Building enterprise POS Android apps forced me to find real uses for AI. I began with ChatGPT for architecture advice and scaffolding code. That changed when tools like Cursor integrated chat-to-code directly into my editor. The workflow shifted from manually creating classes to directing their creation, marking a move from simple autocomplete to execution for complex tasks.

My role has evolved from writing code to architecting systems. With cheap code generation, expensive decisions center on patterns and long-term maintainability. I write TDD prompts for the AI, not just files. Critically, I review every line of AI output; reading "foreign code" is now a core skill. Strong docs and automated tests keep the model grounded and me efficient.

The past year was a cycle of trying things, breaking them, and learning the edges. As AI tools rapidly advance, the developer's job is less about syntax and more about strategic oversight. This shift demands a focus on system boundaries and clean version control to manage the flood of machine-generated work.