How do they balance AI assistance with preserving the learning curve that makes developers actually competent? And what's their internal take on the "10x developer" myth now that AI can generate 100x the code—does quantity finally beat quality, or are we just debugging faster than ever?