Software Development Bottleneck Shifts from Coding to Problem Definition

Thorsten Ball argues that AI-generated code quality exceeds human review, making code review, unit tests, and agile processes obsolete. The bottleneck moves to problem definition; human role shifts from coder to architect.
Key insights
Code Quality Standards Obsolete
Code review is dead because humans cannot find issues in AI code; unit tests may become unnecessary as AI writes perfect code (e.g., 900-line Arduino code that compiled and ran instantly). The concept of 'good code' (readable, clean, format) serves human readers; for AI, such standards are meaningless. Open source still works but eyeballs become AI.
Human Role Shift: From Crafter to Architector
The craft of writing code will disappear, but building software craft becomes crucial: understanding business problems, referring solutions, timing releases. Most bugs become 'defining the wrong thing' bugs rather than coding bugs.
Process and Organization Reshape
PM/design/engineering triopoly becomes obsolete; agile, Scrum, and 'human proxy' engineers lose value. The gap between large and small companies widens; startups copying Google processes seem more ridiculous. Human performance optimization remains marginal; 99% of software has no users.
Tools Paradigm: Terminal Dies, UI Instant Generatin
Terminal, text editors, CLI tools become like perl one-liners. We are bound by 80 years of deterministic computing; entering a post-binary era built on tokens. UI will be instant generated for machines; menus and dashboards are 'API for dumb machines' that will shrink.
Economic Impact: Token as New Production Factor
Token access becomes the new production factor; those without enough tokens compete in second league. Models will not stay cheap and dumb; smarter models make fewer errors and require fewer rounds. Replacement takes a generation, but most will want to be in the new paradigm.