Hyper-automated research: AI as a co-pilot in the scientific method

This paper presents an autoethnographic case study of a custom AI research co-pilot developed and deployed by authors, university professors and senior ICT professionals working in artificial intelligence, across three research projects between January 2024 and June 2026. The co-pilot, a three-layer system comprising a retrieval and grounding layer, a reasoning and generation layer, and a verification and provenance layer, was integrated into the full research lifecycle: literature synthesis, hypothesis formulation, experimental design, data analysis, drafting, and revision. The study documents substantial efficiency gains-literature synthesis time reduced by approximately 71%, time-to-first-draft by 44%, and total project duration by 28%-alongside five ethical dilemmas that emerged from practice rather than from anticipation. These comprise: the comprehension debt problem, in which the author published work whose full analytical basis he did not personally hold; the authorship attribution problem, which resisted resolution under existing institutional policy; the peer review integrity problem, in which the author both received and was asked to produce reviews of contested provenance; the reproducibility problem, arising from non-deterministic analytical pipelines; and the literature dilution problem, in which AI-assisted production increased submission volume without corresponding increases in review capacity. The paper argues that hyper-automated research is not merely faster research but epistemically different research, and that the scientific community has not yet developed the norms, verification infrastructure, or institutional governance required to sustain it. Seven practical recommendations are advanced.