What was important about good software engineering before AI is even more important now with agentic coding.
-
Clear acceptance criteria. Can you articulate exactly what you want to build.
-
Deterministic validation. a.k.a tests. How can you prove the new code does what its supposed to do. How can you prove it didn’t break something?
-
Documentation. specifically in markdown. Your codebase architecture, patterns, do, dont’s, everything needs to be written down, concisely.
-
Seperation of concerns. Context windows are getting bigger yes, but if you split your code at the right edges, LLMs can reason better with a more constrained context.
Agentic coding is force multiplier in both directions which means, codebases that strived for these practices before AI are accelerated. Those that didn’t are even more hindered.