The paper asks whether AI can make software engineering — the discipline — obsolete, and whether it can make software engineers — the professionals — redundant. Its answer is no on both counts: it argues that the rise of capable AI coding agents makes software engineering and software engineers essential, as the missing foundation without which AI-assisted development produces software that is misleadingly plausible, unverifiable and ultimately untrustworthy.
The argument runs from the nature of large language models to the role of the engineer. The paper identifies three structural properties of these models — probabilistic generation, agnosticism and semantic statelessness — that, in its account, create a “structural vacuum” no amount of training can eliminate. Filling that vacuum, it proposes, requires four knowledge levers, all of which must be reified as persistent artifacts, and each of which requires the software engineer in three roles.
Key Points
- The paper claims that capable AI coding agents make software engineering and software engineers essential, not obsolete.
- Without that foundation, according to the paper, AI-assisted development produces misleadingly plausible, unverifiable and ultimately untrustworthy software.
- It names three structural properties of large language models — probabilistic generation, agnosticism and semantic statelessness — as creating a structural vacuum that no amount of training can eliminate.
- It proposes four knowledge levers to fill that vacuum: methodological knowledge, domain knowledge, design choices and process choices.
- It holds that all four levers must be reified as persistent artifacts, and that each requires the software engineer acting as methodologist, mediator and custodian.
Notes
The paper is a six-page argument rather than an empirical study. It was posted to arXiv on 7 October 2026 and revised on 11 October 2026, the second version being the final one submitted to the publisher, with its first section updated with evidence published since July 2026.