AI Shifts Software Development Bottleneck from Coding to Problem Definition

Thorsten Ball and a16z's Josh Elman argue that as AI code generation becomes dominant, the bottleneck in software development is shifting from writing code to defining the right problem. Ball suggests traditional practices like code review and unit tests may become obsolete, while Elman emphasizes that the cost of correct product judgment remains high.
Key insights
The Death of Traditional Code Craft
Thorsten Ball argues that when AI-generated code surpasses human review capability, traditional practices like code review, unit tests, and standards for 'good code' (readability, formatting) will lose their foundation. He states that most bugs will no longer be coding bugs, but 'you asked for the wrong thing' bugs. The terminal, CLI tools, and text editors will become as archaic as Perl one-liners.
The New Value: Architecture and Problem Definition
Ball argues that the 'craft of writing code' will disappear, but the 'craft of building software' will become more important than ever. This includes understanding business problems, system architecture, release timing, and feedback loops. The engineer's role transforms from a coder to an architect. a16z's Josh Elman reinforces this, stating that while code construction costs are dropping, the cost of correct product judgment has not decreased at all.
Token is the New Compute Paradigm
Ball posits that we are entering a 'post-binary' era where everything will be rebuilt on top of tokens. Token cost is the new unit of compute. He argues that smarter models are always better because they make fewer errors, and that the real divide will be between those who have access to enough tokens and those who do not. UI will be generated on the fly, as menus and dashboards are just 'human APIs for dumb machines'.