“Having Lunch Now” : Understanding How Users Engage with a Proactive Agent for Daily Planning and Self-Reflection

Adnan Abbas, Caleb Wohn, Arnav Jagtap, Eugenia H Rho, Young-Ho Kim, Sang Won Lee · 2025

Read “Having Lunch Now” : Understanding How Users Engage with a Proactive Agent for Daily Planning and Self-Reflection by Adnan Abbas, Caleb Wohn, Arnav Jagtap, Eugenia H Rho, Young-Ho Kim, Sang Won Lee (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-24073.txt.

Introduction

Adnan Abbas and 5 co-authors watched how people live with a chat agent that speaks first. For two weeks, twelve volunteers used an agent that checked in to help plan the day and reflect on it. The agent acted as a coach, not a clerk that waits for orders.

People took up tips, pushed back on them, or haggled over terms. Over time, some built a shared sense of how the agent thinks. Others drifted away or resisted. The agent also misstepped: it could be rigid, cut in too soon, and promise more than it could do. The authors draw design lessons for agents that aim to change habits.

Catalog record

CREATORAdnan Abbas, Caleb Wohn, Arnav Jagtap, Eugenia H Rho, Young-Ho Kim, Sang Won Lee
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE58 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2509-24073

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

“Having Lunch Now” : Understanding How Users Engage with a Proactive Agent for Daily Planning and Self-Reflection Adnan Abbas, Caleb Wohn, Arnav Jagtap, Eugenia H Rho, Young-Ho Kim, Sang Won Lee Abstract . Conversational agents have been studied as tools to scaffold planning and self-reflection for productivity and well-being. While prior work has demonstrated positive outcomes, we still lack a clear understanding of what drives these results and how users behave and communicate with agents that act as coaches rather than assistants. Such understanding is critical for designing interactions in which agents foster meaningful behavioral change. We conducted a 14-day longitudinal study with 12 participants using a proactive agent that initiated regular check-ins to support daily planning and reflection. Our findings reveal diverse interaction patterns: participants accepted or negotiated suggestions, developed shared mental models, reported progress, and at times resisted or disengaged. We also identified problematic aspects of the agent’s behavior, including rigidity, premature turn-taking, and overpromising. Our work contributes to understanding how people interact with a proactive, coach-like agent and offers design considerations for facilitating effective behavioral change. 1. Introduction With the presence of large-language models (LLMs), Conversational Agents (CAs) are becoming more human-like, exhibit better conversational intelligence and novelty in interactions ( Rayan et al. 2025 ; Li et al. 2025 ; Sorin et al. 2024 ) . LLM-powered CAs can interpret intent, adapt their tone, and maintain contextual continuity across multiple sessions ( Mahmood et al. 2025 ; Liu et al. 2024 ) . Leveraging these capabilities, researchers have studied coaching-style agents: systems designed to advise, persuade, and guide everyday behavior using the breadth of their knowledge ( Jörke et al. 2025 ; Liu et al. 2024 ; Arakawa and Yakura 2024 ; Wilhelm et al. 2025a ) . Oftentimes, such agents are proactive in a way that they initiate interactions by reminding, suggesting, or even steering users toward specific actions ( Liu et al. 2024 ; Xu et al. 2025 ; Baraglia et al. 2016 ) , compared to reactive agents that respond only when prompted. Studies show that proactivity in CAs has several utilities in human–AI interaction: introducing new topics ( Liu et al. 2025 ) , offering richer suggestions ( Peng et al. 2019 ) , and persuading users toward target behaviors ( Lei et al. 2022 ) . One domain where proactive CAs have been employed is supporting workers’ productivity and well-being ( Kimani et al. 2019 ; Ahire et al. 2024 ; Kocielnik et al. 2018a ; Williams et al. 2018 ; redacted for anonymity [n. d.] ) . In these studies, rather than CAs responding to a user’s request, CAs take the role of a productivity coach or well-being coach and regularly prompt workers to articulate plans or reflect on their day, proactively leading the conversation. By initiating these check-ins, CAs can help make users’ intentions more concrete and actionable, thereby improving adherence to planning ( Luo et al. 2020 ; Kimani et al. 2019 ; Grover et al. 2020 ) . Through reflective prompts, these agents can increase users’ awareness of everyday habits and foster behavior change ( Mols et al. 2016 ; Kocielnik et al. 2018a ; Ho et al. 2018 ) . Many of these systems relied on daily check-ins (e.g., morning or evening check-in) with a specific purpose, such as to remind users of and revisit their initial plans or progress ( Kocielnik et al. 2018a ; Kimani et al. 2019 ; Ahire et al. 2022 ; redacted for anonymity [n. d.] ) . These studies report positive outcomes such as greater engagement with work, improved work–life balance, better physical health, and more structured planning ( Kocielnik et al. 2018a ; Grover et al. 2020 ; Kimani et al. 2019 ; Ahire et al. 2024 ; Ahire et al. 2022 ; redacted for anonymity [n. d.] ) . However, we still lack an understanding of how people actually communicate with CAs, leaving important gaps in knowledge about the mechanisms through which CAs support workers. This gap limits the ability of prior work to inform how CAs should facilitate desirable behavioral changes in the context of productivity and well-being. To address these gaps, we aim to closely observe and investigate how users communicate with a proactive coach that checks in twice per day to facilitate planning in the morning and self-reflection in the evening. We use thematic and dialogue act analyses of user interactions to examine how users cooperatively interact with CAs in order to understand the mechanisms by which a proactive agent facilitates desirable changes in productivity and well-being. Other studies have used conversation analysis to provide deeper insight into how CAs and users communicate ( Rapp et al. 2023 ; Li et al. 2020 ; Seering et al. 2020 ; Porcheron et al. 2018 ; Pelikan and Broth 2016 ) . This method is particularly well-suited for uncovering the sequential and interactional dynamics of dialogue ( Meredith 2019 ; Liddicoat 2021 ) , allowing us to capture subtle cues in how proactivity is received and negotiated in real use. Specifically, we ask the following research questions: • RQ1 : How do users engage with a proactive agent that facilitates daily planning and reflection? • RQ2 : What conversational behaviors of the coaching agent cause communication breakdowns in conversational interaction? To answer these questions, we conducted a two-week field deployment of a proactive, coaching-style agent (PITCH) with 12 graduate students, analyzing 336 conversations (3,181 turns) through thematic and dialogue act analysis. Our analysis shows that the most common user behaviors were planning upcoming tasks, reporting progress, reflecting on routines, and sometimes ignoring the agent’s prompts. Users often cooperated by adopting suggestions or providing contextual explanations when advice did not fit, while also refusing or negotiating to assert their own work styles. Notably, participants complied more often when the agent pursued explicit agendas, but they also refused more often when those agendas felt misaligned. Beyond scheduled check-ins, users treated the chat as an external memory—recording plans, reflecting on habits, making commitments to change, and even offering unsolicited updates as a running log of their day. Finally, we identified problematic aspects of the agent’s behavior, including rigidity, premature turn-taking, and overpromising its capabilities, which point to design opportunities for more flexible and context-aware coaching agents. This study makes four key contributions: (1) Empirical insight: We provide an in-situ account of how users communicate with a proactive coaching agent in daily work contexts, based on a two-week field deployment with graduate students. (2) Conversation analysis: We conduct a fine-grained analysis of 336 interactions (3,181 turns), uncovering how users cooperate, negotiate, or resist agent prompts, and how these behaviors shape productivity and reflection. (3) Dataset: We release an anonymized dataset of 336 conversations (3,181 turns), which can serve as a resource for future research on conversational agents, proactive interaction, and productivity support, available at [redacted for anonymity]. (4) Design implications: We identify opportunities (e.g., leveraging explicit agendas and external memory functions) and challenges (e.g., rigidity, premature turn-taking, and overpromising) for designing proactive CAs that better support planning and self-regulation. 2. Related Work 2.1. Planning and Reflection with Conversational Agents at Work HCI researchers have explored the role of conversational interaction to

Read the complete text