Addy Osmani, a Google engineer, highlights the dual nature of AI in coding. While AI tools can accelerate code generation, they often produce outputs that lack depth and reliability. Osmani notes that over 30% of code at Google comes from AI suggestions, yet many developers report low trust in these AI-generated solutions, with confidence dropping from 70% to 60% in just two years. The real challenge lies in addressing what he calls the "70% problem," where developers must validate and debug the AI's output. As code becomes more complex, fixing one issue often leads to new problems, making the review process longer and more cumbersome. To tackle these challenges, Osmani suggests that developers take time to understand generated code and document their decisions. He advocates for better context engineering, meaning developers should provide AI with comprehensive project information to improve its output. Writing tests for code becomes even more critical, as these tests can serve as a feedback loop for AI. However, human oversight remains essential; relying solely on AI for testing can lead to oversight of critical issues. The increasing reliance on AI tools is creating a new bottleneck in the code review process. As more code is generated, senior engineers, who are typically in limited supply, face mounting pressure to review this output. This shift indicates a need for evolving code review practices to accommodate the growing volume of AI-assisted coding.