GenAI Impact Study: Seven Principles of Agent-Based Software Development

When using agent-based software development, seven principles make all the difference:

1. Short cycles beat long sprints: Traditional agile development operates in sprints lasting two to three weeks. Agent-based teams create cycles of two to four days across the entire software development lifecycle: from requirements through implementation and testing to review and operation. This makes a difference both in terms of quantity and quality. Shorter cycles change how decisions are made, how bugs are discovered, and how quickly a team responds to changing requirements. Agents work continuously—at night, on weekends. Releases become more frequent, and iterations smaller. Once you’ve worked at this pace, you won’t want to go back.

2. Different Roles: Human and Agent: The most common mistake when getting started: AI agents are treated like very fast developers. Agents are not junior developers who can simply be set loose on tasks. They are executing instances that need precise guidelines to function. Humans provide the „why“ and „what“: architectural principles, business requirements, and quality criteria. The agent handles the execution. This includes: generating code, running tests, fixing bugs, and writing documentation. All of this takes place within continuous feedback loops, without the need to manually trigger each step.

3. The quality of the input determines the output: Halluzination is not a technical problem that can be solved with a better model; it is a contextual problem. Vague requirements for AI agents yield vague results. Structured artifacts such as clear specifications, architectural principles, and contextual documentation are not a tedious chore, but rather the actual control layer. A good specification document is more valuable in agent-based development than a thousand lines of hand-written code. It is the difference between an agent that knows what it is doing and one that is guessing.

„The question is not whether companies act as agents
will work, but rather how quickly they can
”Create the conditions for this."

Benedikt Bonnmann,
Member of the Management Board,
Adesso

4. Guided automation instead of full autonomy: One of the most persistent misconceptions: Agents program on their own while humans look on. But agents do not act freely; rather, they operate within structures defined by humans. In practice, this means: Highly specialized agents with a clearly defined focus and precise context operate within explicitly defined handoff points, automated quality gates, and clear acceptance criteria. As the oversight authority, humans handle tasks that require human judgment: architectural decisions, strategic direction-setting, and final approvals—not line-by-line review.

5. Governance: A Requirement, Not an Option: Agent-based development raises new questions for compliance officers. Which models can be used in which projects? How can we ensure that no code containing protected IP ends up in training data? Who bears the LLM costs, and how are they allocated on a project-by-project basis? These are solvable questions, but they must be answered before the first productive deployment—not after. A standardized, legally compliant approach to LLMs with a transparent cost structure and defined governance rules is the foundation for scaling agent-based development.

6. Legacy is not an obstacle: The assumption that agent-based software development only works in a greenfield environment is incorrect. In practice, it has been shown that existing codebases, legacy systems, and mixed teams can be integrated into agent-based processes. The prerequisite is a methodology designed for this purpose. The key principle is parallelism. Traditional and agent-based teams work within the same system, with clearly defined handoff points and a coordination mechanism such as Kanban, which prevents conflicts between the different cycle frequencies. A traditional team works in two-week sprints, while an agent-based team in the same project operates on two- to four-day cycles. This is not merely a theory; it has already been tested in client projects.

7. Software is just the beginning: The principles of agent-based work describe a basic model for collaboration between humans and autonomous systems, including clear role assignments, structured input, short cycles, guided automation, and governance by design. Agent-based approaches can be applied wherever processes are iterative, documentable, and breakable down into distinct steps. In quality assurance, document processing, and automated analysis. And in business processes that companies still handle entirely manually today—not because it makes sense, but because there has been no viable alternative so far.

To the partner entry: