Human Resources: Business Value Instead of Code

Do you know the legend of the grains of rice? Here’s the short version: An Indian ruler granted the inventor of chess one wish. He wished for grains of rice: one grain on the first square of the chessboard, twice that amount on the second square, twice that amount again on the third, and so on. The ruler laughed at what he thought was modesty. He was wrong, because the wish amounted to 18.4 trillion grains of rice—enough to cover all of Germany to a depth of one meter.

AI vs. the Human Brain

We still underestimate the problem of exponential growth, AI researcher Christian Bauckhage explained recently. With artificial neural networks doubling in complexity every year since 2016, they are expected to reach the complexity of the human brain by 2030, according to the professor, who conducts research at the Fraunhofer Institute for Intelligent Analysis and Information Systems, among other institutions. Examples such as Google’s protein folding program AlphaFold or the ongoing solution of „unsolvable“ mathematical problems already demonstrate today how AI is surpassing human capabilities and leading us into an „agent-based“ future, Bauckhage said.

Spec is the new Code

Software development is changing, and the required skills are shifting—it’s already happening, and at an ever-increasing pace. In the future, AI and AI agents will take on additional tasks in the development process alongside coding. In AI Spec-Driven Development, we developers are already describing the application in natural language. The specification serves as both the documentation and the single source of truth—not the code. This is because the code is modified by changing the specification. And we take a highly iterative approach in doing so. At Snap, we first describe the business requirements without any reference to technology and then supplement the project with business, legal, and operational constraints that the AI takes into account.

Phase two involves the technical design—documenting technical concepts and architectural decisions. Planning takes place in phase three—it makes sense not to implement the entire scope right away (which the AI could certainly do), but rather to proceed step by step. We take an iterative approach in the smallest possible steps, review the results, and continuously adjust the specifications or our process model. To avoid the risk of AI fatigue from reading hundreds of text files, we have the AI generate continuously updated, „human-friendly“ diagrams for reviews and approvals. Here, too, we don’t work through countless test cases—the AI handles that for us as well—but simply verify whether the result correctly implements our requirement. If not, we formulate the change via prompting, and the AI ensures that everything remains consistent, from the code to the specification.

However, even with AI support, the most important thing remains the same: customers and developers must understand their requirements. Let’s be clear: the ABAP world is also changing due to AI. Perhaps in a different way and possibly a little later, but inevitably. AI can only provide support here and cannot code autonomously, because code in ABAP simply isn’t as easy to manage as it is with Python or JavaScript. But even here, tasks, processes, and workflows are already changing. At Snap, for example, AI is already helping us with the standardization and quality assurance of our SnapWare products. Qualified human experts remain crucial, even when AI is used.

As “conductors,” we (former) developers will likely take on an even more central role in the future. And thanks to AI support and expanded skill sets, our productivity—and thus the business value we generate—will increase. In addition, this will free up time for more complex and value-adding activities—such as deepening our understanding of exponential growth.

To the partner entry: