Good Code Still Isn't Enough in the Agentic Age

Agentic software development is transforming the way software is created: AI agents are already taking over large parts of the development process. But excellence does not come from code alone. It depends on how well companies adapt their roles, decision-making processes and management structures.

For IT service providers and software vendors, this means that their business model, which has so far been based on developers' working hours, must adapt to these new possibilities. After all, a new tool does not automatically change the way a company operates. Those who develop with agents but leave their organisation unchanged realise only part of the potential. Agents can write code quickly today, but it is the organisation and the people around it that determine whether this code does the right thing and stands the test of time in operation.

According to IT service provider adesso, five prerequisites determine whether a company's transformation towards the agentic era succeeds:

1. Redefine roles deliberately. An "AI Officer" on a business card is not enough if the underlying task remains unchanged. New roles such as the Forward Deployed Engineer, who links technical expertise directly to business processes and requirements on the ground, already show this. Existing roles are shifting too: those who process an insurance claim or answer a customer enquiry today will increasingly review an agent's results tomorrow and intervene only in exceptional cases. This requires a different kind of judgement than mere processing. Anyone who manages an agent takes on a leadership role at its core and is responsible for results they no longer produce themselves. This calls for a deliberately redefined job description, to which targets and performance reviews are also aligned.

2. Clarify responsibilities and approval processes in advance. Most approval chains have evolved over the years and have rarely been deliberately designed. They often reflect who was trusted at the time. As soon as an agent takes over part of the decision-making, this lack of clarity becomes a risk. Who is liable if an automated decision is wrong? Who must give prior approval, and who is merely informed afterwards? These questions are best answered before go-live, with documented boundaries within which an agent may act independently.

3. Adapt spans of control to new tasks. Many management structures are built around a fixed amount of coordination, in which a manager coordinates, prioritises and resolves conflicts. If an agent takes on part of this coordination, the actual span of control changes. If the structure stays the same, duplicate structures emerge: the agent coordinates, while the manager, out of habit, reviews every decision again. This costs time, and the hoped-for acceleration never materialises.

4. Build trust through transparency. Employees are most likely to accept an agent if they can understand what it does, why it does it and where its limits lie. Without this transparency, mistrust grows. In day-to-day practice, it shows in whether the agent actually takes workload off employees without them losing control, or whether it is imposed on them without explanation.

5. Embed training permanently. As agents take on more tasks, the demands on staff keep changing. A workshop at the start of a project explains the technology, but not how one's own role will evolve over the following months. Companies that take this seriously develop training programmes that evolve alongside the agents.

All five points concern organisational development. Yet this is precisely the work that many companies skip or overlook. Those who tackle these issues early gain a head start that no model update can replace.

Among other things, adesso describes here how it is aligning its own organisation for the agentic era.