← ALL NEWS

DIGI.NO · 16 Nov 2026 · NORDICS

Architects are shaped by working with AI

This is a Norwegian opinion piece (a debate reply) by Ole Rabanal, architect and senior developer at Sonat, and Silvija Seres, technologist and strategic adviser at Technorocks. It answers Jasvinder Sadana's question about who will understand future systems if junior developers never learn how they are built. The authors share the concern. They say AI is raising code output while companies cut the people who would learn to understand that code.

Their central argument is that management misplaces the bottleneck. The hard part of building a system is not writing enough code but working out what the code should do: which system owns customer data, why one customer group is treated differently, and who is affected when an access rule changes. The answers are scattered across code, old decisions and employees' memories. AI can produce a technically plausible solution before anyone has found them, so costs shift to wrong data, more dependencies, security work and changes that are hard to make later. A system can work at launch and still be poorly architected.

With AI, architectural choices are made earlier and more often: in the task description, the choice of data sources, the limits given to an agent, and the decision on which suggestions to accept. The authors say an architect must therefore understand both the business and the technical consequences. AI can help map a codebase, find dependencies and compare options, but it cannot know which customer data is authoritative or which old exception the business must still support. People must decide what the findings mean.

On juniors, the authors do not endorse Sadana's idea of reserving tasks for them that AI could do faster. Instead, a junior should investigate how a billing change affects three connected systems, using AI to find code and sketch options, talking to the process owners, testing assumptions and presenting a recommendation. A senior architect should challenge it before building, and the junior should afterwards see what happened in operation.

They also describe a harder kind of debt: systems that work, but whose rationale exists neither in documentation nor in the people who own them. Cutting junior roles adds competence debt on top. A company may save a junior's salary this year and lose an architect in five years, so the authors argue that junior work and senior mentoring must be built into project plans and pricing, and that customers buying development must pay for teams that build skills along the way.

Read the original ↗