AI coding shifts value from code to problem definition

Developers argue that as AI generates code surpassing human review quality, the bottleneck shifts to defining the right problem. Thorsten Ball predicts code review, unit tests, and agile processes become obsolete. a16z's Josh Elman says product judgment costs remain high despite cheap code generation.
Key insights
Traditional code quality processes are becoming obsolete
Thorsten Ball argues code review is dead because humans cannot review AI-generated code at scale. Unit tests become training wheels when AI produces error-free code (e.g., 900 lines of Arduino C that ran first time). 'Good code' standards designed for human readability become irrelevant when AI reads and modifies code. Open source's 'many eyeballs' now means AI eyeballs.
Human value shifts from coder to architect and problem definer
Ball states the craft of writing code will fade, but the craft of building software (understanding business problems, choosing solutions, timing releases) becomes more important. Most bugs will be 'you asked for the wrong thing' bugs. Engineers who merely shuffle tickets between agents and humans lose value. Performance optimization remains niche.
Product management remains critical as code becomes cheap
Josh Elman contends that code generation cost drops, but the cost of correct product judgment does not. PMs still need to decide what to build, for whom, and why. The role does not disappear; it becomes more strategic.
The terminal and deterministic computing paradigm are ending
Ball predicts the terminal, shell, CLI tools, and even UI will decline. We are entering a 'post-binary era' where everything is rebuilt around tokens. UIs become dynamically generated for 'dumb machines', but as machines grow smarter, UI surface shrinks. Token access becomes the new economic divide.