Addy Osmani from Google's Chrome Developer Experience team shared insights on the "70% problem" in AI coding during a recent session. Over the last two years, he noted that AI contributes to over 30% of code at Google, yet there's a significant hurdle: AI can generate 70% of a solution quickly, but the last 30%—which includes debugging, security, and integration—remains complex and time-consuming. Trust in AI-generated code is also declining, dropping from 70% to 60% in two years, with about 30% of developers expressing little to no trust in what AI produces. Osmani pointed out a troubling pattern where trying to fix one bug with AI often leads to multiple new issues. He emphasized the importance of understanding the code being generated, as developers must remain in control of the architecture and security of their projects. While some claim high productivity gains from AI tools, these numbers often come from clean, new projects without the technical debt that complicates established systems. He warned that code review has emerged as a bottleneck due to the volume of changes AI can introduce. For junior developers, Osmani advised using AI as a learning tool while ensuring they comprehend the generated code. He highlighted the need for curiosity and a willingness to learn from mistakes, echoing concerns that some individuals previously relied too heavily on platforms like Stack Overflow without fully grasping the solutions they implemented. Understanding the intricacies of code remains essential, especially in an environment where AI can significantly aid but not replace human oversight.