The usefulness of an AI-generated answer depends in part on how the task is defined. As large language models become part of professional workflows, designing and evaluating their instructions deserves the same attention as reviewing their outputs.
Kirill Patyrykin, Managing Director of Surety Marine, and Aleksey Dudkin, IT Officer at Surety Marine, explore this subject in "Instruction Architecture in Generative AI: A Framework for Prompt Engineering and Reasoning Control in Large Language Models," published in Acta Innovations on 30 October 2026.
The paper examines prompt engineering as a structured discipline. It brings together the architectural, cognitive and probabilistic foundations of large language models with practical approaches to designing, refining and evaluating prompts. Its scope includes applications in research, software engineering, education and decision-making.
A central concern is the gap between widespread use of prompting and the often informal, trial-and-error methods used to develop instructions. The paper offers a methodological framework for approaching this work more systematically.
For marine insurance, the topic raises a practical question: how should professionals frame tasks for AI when the information is complex and the consequences of an error matter? Clear instructions can help organise the work, but they do not remove the need to check evidence, recognise missing context and review conclusions.
At Surety Marine, we believe technology should strengthen professional judgement. Whether AI is used to support document review or analyse information, experienced professionals remain responsible for interpreting the evidence and making decisions.
Read more about the paper on Kirill Patyrykin’s blog.
← Back to Press Center