Human-Human-AI Triadic Programming: Uncovering the Role of AI Agent and the Value of Human Partner in Collaborative Learning DOI: XXXXXXX.XXXXXXX Conference: CHI Conference on Human Factors in Computing Systems; April 13–17, 2026; Barcelona, Spain CCS: Do Not Use This Code Generate the Correct Terms
Taufiq Daryanto, Xiaohan Ding, Kaike Ping, Lance T. Wilhelm, Yan Chen, Chris Brown, Eugenia H. Rho · 2026
Read Human-Human-AI Triadic Programming: Uncovering the Role of AI Agent and the Value of Human Partner in Collaborative Learning DOI: XXXXXXX.XXXXXXX Conference: CHI Conference on Human Factors in Computing Systems; April 13–17, 2026; Barcelona, Spain CCS: Do Not Use This Code Generate the Correct Terms by Taufiq Daryanto, Xiaohan Ding, Kaike Ping, Lance T. Wilhelm, Yan Chen, Chris Brown, Eugenia H. Rho (2026) 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-2601-12134.txt.
Introduction
Taufiq Daryanto, Xiaohan Ding, Kaike Ping, and their co-authors study coding work done by two humans plus one AI agent. Past work treats AI as a swap for a human partner. That view misses the social and learning gains of joint work.
Their study asks what the AI adds and what the human partner adds in a trio. Each may play a unlike role in code work and in learning. The results can shape class tools that keep peer learning alive. The work is set for CHI 2026. The trio view may change how class coding tasks are set.
Catalog record
| CREATOR | Taufiq Daryanto, Xiaohan Ding, Kaike Ping, Lance T. Wilhelm, Yan Chen, Chris Brown, Eugenia H. Rho |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 63 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2601-12134
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Human-Human-AI Triadic Programming: Uncovering the Role of AI Agent and the Value of Human Partner in Collaborative Learning DOI: XXXXXXX.XXXXXXX Conference: CHI Conference on Human Factors in Computing Systems; April 13–17, 2026; Barcelona, Spain CCS: Do Not Use This Code Generate the Correct Terms for Your Paper CCS: Do Not Use This Code Generate the Correct Terms for Your Paper CCS: Do Not Use This Code Generate the Correct Terms for Your Paper CCS: Do Not Use This Code Generate the Correct Terms for Your Paper Taufiq Daryanto, Xiaohan Ding, Kaike Ping, Lance T. Wilhelm, Yan Chen, Chris Brown, Eugenia H. Rho Abstract . As AI assistance becomes embedded in programming practice, researchers have increasingly examined how these systems help learners generate code and work more efficiently. However, these studies often position AI as a replacement for human collaboration and overlook the social and learning-oriented aspects that emerge in collaborative programming. Our work introduces human-human-AI (HHAI) triadic programming, where an AI agent serves as an additional collaborator rather than a substitute for a human partner. Through a within-subjects study with 20 participants, we show that triadic collaboration enhances collaborative learning and social presence compared to the dyadic human–AI (HAI) baseline. In the triadic HHAI conditions, participants relied significantly less on AI generated code in their work. This effect was strongest in the HHAI-shared condition, where participants had an increased sense of responsibility to understand AI suggestions before applying them. These findings demonstrate how triadic settings activate socially shared regulation of learning by making AI use visible and accountable to a human peer, suggesting that AI systems that augment rather than automate peer collaboration can better preserve the learning processes that collaborative programming relies on. Figure 1 . Rethinking AI for Programming Learning. Rather than replacing human partners, can AI augment a human–human pair programming while preserving the social and pedagogical benefits of collaboration? To answer that, we explore “human–human–AI triadic programming”, where two humans work together with an AI agent. Our study compares this approach to human–AI pairs and examines design considerations in this triadic interaction, including whether AI should act as a shared collaborator or personal support. 1. Introduction AI assistance has become increasingly prevalent in programming, with tools such as ChatGPT ( 84 ) and GitHub Copilot ( 41 ) now common in everyday development practices ( 109 ) . Prior work shows that these tools can accelerate productivity and improve task completion ( 109 ) , and this trend has extended into educational settings where AI is frequently used for programming activities ( 108 ) . Many studies adopt a human–AI pair programming setup in which a learner works with an AI model on the same task ( 74 ; 36 ) . These studies often position the AI as a replacement for a human partner ( 74 ) , and are frequently compared to collaborative programming practices such as human–human pair programming ( 74 ; 36 ) . Although substituting a peer with an AI can speed up progress and improve coding outcomes ( 36 ; 74 ) , this approach can overlook the social and collaborative dimensions that support learning in peer-based programming between two human partners ( 36 ) . This raises a central question for computing education about whether AI can augment human collaboration in ways that preserve the social and pedagogical benefits of peer learning. Existing studies largely examine the benefits of dyadic human–AI programming, with most work emphasizing speed and code quality as key evaluation metrics ( 74 ; 36 ) . However, these metrics tend to overlook the social and collaborative experiences in learning that shape how students learn ( 81 ) . In collaborative programming, learners work together by articulating their thought processes and sharing knowledge ( 6 ) as they jointly contribute to a shared task ( 82 ) . Even in the era of AI, these collaborative skills remain valuable, particularly as modern work places increasingly adopt human-AI teaming practices ( 49 ; 80 ) where teams of humans and autonomous agents can work together toward shared goals ( 9 ) . Students similarly rely on peers for shared reasoning, clarification, and support during programming activities ( 112 ) , yet they also recognize that AI can provide distinct forms of assistance that are difficult for peers to offer ( 24 ) . To better understand the combined strengths of both human and AI partners, we investigate a human–human–AI triadic programming approach where two human programmers collaborate with an AI agent as a third teammate. Moving from human-AI dyads to a triadic human–human–AI setting introduces distinct design considerations ( 22 ) . First, should the AI act as a shared collaborator for the human-human pair, or as personal support for each human partner ( 55 ; 34 ; 119 ; 71 ) ? Second, beyond providing on-demand answers, what types of AI interventions are useful, for example, should the AI intervene proactively when the pair is stuck ( 87 ; 18 ; 55 ) ? Following recent work that defines proactive support as system-initiated assistance in the programming environment ( 87 ) , we refer to a proactive AI agent as one that has the capability to initiate suggestions without always being explicitly prompted by the user. Prior work shows that such proactive AI behaviors in programming settings can be perceived as helpful partners and improve collaboration experiences, but may also be disruptive ( 87 ; 18 ) . However, such work has focused exclusively on dyadic human–AI interactions. We know little about how a proactive AI agent should participate in synchronous programming with two humans, where conversational flow, role coordination, and mutual accountability add additional layers of complexity ( 51 ; 113 ; 14 ) . Clarifying when and how an AI should intervene, and how its positioning reshapes collaboration between two human partners, remains an open challenge. In this work, we address this gap through the design and evaluation of a human–human–AI triadic programming system that allows us to compare shared versus personal AI positioning and proactive versus on-demand support. To investigate this, we built an integrated workspace that connects two humans and an AI agent in real time, combining a collaborative code editor with a conversational interface that supports text and voice (Figure 2). The system is designed to support collaborative programming practice between the two humans, allowing them to actively engage in continuous dialogue to articulate their thought processes and coding. The AI agent complements this interaction and can assist on demand through hints and code analysis, as well as through lightweight proactive interventions (e.g., providing hints during idle periods), while avoiding premature solution reveal. We then conducted a within-subjects study of 20 participants (10 pairs) to examine key design considerations for human-human-AI triadic programming, such as the scope of AI interactions (shared collaborator versus personal support), the usefulness of different intervention types, and user perceptions towards human-human-AI triadic programming in general. Specifically, we compared three conditions: (1) Shared AI , where the AI addresses the pair as a third collaborator; (2) Personal AI , where each human has an individual AI partner; and (3) Human–AI , a baseline condition where each human