Type: url
Original: https://arxiv.org/pdf/2604.15468
Status: generated
Summary
This arXiv paper, the written companion to a keynote given at the Agentic Engineering 2026 workshop, argues that AI-based agentic systems do not shrink software engineering but expand what it engineers, from executable code toward semi-executable artifacts — prompts, workflows, controls and organizational routines whose enactment depends on human or probabilistic interpretation. It introduces the Semi-Executable Stack, a six-ring diagnostic reference model running from executable artifacts to societal and institutional fit, applies it to three worked cases, and reframes common objections to agentic software engineering as engineering targets. It closes with a preserve-versus-purify heuristic for deciding which legacy software engineering practices to keep and which to simplify or redesign, and describes itself as diagnostic and agenda-setting rather than empirical.