AI in Software Development: From Code Writing to Problem Definition

Thorsten Ball argues that as AI-generated code surpasses human review capability, traditional software engineering practices like code review, unit tests, and agile processes will become obsolete. The new bottleneck is problem definition, with humans transitioning from craftsmen to architects. Token becomes the new computing paradigm.
Key insights
Traditional Code Quality Standards Become Obsolete
Thorsten Ball states that code review is dead because humans cannot find issues in AI-generated code in reasonable time. Unit tests may become unnecessary as models produce error-free code. The concept of 'good code' (readable, formatted) is for human readers; when AI is the primary modifier, these standards lose meaning. Open source will still work, but with AI eyes instead of human.
Human Role Shifts from Craftsman to Architect
The craft of writing code will disappear, but the craft of building software becomes more important: understanding business problems, leveraging existing solutions, and managing feedback loops. Most bugs will be 'you asked for the wrong thing' rather than coding errors. Performance optimization remains but only for edge cases.
Existing Collaboration Paradigms Will Dissolve
The PM/design/engineering tripartite division becomes unreasonable. Agile and Scrum will die. Engineers who merely pass tickets to agents and report back lose value. The gap between large and small companies widens as startups can move faster with fewer constraints. Token becomes the new means of production.
Product Managers Remain Essential for Problem Definition
a16z's Josh Elman argues that while code generation costs drop, the cost of making correct product decisions does not. PMs are still needed to define what to build. The ability to understand user needs and prioritize features becomes even more critical as prototyping becomes cheaper.