An AI-Enabled Software Development Lifecycle with Governed Agent Execution

Artificial intelligence (AI) contributes to engineering, delivered behaviour and delegated execution. This paper proposes a software development lifecycle (SDLC) integrating people, AI agents and conventional automation. Three independent applicability questions—Assist, Embed and Operate (A/E/O)—distinguish participation from autonomy and maturity. Seven recurring activities connect tasks, verification and handoffs to accountable decisions. Three operational measures cover accepted-change elapsed time, review effort and delivery cost per accepted change alongside quality and safety guardrails. A purposive synthesis of 58 external source records informs fourteen core requirements, reproduced with verification notes in Appendices A–B. A SupportFlow sandbox demonstration exercises selected application controls using sample records and supplied AI-output fixtures; Appendix C identifies the evidence boundaries. Packaged-software adoption remains illustrative. Internal consistency checks and a prospective comparison with existing DevSecOps address different evaluation questions; comparative effectiveness remains untested. Conceptual research preprint, version 4.0. Not a peer-reviewed journal publication or an empirically validated standard. The paper includes Appendices A–C. Supporting materials provide the current criteria, source records and reproducible internal consistency checks. CC BY 4.0 applies to the paper; supporting-file rights are specified in LICENSES.md.