The Contract Discovery Bottleneck

As AI makes code generation cheaper, the primary challenge in software engineering is shifting from implementation to verification and, more importantly, contract discovery. Author Ken Walger argues that while AI can build features based on prompts, it often fails to account for implicit requirements—such as ensuring password reset links are single-use or handling concurrent requests. The author suggests that the real bottleneck is defining what 'correct' behavior actually means. Because humans often lack a complete specification beforehand, the development process must become a cycle of iterative refinement where discovered invariants are captured in a durable, versioned, and reviewable contract. This approach moves beyond simple testing, emphasizing that the definition of correctness should be independent of the implementation. Ultimately, as agents take on more autonomous roles, establishing explicit, inspectable behavioral contracts is essential to prevent systemic failures and ensure that intended outcomes are reliably achieved.