How to Use AI to Help with Software Engineering Tasks (newsletter.eng-leadership.com)

Steven Levey’s new C.R.A.F.T.E.D. prompt framework gives software engineers a repeatable, developer-centric recipe for getting reliable outputs from AI models. The acronym—Context, Role, Action, Format, Tone, Examples, Definition of Done—maps to a typical dev workflow: prime the model with code, errors, and specs (Context); set the expert lens (Role); issue a single, precise task (Action); demand a machine-friendly structure (Format); control voice for the audience (Tone); include few‑shot examples (Examples); and finish with hard constraints (Definition of Done). The article includes concrete prompts and responses (JS refactor, IAM policy review, pytest unit tests, minified JSON output) to show how each element reduces ambiguity and re-rolls. For the AI/ML community this matters because it operationalizes prompt engineering into practices that improve reproducibility, safety, and automation. Technical implications include better priming of models to avoid hallucinations, explicit persona-setting to narrow scope, strict output formats that enable programmatic consumption (e.g., returning only code or a compact JSON), and using constraints at the end to exploit the recency effect. Paired with measurement guidance (DX’s report on how companies track speed, quality, maintainability and novel metrics like “Bad Developer Days”), C.R.A.F.T.E.D. supports integrating LLMs into CI/CD, code review, security checks, and metric-driven evaluation of AI’s engineering impact.