How far can we push AI autonomy in code generation?

Autonomous AI code generation remains challenging despite advances in LLM capabilities. An experiment testing AI's ability to create a Spring Boot application revealed persistent issues even with specialized strategies including multiple agents, stack-specific instructions, code examples, and review cycles. Problems included overeagerness, assumption-filling, brute force fixes, and quality issues flagged by static analysis tools. The future likely involves human-AI collaboration rather than full AI autonomy for business-critical applications.