In this podcast, Michael Stiefel spoke to Joy Ebertz about how the use of artificial intelligence is changing the way architects and developers do their jobs. Focusing on the long term success of the business is one of the most neglected tasks of an architect. This task becomes particularly important as the nature of artificial intelligence changes the landscape of software development. Architects must be especially aware of requirements and their inherent ambiguity. More than ever, they must be expert at explaining technology to the business people, and the business to the technology people. On the flip side, goals must be clearly explained to AI agents, and tests must be unambiguously written. Industry practice is uncertain on how to incorporate changes due to bug reports or new requirements. Should it be done incrementally, or should the module’s or application’s code base be regenerated? This is especially challenging in existing systems, especially old ones where nobody knows what the real requirements were. Unambiguous requirements definition may turn out to be the gating factor in AI driven software development because code reviews seem not to be effective, and verification of the AI work is a great burden.
The conversation then shifted to talk about the future of software developers. The industry needs to find a way to use AI so that human understanding of how software systems work does not atrophy. The flexibility that AI coding gives you reinforces the idea that constraints, and the recognition of the assumptions you are making, are critical concepts on which future developers need to focus. If you look at how AI is actually used, it seems to boil down to two use cases - vibe coding for things that do not have to be perfect such as many internal tools, and essential systems where you specify to the AI, and validate the results very carefully. In the latter scenario, you are effectively pair programming with the AI. Here those constraints and assumptions must be understood. In addition, the industry has not figured out a way to train future software engineers since AI is doing the tasks on which junior engineers were traditionally trained.