BoothIQ reports using AI to write 150,000 lines of Elixir for its trade show badge scanner. The team found AI excels with Elixir's small, terse syntax, requiring fewer tokens and decisions. Tools like Tidewave help by providing application context, enabling longer unassisted coding sessions.

The main drawback is AI's tendency toward defensive, imperative code. It struggles with OTP concurrency and doesn't grasp Ecto Sandbox test isolation, leading to debugging dead-ends. Human oversight is essential for architecture and correcting these ingrained patterns from its training data.

Despite these friction points, productivity gains are substantial. The key is maintaining a coherent codebase architecture. The goal is to automate the entire development lifecycle, from problem statement to a merge-ready PR, pushing the boundaries of AI-assisted software engineering.