A senior engineer shares a practical framework for deciding when to delegate work to AI and when to retain human ownership. The framework centers on four delegation criteria: work must be verifiable, low risk, boring, and easy to define as done. Drawing from personal mistakes — including shipping poorly understood AI-generated code and running parallel AI tasks that backfired — the author argues for deliberate rather than liberal AI use. Key rules include working on one thing at a time, owning the design phase, critically evaluating AI suggestions, and thoroughly reviewing all AI-generated output before it ships. The author breaks the software lifecycle into three stages (design, implementation planning, implementation) with different levels of AI involvement at each.
