Goodbye SaaS, Hello AI Agents

The transition from SaaS to Services as Software with AI agents is upon us. AI agents are here, and a new need for orchestration will come, much like Kubernetes emerged as containers scaled. This will require ways to allocate resources and manage workflows between agents, among other requirements.

AI agents, still incredibly new, will scale over time as two paths emerge in the course of software development:

• Data evolution: We can think of the evolution starting with spreadsheets and progressing to relational databases, data warehouses, big data, predictive analytics, generative AI and AI agents as the next frontier.
• Computing evolution: The story starts in the mid-20th Century, with mainframes: the desktop, client servers, web/mobile, Software as a Service, and now the emergence of agentic workflows

This means some questions about the evolution from SaaS to Services as Software, said Janakiram MSV, a noted analyst and long-time contributor to The New Stack, in this episode of The New Stack Makers.

“SaaS is obviously the frontend for the data,” MSV said. “Without data, Salesforce, Workday or SAP, or any of these applications, they mean nothing, right?”

So two important trends are going to converge.

“On one side, data is becoming more and more actionable. And on the other side, compute is becoming more and more dynamic and agentic. When these two converge, we have a new paradigm, which is called Service as Software.”

AI Agents Will Need Orchestration

Today’s SaaS platforms require human action. AI agents will start to replace the work done by humans. They’ll send emails, follow up on sales calls, etc. Their architecture and components will continue to require the large language models, the tools and APIs, and the memory.

“I’m sure you heard of Satya Nadella” — the Microsoft CEO and chairman — “claiming SaaS is dead and AI agents are going to take over SaaS,” MSV said. “It made quite a big storm in the industry. So what he actually meant was precisely this: SaaS comes all the way till the end and leaves the action to the end user, but agents will close the loop by actually performing the action.”

The challenge comes with orchestration. It’s like moving from a single container to a Kubernetes infrastructure. It’s the scale that takes time. What we need is a Kubernetes for agents.

What’s the impact on the enterprise? As we saw with containers, the scale-out providers will see the need first. Enterprise software managers will face less impact. But the infrastructure will require changes. The statefulness of agents will require enterprise managers to consider issues such as memory management.

The difference with containers is that they’re largely deterministic. Agents are different. They’re built on data that goes through a whole training process.

There’s more! Please listen to this episode of The New Stack Makers.