An AI-assisted development workflow
A grounded workflow for using AI as a fast collaborator while keeping architecture, taste, and verification in human hands.
Use AI where feedback is cheap
AI is most useful when the cost of checking its work is low. Generating test cases, exploring APIs, drafting UI variants, and refactoring small modules can all be fast wins.
It is less useful when correctness depends on hidden product context that has not been written down.
Keep architecture human-owned
The model can propose structure, but the engineer should own the boundaries. Naming, ownership, data flow, and failure modes are where product knowledge lives.
I try to make AI accelerate implementation without outsourcing judgment.
Verify like nothing changed
AI-assisted code still needs the normal rituals: tests, builds, screenshots, profiling, and careful review.
The workflow gets faster, but the responsibility does not move.