01. Jul 2025
Agentic AI: The future of artificial intelligence in software engineering
Software engineering technology is developing rapidly. A new concept is fundamentally changing how companies - and we as an IT service provider - use artificial intelligence: Agentic AI. But what exactly is Agentic AI and why should modern companies get to grips with it?
What characterizes Agentic AI and distinguishes it from other forms of AI
Agentic AI refers to autonomous intelligent systems that can make decisions, act and learn independently. In contrast to generative AI, which reacts to human commands, AI agents act on their own initiative. They observe, analyze and orchestrate processes without the need for a constant command-reaction cycle.
In short: Agentic AI makes independent decisions, acts proactively and observes - while we evaluate and control the results.
A decisive advantage: Agentic AI covers all areas of software engineering, from design and development to operation and support. It therefore goes far beyond pure code generation and offers a new quality of collaboration between man and machine.
Generative AI vs. Agentic AI an overviewTwo AI systems: Generative AI and Agentic AI in comparison
| Aspect | Generative AI | Agentic AI |
| Systemtypa | Reactive system: Executes tasks when instructed to do so. | Proactive system: Acts on the basis of an input and adapts dynamically. |
| Dependence on prompts | Requires clear prompts (commands) and is based on statistical patterns learned through training. | Responds to prompts, decides independently on execution and learns from the results. |
| Autonomie | Only performs the tasks that are explicitly specified. | Independently continues the command-reaction cycle and acts without constant intervention. |
| Focus | Supports everyday tasks, such as the generation of (creative) content. | Can plan and carry out complex, multi-stage processes independently, e.g. purchasing processes. |
| Working with people | The AI creates proposals that are revised and controlled by humans(humans remain active). | The AI breaks down tasks into logical steps, recognizes problems itself and makes its own decisions. |
| Chain of Thought Reasoning | No chain of thought reasoning; control and interpretation entirely by humans. | Uses "Chain of Thought Reasoning" to break down tasks logically and find solutions independently. |
| Basic technology | Uses Large Language Models (LLMs) for simple applications such as chatbots. | Uses LLMs as the basis for the underlying reasoning engine. |
Why we benefit from Agentic AI and AI agents in software engineering
We are currently using AI agents successfully in both the modernization and new construction of software systems.
Modernization rethought: AI agents in use for migration
In the area of modernization, we benefit from the fact that AI agents not only reliably create inline documentation, but can also analyse entire code bases and generate comprehensive, comprehensible documentation from them. This makes maintenance easier and enables a deeper understanding of complex, evolved systems.
Another focus is on the migration of old technologies: AI agents support us in efficiently replacing outdated programming languages and libraries. We have already successfully implemented modernization projects - such as the migration from Pro*C to Kotlin or from PL/SQL to Kotlin. The quality and efficiency of the automated code transformations significantly reduce manual effort.
New construction - without programming a single line of code
AI agents also open up new possibilities when building new software systems . They accompany the entire development cycle: from requirements gathering, code generation and test automation through to execution and continuous improvement. I am particularly pleased about our first project, which we implemented with a consistent "no code" policy. The aim is to stop writing code by hand - and the first successful run shows that this can already be implemented in practice with today's AI models.
Despite all the automation, our standards for software quality remain high. We check the results generated by AI very carefully. Maintainability, comprehensibility and principles such as SOLID, separation of concerns and clean code remain the focus. With the right prompt, these standards can also be adhered to surprisingly well with AI-generated code.
More freedom for expertise and innovation with AI agents
Overall, the AI models in agent mode published at the beginning of the year show a significant increase in efficiency - both when modernizing and building new software systems. We can use the time gained to focus more intensively on the technical requirements and strategic consulting of our customers. In this way, we create scope for real innovation and sustainable value creation.
What are your experiences with the new generation of AI agents? We look forward to the exchange.
