This weekend, an army of AI agents wrote thousands of lines of code for me. And that is exactly the problem.
For years, the bottleneck was the speed at which IT teams could build new functionality, but AI agents are fundamentally changing that. Not because they "help a little", but because they accelerate and parallelise execution at scale. This weekend, my "private army" of AI agents was at work, each with its own set of tasks, operating like a full-fledged development team. Thousands of lines of code in hours, not weeks.
And because we are prompting rather than coding, this is also becoming accessible to business professionals. With the right tools and guidance, they can produce workflows, applications and code. This expands not only the speed of execution, but also who can contribute to digital delivery.
Productive? Absolutely. But that is precisely where a new bottleneck emerges: the quality of direction. Anyone can generate code, but it still needs to work and remain reliable in production. The focus shifts to quality assurance, architecture, security and governance. In short: the ability to translate speed into sustainable business value.
Should we be willing to question the traditional structure of development teams? Work that used to be assigned to junior profiles is becoming automatable. Senior professionals create value not so much through sheer output, but through judgement: architecture, prioritisation, review and course correction. And if people from the business can contribute as well, we need to ask how we organise this, who provides direction, who safeguards quality and coherence, and who ultimately takes responsibility.
There is another question too: how do you develop future senior professionals with sufficient technical depth if hardly any code is still being written? And if the business can produce as well, how do you avoid fragmentation, shadow IT and solutions that are not sufficiently robust?
On paper, it sounds attractive: everyone generating code. But in practice, it creates a risk of "technical debt", code that no one fully understands anymore, and dependency on a small group of people who have to oversee everything. AI is not the problem here. The problem is trying to accelerate without control while leaving the operating model unchanged. AI in software development is not just about efficiency, but about redesigning responsibilities, team structures, collaboration between business and IT, and how juniors are supported in developing core skills.
The relevant question for IT and business leaders is therefore not:
"How do we use AI in development?"
But:
"How do we organise ownership, governance and collaboration when production is no longer the limiting factor, and how do we still develop the next generation of capable senior IT professionals within that model?"
