The article analyzes the use of AI, specifically Claude, to generate code in Elixir for BoothIQ's lead retrieval app. With 150,000 lines of code, the author highlights both the strengths and weaknesses of relying on AI for software development. The good aspects include Elixir's simplicity, which allows the AI to make fewer decisions and produce cleaner code. Elixir's terse syntax and structure mean that the AI can operate with less context loss during coding sessions. The introduction of Tidewave enhances the AI's capabilities by allowing it to access logs and schemas, leading to fewer errors. On the downside, AI struggles with organizing code architecture. It tends to generate redundant files and inconsistent structures, necessitating human oversight for maintaining coherence. Claude's training on imperative languages leads it to produce defensive code that's at odds with Elixir's functional style. The AI's handling of Git operations is also inefficient, as it consumes valuable context without adding much value. The article dives into the AI's significant shortcomings in debugging asynchronous processes and test failures. Claude's misunderstanding of how tests interact in Elixir can lead to incorrect assumptions about data states, creating confusion. Despite these issues, the author finds that AI-generated code greatly boosts productivity. While the challenges are manageable, maintaining a structured codebase is essential to avoid chaotic development. The goal moving forward is to automate more of the software development process, allowing Claude to handle everything from initial problem statements to creating fully tested pull requests ready for deployment.