I have worked in architect roles for most of my IT career, so I will start by looking at how AI is changing that role. Separate posts about other software development roles will follow.
So how is the architect role changing now that we are using AI to develop much of the software, while also implementing AI as part of the products themselves?
First, architects need to plan systems that are compatible with AI-assisted development. Ideally, they should be optimized for it. Since AI still struggles with very large contexts, moving away from monolithic solutions is probably still a good idea if you have not done so already. But somewhat counterintuitively, a pure microservices architecture might not be the ideal answer either.
Highly fragmented systems require the AI to understand a large number of services, which again creates context challenges. They also increase the risk of duplicated logic and overlapping solutions if you are not careful. And once services become small enough, you cannot really leverage AI autonomy to the same extent as you can with larger, more self-contained services. At that point, the AI mostly just wires things together and asks what to do next.
This suggests a middle ground. Architectures where the context does not overwhelm the AI, but where the AI can still work autonomously within a service for meaningful periods of time. Modularity still matters, and fine-grained testing becomes even more important when a non-human contributor is writing a noticeable portion of the code.
When AI is part of the product, architects also need to understand how it can actually be built. Can the requirements be implemented by calling an LLM via an API and providing instructions? Or do you need retrieval augmented generation, multistage agentic pipelines, fine-tuned models, or something else entirely? Do you need security layers to validate inputs and outputs, or data protection solutions to control what the model can see? Is this even an LLM problem, or would a more traditional machine learning approach be a better fit?
Not everything needs to be AI. Still, a large portion of new projects now include some AI component, whether it is justified or not. Architects should understand the basic building blocks well enough to plan realistic systems and ask uncomfortable questions early on.
