From Code Scarcity To Verification Scarcity: The New Bottleneck in Agentic Software Engineering

I still remember how studying computer science felt in the early 2000s. Coding was a beast that overwhelmed almost everybody. Thinking computationally, then turning that thinking into syntax, was not an easy mental operation. We looked up to talented students who ranked in ACMICPC programming contests; they were known simply as the coders. After graduation, some of us even chose different tech careers just to avoid coding. For those who stayed, producing and debugging code could take weeks, and much of the development lifecycle was organized around waiting for it. Fast-forward to today, and AI agents can generate substantial amounts of code in minutes or hours. Yet many organizations still focus on how much more code or how many more features they can produce. Beyond that hype, a more important question is emerging: are we generating the right change, and will it remain right as the system evolves? The question, then, is no longer simply how much code we can generate, but how much software change we can generate while maintaining justified confidence that it is the right change. This article examines the emerging challenge of establishing that confidence as code generation becomes increasingly abundant.