripts. They can complete complex, end-to-end tasks across different systems and make decisions based on their own reasoning. 

The existential threat they pose to SaaS, in theory, is that an AI agent could complete the same multi-step workflows that SaaS would otherwise be responsible for – project management, data analysis, predictive IT maintenance, campaign orchestration, to name a few. 

While I think agentic AI will reshape our tech stacks, I don’t think it comes down to a simple game of “this or that.” Historically, older technologies have been tenacious, evolving in lockstep with new technologies, rather than disappearing entirely. 

The more interesting questions, in my opinion, are: how will agentic AI force SaaS to adapt? And what does that mean for the modern tech stack? 

The future of SaaS is open and programmatic 

From the 1990s onwards, SaaS revolutionized how businesses used software. Unlike the expensive, self-hosted, on-prem solutions from before, SaaS offered a refreshing alternative due to its cost-effectiveness, fast implementation, and reduced IT maintenance. In fact, SaaS was heralded for the very thing that AI is being praised for today: removing barriers to entry. 

Yet the explosion of this new software delivery model came with some tradeoffs. SaaS can be difficult to integrate with other systems, which creates silos. It can also be limiting: workflows are usually preconfigured and may not offer much leeway in terms of customization.  

These disadvantages in SaaS are where AI agents are meant to excel: its interoperability and tailor-made solutions. You could ask an AI agent to create a go-to-market plan based around a specific deadline, and from that simple prompt it could pull in meeting transcripts, your product roadmap, and team bandwidth to create a step-by-step strategy. Or maybe you want to understand if there are any underlying patterns around churn rates: an AI agent could analyze engagement metrics, Net Promoter Scores (NPS), and customer support tickets to help identify root causes. 

In both examples, an AI agent would be accessing several different systems, owned by different teams, and synthesizing this information in real time.

SaaS is competitive because it offers an intuitive, easy-to-use interface for humans to do complex work. Now, the value may shift to how well an AI agent can interact with systems programmatically. Since this would be through code (rather than buttons and dashboards), businesses should be prioritizing structured APIs, open platforms, and standardized, machine-readable data to become “agentic AI ready.” 

For SaaS, there’s also an opportunity to integrate agentic AI into their software, and many have already started. SaaS companies are updating features with LLM resources, and are considering how agentic AI could help speed up time-to-value for their customers. Not only could AI agents help users with complex tasks, but with access to the platform’s core data and process design, agents could replicate workflows entirely. This is another opportunity for SaaS: training their own AI agents based on their proprietary data and systems, and their extensive industry knowledge. 

There will also be times where it might not make sense for an AI agent to take over. Take payment processing, which is often routine, and requires strict security and regulatory requirements. Does it make sense to have an AI agent handle this from a computational and cost perspective? Maybe not. Instead, an AI agent would excel in the more complex, cross-functional, and predictive workflows – like fraud detection, 24/7 billing support, or audit preparation. 

What’s ahead

The value of agentic AI isn’t overblown. But often there’s a pressure to chase trends, which comes at the expense of strategy. 

First, focus on your foundation. Deploying an AI agent won’t do much if it’s unable to interact with other systems. API-first architectures, strong data governance, and clear protocols around what data the AI can access (and when a human should be brought in) are critical. 

Then, look at the business outcomes you want to achieve. What use cases could an AI agent help scale, streamline, or simplify? Always keep the ROI in mind. You’ll need to consider the infrastructure costs, engineering and training resources, and operational oversight needed to make sure the AI is functioning as intended. There might even be some use cases in which multiple AI agents are needed to deliver the outcome you’re aiming for. 

Recently, it seems like there’s a new “fundamental shift” every other week. It can make it difficult to see the best path forward. But there’s a popular piece of advice on this: focus on what will inevitably change and what will never change. We know tech stacks will always evolve – there will be new systems, new AI models, new predictions and disruptions. What won’t change is the need for these different systems to work and scale together, and for businesses to deliver value to customers. What will change is when and how often an AI interacts with a system versus a human being. 

In this way, the whole agentic AI vs. SaaS debate feels like tunnel vision. Agentic AI won’t just change SaaS, but every layer of the tech stack.