The energy consumption of households accounts for approximately 30% of the total global energy consumption, leading to a significant portion of CO2 emissions from energy production. Enhancing energy efficiency by managing demand, such as through load shifting, presents a viable strategy for reducing CO2 emissions. This study introduces an innovative activity-based multi-agent recommendation system aimed at reducing CO2 emissions in households. By shifting household activities rather than individual appliance usage, we propose a more intuitive approach to energy efficiency grounded in the social practices of domestic life. Using real-world data, the system provides personalized, actionable recommendations. Our contributions encompass the development of an Activity Agent, the introduction of a performance measure, and a practical implementation strategy requiring minimal user input. Our approach not only encourages sustainable behavior among households but also contributes to the IS field by demonstrating how AI can play a pivotal role in addressing climate change challenges.

Technical advances in machine learning (ML) and artificial intelligence (AI) are shaping the transformation in organisations, society and research. Yet, adoption lags behind as implementation is costly and requires experts which are scarce on the market. Automated ML (autoML) promises to overcome these barriers and help to democratize ML by empowering domain specialists to develop ML models in an easy and cheap way. However, the usage of autoML by non-experts in science raises concerns regarding reproducibility, undermining research credibility. This paper examines the extent to which users without in-depth ML knowledge are supported by no-code autoML tools in ensuring research reproducibility. The results of this study uncover human-related and tool-related opportunities and challenges. Addressing these requires a multifaceted design-oriented approach that incorporates open science principles. In this way, the full potential of no-code autoML tools can be realized while ensuring reproducibility and ultimately the credibility of research.

User frustration is one negative consequence of human-computer interaction caused by bad interpretations and insufficient adaptation to user preferences. In this scope, genetic algorithms (GAs) might offer some insights to mitigate this problem. We conducted a systematic review to identify the implementation of GAs in the field of the design of conversational agents (CAs). Our results displayed that the literature focuses on three clusters mainly using evolutionary algorithms, and binary-coded GAs for natural language processing (NLP).