AI Tools Are Reshaping How Engineers Plan, Build, and Review Code

Traditional software development required engineers to pre-plan and split work into small pull requests before writing any code, primarily to keep code reviews manageable. AI assistants have dramatically reduced the cost of building, designing, and decomposing large branches of code into smaller PRs after the fact. This shift allows developers to build a complete feature first, demo it early, and only then split the work into reviewable chunks along boundaries the code itself reveals. However, two things remain expensive: the judgment-intensive half of code review and product validation, which still require human understanding and real-world testing. The proposed new workflow emphasizes rigorous upfront planning and spec-writing, but defers structural decomposition until after the feature is built and validated.