How AI Is Changing Software Development Part 3 -
The CTO´s Perspective

Maintainability Becomes a Competitive Advantage

Software companies have long hoped that one day they would be able to develop new functionality faster. Continuous Integration, DevOps, and most recently Agentic Coding ultimately pursue the same goal: They make development teams more productive and reduce existing bottlenecks. As a result, a product organization’s ability to deliver new functionality increases.

However, this also raises the question of what we actually want to use the potentially higher output for. As development capacity increases, so does the number of possible features, and suddenly many things become feasible that could previously never be implemented due to capacity constraints. Even individual customer requests may suddenly become economically viable, while internal ideas no longer automatically disappear into the backlog. At first, this sounds almost too good to be true.

For product organizations, however, this creates a new risk: The easier it becomes to create features, the more difficult it becomes to decide which features should actually be created in the first place. As a result, many software companies will probably build a feature graveyard in the coming years with their newly acquired delivery capacity, paradoxically limiting their ability to deliver more and more.

Not because individual features were poorly developed.

But because each individual feature appears reasonable when considered on its own. Taken together, however, products become increasingly complex, offer more and more configuration options, and become progressively harder to use.

More Features Do Not Automatically Mean More Customer Value.

There is a second effect as well: Every additional feature increases testing effort, expands documentation requirements, increases the number of possible interactions, and makes future changes more difficult. Product maintainability gradually declines. This effect is already visible in many organizations today. Agentic Coding is likely to amplify it further if development capacity grows faster than the ability to reduce complexity again, particularly for customers.

That is why I believe another type of agent will ultimately be at least as interesting. Today, we primarily talk about how agents can generate new code. In the future, however, we may need more agents that simplify existing code, identify duplicate functionality, or suggest unnecessary features for removal.

Maintainability does not come from good architecture alone. It also comes from regularly removing complexity from a product.

This also changes the requirements placed on developers. For many years, their performance was often measured by how quickly and reliably they could implement solutions. This ability will remain important. At the same time, another skill is becoming increasingly valuable: the ability to make good decisions.

In the future, developers will more frequently need to assess — together with Product Management, of course — whether a proposed solution fits the existing architecture, what its long-term impact on maintainability and security will be, and whether a new feature actually increases the value of the product or merely adds complexity.

In my view, this makes sound judgment one of the most valuable skills within a product organization (see also Part 2 of the series “How AI Is Changing Software Development”). In the long term, I believe only those companies that use their additional development capacity without losing control over their products will be successful.

The three previous parts of this series had one thing in common: They described how agents are changing everyday work, the organization, and the product. One question therefore remains unanswered. If all these underlying conditions are changing, our understanding of what makes a good developer will inevitably change as well.

That is the question addressed in the fourth and final part of the series.