The Semi-Executable Stack: Agentic Software Engineering and the Expanding Scope of SE
A research paper proposes that AI-driven software engineering doesn't threaten the field but rather expands its scope to include 'semi-executable' artifacts—combinations of natural language, tools, and workflows requiring human or probabilistic interpretation. The Semi-Executable Stack model provides a diagnostic framework across six layers to understand how software engineering practices evolve as AI agents handle routine tasks.
The paper addresses legitimate concerns among developers about AI's impact on software engineering by reframing the challenge as an expansion rather than replacement of the field. As large language models and agentic systems handle scaffolding, test generation, and bug fixes, the boundary of what constitutes 'software engineering work' shifts fundamentally. The Semi-Executable Stack model offers practical value by mapping six interconnected layers—from executable code through orchestration, controls, operating logic, to institutional fit—helping practitioners identify where bottlenecks and contributions actually reside.
This conceptual framework emerges from a broader trend of AI systems moving beyond code generation into workflow automation and system orchestration. Traditional software engineering assumed deterministic execution of binary instructions; the new paradigm requires managing probabilistic outputs, natural language specifications, and human-in-the-loop decision points. For developers and organizations, this means expertise doesn't depreciate but transforms. Junior developers worried about automation now face a different problem: learning to engineer systems where interpretation layers matter as much as compilation.
The practical impact extends beyond individual developer anxiety to organizational structure. Teams must develop new processes for validating semi-executable artifacts, managing control mechanisms for probabilistic systems, and integrating AI agents into established workflows. The preserve-versus-purify heuristic the paper introduces becomes critical—determining which legacy processes remain essential versus which ones require fundamental redesign in an agentic context. This diagnostic lens helps organizations navigate technological transition without wholesale rejection of accumulated engineering discipline. Looking forward, successful adoption depends on treating this as an engineering problem requiring systematic methodology, not merely a technology implementation challenge.
- →AI doesn't threaten software engineering but expands its scope to include semi-executable artifacts requiring human or probabilistic interpretation
- →The Semi-Executable Stack model provides six diagnostic layers for understanding how AI agents change software engineering practice
- →Developer expertise remains valuable as engineering work shifts from code execution to managing probabilistic systems and control mechanisms
- →Organizations must systematically redesign legacy processes rather than wholesale adopting or rejecting AI-driven workflows
- →The field faces a fundamental shift from deterministic to probabilistic engineering requiring new validation and coordination approaches