Before automating anything, I talk through the intended process with AI: where data comes from, what triggers what, and where it should end up. The exchange turns scattered logic into a clear system blueprint that can later be implemented in code.
I learned to break asks into small, verifiable chunks: one file, one function, one step. Big prompts invited confident mistakes; modular prompts reduced errors and sped up iteration. This is the core engineering pattern for co‑building with models.
One day, a week’s worth of refactors collapsed into hours. Chunked tasks plus agentic assistance meant I shipped two steps by lunch and had energy left to test. The gain wasn’t speed alone; it was momentum.