Agent Loop Architectures as Methodological Instruments: A Problematization Review of Hybrid Research Inference in Information Systems

Abstract

Methodological scholarship in information systems (IS) and the social sciences rests on three assumptions that have rarely faced direct empirical challenge: that data type ought to govern method selection, that interpretation is constitutively tied to human agency, and that research inference must advance through a defined sequence of stages. This paper applies a problematization review to 94 studies published between 2018 and 2025, testing each assumption against the documented behaviour of AI agent loop architectures. Analysis centres on the Anthropic Claude Code queryLoop() function as an engineering realisation of iterative interviewing logic, and extends to a cross-domain case study of recursive speaker diarization optimisation for Sarawak Malay, which demonstrates that all three assumptions are contested well beyond text-centric AI systems. The paper concludes by proposing four research pathways: validity theory for data-type-agnostic inference, human-agent interpretive division of labour, reporting standards for recursive research processes, and ethical governance of autonomous methodological agents.