As AI, particularly Large Language Models (LLMs) and agentic systems, becomes more capable of handling routine coding tasks like scaffolding, test generation, and bug fixing, there is growing anxiety that traditional software engineering (SE) is losing its relevance. However, this paper argues the opposite: the discipline is not shrinking, but rather expanding its scope. Instead of focusing solely on deterministic, executable code, SE is evolving to engineer “semi-executable artifacts.” These are complex blends of natural language, tools, workflows, and controls that rely on human or probabilistic interpretation.
Key Concepts and Findings
To navigate this transition, the authors introduce the Semi-Executable Stack, a six-ring diagnostic reference model. This framework helps teams reason about the expanding scope of their work by mapping out:
- Executable artifacts
- Instructional artifacts
- Orchestrated execution
- Controls
- Operating logic
- Societal and institutional fit
The model is designed to help organizations locate contributions, bottlenecks, and transition points across these adjacent rings. Furthermore, the authors reframe familiar industry objections to AI tools not as reasons to dismiss the technology, but as distinct new engineering targets to be solved.
Conclusions and Takeaways
A major practical takeaway is the introduction of a “preserve-versus-purify” heuristic. This tool aids engineering leaders in deciding which legacy SE processes, controls, and coordination routines should be kept intact and which need simplification or complete redesign for an agentic workflow. Ultimately, this conceptual, agenda-setting paper encourages practitioners to embrace their evolving roles as architects of probabilistic systems, ensuring their hard-won expertise remains highly valuable in a changing landscape.
Mentoring question
How might you apply the ‘preserve-versus-purify’ heuristic to your current development workflows to determine which legacy practices to keep and which to adapt for AI agents?
Source: https://arxiv.org/abs/2604.15468?utm_source=alphasignal&utm_campaign=2026-04-21&lid=rrYke7wqGoScj8cy