Software Development with AI

Now we have a tool to both significantly improve and significantly worsen the codebase legacy: How Can AI Affect Code Quality

Human language is a pretty inefficient way to describe logic of complex systems in prompts because english is not the hottest new programming language.

Development workflows have to be re-adjusted. There should be more quality gates between any code change and production. Ensuring that at least some of the quality gates are organic.

Software development does not entirely consist of hard problem-solving that requires deep expertise. There are a lot of shallow tasks that can be offloaded to AI in the form of "intelligent" code generation.

Sorting out task types and selecting those well-suited for delegation to AI is an art.

Software Development with AI needs an extensively well-thought-out application architecture. Flexible enough for different use cases, but restrictive and consistent to ensure that everything works as expected at the high level and hard-to-spot bugs are secured. That's a hard trade-off to balance.

At the moment, it feels like the industry is swept up in a FOMO-driven rush, overlooking the security vulnerabilities we’re exposing ourselves to.

Even in the worst-case scenario (AI turns out to be useless) we still end up with densely documented codebases and probably extensive test coverage.