Senior Engineer Reflects on How AI Is Shifting His Role From Builder to Validator

A software engineer writes that over the past year, AI tools have dramatically reduced the time he spends writing code, with first implementations now generated quickly from context and descriptions. He notes that the genuinely difficult work has shifted to evaluating whether AI-generated code actually behaves correctly across real-world edge cases, not just whether it runs or passes tests. This represents a broader inversion in software engineering, where generating an implementation is becoming cheaper while determining whether to trust it is growing more complex and costly. The engineer describes being asked to build an evaluator for a new feature, initially resisting the idea as resembling QA work, but concluding that defining what an evaluator should measure is itself a hard intellectual problem. He argues that the core engineering skill is increasingly about specifying correctness, identifying failure modes, and designing trustworthy checks rather than writing the implementation itself.