Four Modes of Software Evolution for AI Assisted Development

This paper proposes a four-mode framework for AI-assisted software evolution: Creation, Repair, Change, and Refactoring. Each mode is defined by its objective, reference baseline, preservation obligations, execution strategy, and acceptance evidence. Creation establishes new capabilities; Repair corrects deviations from accepted requirements; Change revises required behavior or constraints; and Refactoring improves internal structure while preserving declared observable behavior. The four modes are recurring development activities rather than mandatory stages in a fixed sequence. Execution, evaluation, and feedback connect them into an ongoing software evolution process. Evaluation determines whether to accept a result, continue within the current mode, gather additional evidence, transition to another mode, or stop. The paper describes a minimal work record, a transition protocol, an illustrative development episode, and a proposed longitudinal evaluation design. The framework separates development strategies from organizational roles and is applicable to both single-agent and multi-agent systems. This work is a conceptual proposal. No empirical evaluation is reported, and no improvements in reliability, efficiency, or maintainability are claimed as demonstrated results.