The emergence of autonomous AI software engineers—defined as agentic systems capable of participating in full software engineering (SE) lifecycle activities beyond mere code generation 1—has precipitated a fundamental jurisprudential paradox. These systems now possess the technical capacity to execute socio-technical responsibilities traditionally requiring fiduciary judgment, including requirements negotiation, architectural design under ethical constraints, and long-term maintenance decisions affecting human welfare. Yet they remain devoid of moral patiency, legal personhood, and the capacity to bear ethical responsibility. This creates the Fiduciary Compression Paradox: the ontological compression of professional ethical duties into executable technical specifications without corresponding expansion of ethical jurisdiction or accountability mechanisms. While I precedently 2 addressed the prudential risk implications of autonomous coding through the lens of financial regulation and causal attribution deficits, this paper distinguishes itself by examining the ethical boundary conditions where AI agents assume professional responsibilities requiring fiduciary duties of care, loyalty, and confidentiality. The proposed framework addresses risk allocation in capital markets; this analysis addresses moral agency limits in professional practice. I argue that current governance frameworks commit a category error by treating ethical constraints as verifiable technical properties—dimensions of "trustworthiness" 1—rather than jurisdictional limits inherent to the moral status of the agent. This paper proposes that the limits of AI software engineering are not merely technical or prudential, but fundamentally ethical and fiduciary in nature.
Marcel Osmond (Fri,) studied this question.