Is AI killing traditional SaaS ? — A Technical view

This statement “The notion that business applications exist, that’s probably where they all collapse, right in the agent era” from Satya Nadella back in December 2024 made almost everyone ponder at what will happen for the traditional software business after the advent of Gen AI?

I was not clear until recently what he meant to explain and if that is really the case.

Traditional Software and AI systems for business applications

The AI agents are nothing but a specific piece of code/tool built with a specific task in mind that uses one or more AI model/s which are more broader in nature with respect to the specific task.

Usually AI provides different outputs for the same set of inputs, for example you will see different output (not exactly the same) for the same prompt when used multiple times where as the AI agent is built to provide consistent output for the same set of inputs no matter how many times you consume/use it with the same inputs.

The Agentic AI system can thus leverage these AI agents and provide the needed business logic to replace the traditional business applications.

Now a new question arises in my mind, why are we trying to make the AI models provide consistent output with these AI agents (remove the intelligence, make them dumb robots), isn’t this what the traditional business applications did, provide consistent outputs for the same inputs ?

For example: Consider the simplest computational requirement, adding two numbers. I can ask why would I need an AI agent to do this task when I can code it easily and it is common sense.

After going through many explanations, I think I have come to a better understanding and my question above is completely irrelevant. Now one may wonder why did I even mention that ?

The reason being simple enough, curiosity. I am sure many of you would have churned through the same/similar question in your minds and here I am trying to clarify the way that I have understood. On another thought, this was also a serious question for me being a software professional, as finding answers would help me understand what I should do next to make my living at this turning point of the technology world.

In the below table, I have summarized some preliminary differences between traditional software apps and AI agent based apps to handle business requirements:

Some innate thoughts on the differences in traditional software and AI based business app

The AI models have impressively gotten better and are getting better continuously, especially the LLMs along with other multi-modal AI models, they have opened up unimaginable possibilities for us.

Instead of iteratively building tools to solve the dynamically evolving problems I can just have these AI models to address them without having to build a newer version of the tool. The build-release cycle of a tool/software is not that easy all the time and usually has some cost as the evolved problem has to be coded, tested and delivered all of which have their own inherent problems and risks in addition to the evolved problem which is being addressed.

It is based on such understanding, you will hear many experts saying SaaS/business apps are becoming irrelevant.

However I have not seen this in the real world yet, no matter how matured these AI models have been, they have not replaced the traditional software tools completely. Reason being: no clear accountability ! Who is responsible if the AI makes a mistake ? The correct answer is the person who decided to use it for that purpose.

It’s like I build a tool to solve my own business problem, and when using it in my business, it is only me who is responsible for that tool. If there is a bug, or the tool utterly fails for a specific edge use case I can only blame myself.

So how is the AI being leveraged/used to make the most out of it? It is the human in the loop (HITL) that completes the circle. Maybe one day we see a total maturity in AI that can remove the HITL, or do we even want it is another question? (We don’t want a Skynet, do we? ;) )

Back on the initial topic, are these Agentic AI systems replacing the traditional SaaS/business apps? Well the answer cannot be a simple yes or no. The agentic AI systems are versatile in addressing deep integration than the traditional business apps and on the other hand the deeply integrated point tools can match or even outperform agentic AI on specific, bounded workflows, but only when they own the integrations, data context, and control-loop needed to deliver end-to-end outcomes with reliability and governance. Beyond narrow scopes, agentic architectures regain the advantage in adaptability and cross-system orchestration.

Today the AI can even build traditional apps successfully, thanks to the ‘Vibe coding’. You can just have the AI do your software development. How effective is this and to what extent? I already see commercial AI services that can effectively build simple web apps and mobile apps. There are even AI based services that can deliver/deploy the apps to environments (production).

Agentic AI systems building traditional software tools/apps

Let’s ponder over the different patterns of AI systems being used in Apps in the current landscape in another post.