AI will change the world of business applications

In a recent post, my colleague Franz Färber shared some profound thoughts on how drastically AI will change the world of business software applications. See

https://www.linkedin.com/pulse/generative-erp-future-business-applications-franz-faerber-rvdbe/?trackingId=TNmKeKuRTUe0DQ7%2Bq6uniw%3D%3D

To summarize, the way and quality of how AI models are able to generate code, based on specifications and high-level architecture descriptions, is just incredible. And what we see today is just the beginning. It is the start of a sea change similar to the transition from writing assembler code to “business languages” like COBOL, ABAP, or Natural. Back then, the machine code became an invisible vehicle to execute a program; nobody looked at it again. Fast forward to 2025, “normal” code, even written in modern languages like Java, Rust, or Python, will see the same fate as prompt-generated software development has become so convenient. The first generations of tools beyond just the usual chat prompt line are already available; take a look at Windsurf or Cursor.  

How this is related to the viral discussion about "agentic AI" that Satya Nadella stirred the tech world with (see https://www.youtube.com/watch?v=9NtsnzRFJ_o), is the topic of my last blog here https://www.everest-systems.com/post/the-role-of-ai-agents-in-the-future-of-erp-systems-revolutionizing-app-development

In this blog, I want to elaborate on what this means for the relationship between business and technology. It's nothing short of a revolution. We all know the examples of how a business requirement degenerated into a useless or unusable piece of software during the process of capturing a business need to deliver a product. 

And yes, this revolution will put more burden on the business side in terms of being asked to be as clear and explicit as possible in documenting what is needed. The point is the ability to write down what is really required and NOT what the technology solution should look like. It is all about working backwards from the user's needs to technology, where the technology (= code) will be completely hidden. The resulting document will be used to ask the AI model to write the first version of a specification, weaving in the “world wisdom” of the model and (public) knowledge of other solutions on the market. This might be subject to some iterations back and forth. The final spec can be checked again with the customer and users. If ready, the specification is used to ask AI to transform this into a software architecture paper, adding information on the existing system environment in terms of APIs and technologies used (e.g., UI libraries). Once this is finished, a capable engineer would feed those documents into these new code generation tools that look like this

After the initial code generation based on the spec, you just “chat” with the system, telling it what it should do to optimize the result, constantly looking at the UI and what the user experience feels like. This is the right-hand side of the screen. So, the spec and the chat history are a complete description of the code and the intent. The specification gets “bound” to the code and cannot become outdated or detached by definition of the process. When the spec changes, we completely re-generate the code - so what? Manually typing in code will not happen anymore. The speed of innovation and the validation iterations with the customer will increase by factors, not percentages. The specification paper becomes a living document that is readable by business people and technologists at the same time. 

But is this really feasible beyond toy applications? Is it truly enterprise software ready? The answer is yes, IF you have a business software foundation in place that serves as a starting point. What does the foundation need to contain? Here are the major cornerstones:

• Cloud infrastructure (database, Kubernetes cluster, network,...) 

• AI framework (Bedrock, OpenAI,...)

• Runtime layer (kernel)

• Application server

• Integration service

• Sandboxing

• Application foundation with all the basic accounting and finance operations, from sales order management over invoicing to billing, that one needs in almost every transactional business app

To AI-generate such a technology foundation is out of scope for the foreseeable future, but apps on top? Absolutely yes! 

Here are some examples that come to mind:

• A new sales forecasting tool that integrates CRM and finance data at the same time

• A professional services automation app

• A performance and talent management tool 

• Dedicated dashboards and analytical apps for specific use cases, e.g., SaaS metrics

The last example is intriguing, without sophisticated design-time environments for dashboard creation, dashboards can be generated and embedded in an ad-hoc manner by business users. This is already product reality without any use of an IDE as above. Tools will be replaced by prompting. 

Imagine the speed, efficiency, and effectiveness of this approach. And it goes way beyond AI agents that “just” wrap an AI model, querying a database or isolated “AI features” that were more invented by software people rather than required by business users. 

The interaction between the software vendor and the business teams at a customer will become much more direct and collaborative. Customers and consultants can use the same tools and procedures, feeding back the result to the software vendor. In the case of an update, the whole code gets re-generated again, based on the most recent documentation coming from all parties. 

To get a glimpse of where AI stands in terms of competitive programming and how fast it is evolving, take a look at this:

The underlying study has been published recently:

One is immediately reminded of a similar development that led to AI becoming unbeatable in Chess and Go. So, the race is on. It'll be super interesting to watch how the software industry will change...