Application building categories in the current AI world

In out previous blog: https://www.dheemai.com/post/is-ai-killing-traditional-saas-a-technical-view we discussed what makes AI apps different than traditional software apps. In this post, we present a classification of software applications in the current AI era across three distinct categories:


AI as the core business application: In this category AI executes the core business logic and provides the output. Rather than being explicitly programmed via coding, the core logic is trained to solve problems through reasoning. Since it relies on learned intelligence, output accuracy varies depending on the maturity of the model, training data and resource constraints (time, data and compute capacity). This category will become increasingly prevalent as we approach AGI. This pattern resembles pre-computer business systems where all the work was done by humans except that the speed of these humans(AI) is scaled up close to real time.

AI Assisted software applications: This is currently the most prevalent category. It combines the programmed logic working alongside AI systems. This category best exemplifies the term 'AI Native'. This represents the natural evolution of traditional applications into the AI era. It balances the AI thinking, reasoning and decisions part with the action part using the existing software tools. It is these interactions between AI reasoning and software actions that led to the development of the Model Context Protocol (MCP). MCP enables seamless AI integration with existing tools.

AI generated software tools/systems: With the advent of genAI and its maturity, the practice commonly known as 'vibe coding' has become reality where AI builds production grade applications, tools and systems. Here, AI conceptualizes the required program logic and writes code much like a human developer.

It started as a co-pilot for the software developers and has become so ubiquitous that development without it is competitively disadvantageous and slow. This represents an evolution beyond code completion toward full application generation. This comes with its own problems

. Applications built through this approach require thorough documentation and deep team understanding so that the troubleshooting and maintenance becomes more manageable. Experience shows that inadequate code comprehension by the team results in troubleshooting costs that far exceed the time savings from rapid generation.

I have highlighted the troubleshooting aspects in these categories with the following image:

Understanding these three distinct patterns is essential for technology leaders making architectural decisions. Each category demands different skill sets, governance models, and risk management approaches. Organizations should evaluate which pattern best aligns with their business requirements, risk tolerance, and team capabilities rather than adopting AI approaches indiscriminately.