Pith. sign in

REVIEW 4 major objections 4 minor 46 references

The Help Ladder: Skill-Adaptive Peer Scaffolding for Real-Time Collaborative Programming

T0 review · 4 major / 4 minor · reviewed 2026-08-01 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A shared IDE that breaks coding problems into four difficulty levels makes students help each other more often and faster.

desk verdict A promising system design with a headline claim that outruns the evidence in a small, under-controlled evaluation. read the letter →

arxiv 2607.23031 v1 pith:XHB7O2VM submitted 2026-07-25 cs.HC

Panayu Keelawat , Sriram Narlapati , Sangwook Lee , Darshan Nere , Griffin Ogura , Xinran Adeline Li , Sang Won Lee , Yan Chen This is my paper · ORCID
classification cs.HC
keywords peerscaffoldingcollaborativeprogramminghelpladderprogressivedisclosureLLMreal-timeIDEcomputer-supportedcooperativework
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper argues that real-time collaborative programming breaks down into parallel individual work not because students are unwilling to help, but because entering a teammate's problem requires too much context at once. To test this, the authors built Canary, a shared IDE that uses an LLM to classify a stuck student's obstacles into a four-level 'Help Ladder' — structural, runtime, local logic, and requirement issues — and surfaces them to available teammates as lightweight starting points that become progressively deeper. In a within-subject study of 12 students in four teams, Canary produced more help sessions (6.00 vs. 2.75), more resolved issues (6.25 vs. 1.50), faster time-to-intervention (0.99 vs. 3.29 minutes), and lower reported effort to initiate help (2.58 vs. 3.58). The paper claims that this ladder, not merely proactive notifications, is what made helping feel manageable.

What carries the argument

The Help Ladder: a four-level taxonomy of programming obstacles ordered by the amount of contextual understanding a helper needs — (1) structural issues like syntax errors, (2) runtime issues like exceptions and infinite loops, (3) local logic issues where code runs but returns wrong results, and (4) requirement issues where the implementation misses the task's intent. The system surfaces lower levels first, provides AI-generated multiple-choice fixes only to helpers who have already shown mastery of the relevant concept, and inserts helpers' contributions as code variants or contextual comments in the helpee's editor. The ladder carries the argument by giving helpers a low-cost first step a

What would settle it

Measure LLM classification accuracy against human-annotated issues from the study's code artifacts; if the classifier is no better than chance at distinguishing the four levels, the ladder's ordering cannot be the causal driver. Alternatively, run the same study with ladder levels shuffled and check whether the behavioral gains disappear.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that decomposing a programming obstacle into a progression of increasingly context-heavy sub-issues — from one-line syntax problems to whole-requirement misunderstandings — lowers the barrier to entry for peer help. The system continuously monitors editor activity to identify who is stuck, asks the LLM to sort their issues into the four ladder levels, and shows helpers an entry point at the level requiring the least context, with multiple-choice suggestions offered when the helper has demonstrated relevant skill. Helpers who begin with a small, concrete fix tend to continue into deeper reasoning, turning single help sessions into multi-issue rescue

Load-bearing premise

The four ladder levels reflect a genuine, monotonic increase in the context a helper needs, and the LLM classifies each issue into the correct level.

Editorial extensions

If this is right

  • More frequent and faster peer help means fewer tasks get stuck: average tasks completed rose from 0.5 to 2.0 across groups.
  • Help sessions become multi-issue: Canary teams resolved 6.25 issues across 6.00 sessions, while baseline resolved 1.50 across 2.75, so a single session can address several related sub-problems.
  • Initiation effort drops most among workload measures, suggesting the design targets the right barrier.
  • The same progressive-reveal principle could apply beyond coding to other real-time collaborative tasks that require entering a teammate's context.
  • AI assistance is positioned as a facilitator of human-to-human scaffolding rather than a direct solution provider.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The ladder's causal role hinges on LLM classification quality, which the paper does not measure; if the classifier simply labels most issues as Level 1, the observed gains could be due to help visibility alone.
  • A testable extension would run the same system with ladder levels shuffled; if the ordering itself does not matter, the monotonic context assumption is not what drives the effects.
  • The four-level taxonomy might generalize to any task where a small, well-defined first step into someone else's work can be specified — code review, document editing, or scientific debugging.
  • Because one group (G3) already knew each other and showed smaller gains, the ladder may matter most for teams of strangers, while real classroom groups with established trust might see dampened effects.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 4 minor

Summary. This paper presents Canary, a collaborative IDE that supports real-time peer scaffolding by detecting when a student is stuck and generating a four-level 'Help Ladder' (structural, runtime, local logic, requirement). The system surfaces these levels progressively to potential helpers, with expertise-adaptive multiple-choice suggestions for low-level issues and open-ended guidance at higher levels. A within-subject study with four teams (n=12) compared the full system with a baseline 'without the adaptive peer scaffolding features,' reporting more help sessions, more issues identified/resolved, faster time-to-intervention, and lower help-initiation effort. The paper argues that the staged ramp-up reduces helpers' cognitive cost and increases collaboration.

Significance. If the causal claim were established, this would be a useful contribution to CSCW and computing-education research: the design insight that helpers need low-context entry points is well motivated by the formative studies, and the system is thoughtfully built. The qualitative data and honest reporting of stalled escalation at higher levels are strengths. However, the present evaluation does not isolate the progressive ladder from the general act of detecting and surfacing issues and suggestions, and a key outcome ('issues identified') is partly system-generated. The central claim needs additional experimental control or a more cautious framing; as it stands, the paper is promising but not yet definitive.

major comments (4)
  1. [Section VII-A, Table I] The evaluation contrasts the full CANARY system with a baseline 'without the adaptive peer scaffolding features.' This design cannot isolate the Help Ladder's progressive low-to-high-context ordering: the baseline lacks not only the staged reveal but also the issue-surfacing/suggestion mechanism, so the observed gains (6.00 vs. 2.75 help sessions, 1.50 vs. 6.25 resolved issues) could be produced by merely presenting detected issues and easy suggestions. A flat condition that reveals the same issues and suggestions in a single step is required to support the abstract's causal claim about the staged ladder. The paper's own data (Section VII-B2: fewer than 20% of higher-level issues resolved; roughly two-thirds of initial interventions low-context) suggest helpers acted mostly on the easy, immediately visible suggestions, which do not depend on progressive ordering.
  2. [Section VII-B1, Table I] The 'Issues Identified' dependent variable is partly generated by CANARY: the system decomposes obstacles into multiple layers, so the count (9.25 vs. 2.75) reflects system output rather than human scaffolding activity. In the baseline, issues identified presumably came from human help interactions. This measurement asymmetry invalidates 'issues identified' as a behavioral outcome. The 'issues resolved' outcome is also conflated where resolution is achieved by applying a system-suggested multiple-choice variant. Please report human-authored issue identifications separately from system-surfaced ones.
  3. [Section VI-C] The four-level Help Ladder is load-bearing, but the paper provides no validation of the LLM's level classification (no accuracy, confusion matrix, or inter-rater reliability) and no evidence that the ordering (structural < runtime < local logic < requirement) corresponds to a monotonic increase in helper context. The categories are described as derived from formative-study artifacts, but the monotonicity assumption is asserted. Without this validation, the observed effects cannot be attributed to the progressive, skill-adaptive ramp-up; they may be due to the mere availability of categorized suggestions.
  4. [Section VII-B, Table I] All inferential claims rest on n=4 teams, yet no effect sizes or confidence intervals are reported. Moreover, the 'Tasks Completed' comparison (0.50 vs. 2.00) is not tested or flagged as non-significant, despite the Introduction asserting 'a higher number of tasks completed overall.' Please report effect sizes (e.g., Cohen's d for paired comparisons) and provide an explicit test (or remove the claim) for tasks completed. Given the small sample, the robustness of the p-values should be discussed.
minor comments (4)
  1. [Section VII-B2] The statement that 'fewer than 20% of higher-level issues were resolved' would be more useful with per-level counts or a small table; currently no numbers are visible for the reader to verify.
  2. [Section VII-B1] Please describe how 'becoming stuck' was operationalized in the baseline condition (e.g., manual video coding, inter-rater reliability). Without this, the time-to-intervention comparison may be biased by different detection mechanisms in the two conditions.
  3. [Abstract] The abstract states that 'this staged approach makes helping feel less overwhelming,' but only help-initiation effort reached significance; several workload dimensions did not differ significantly. Consider tempering the claim or adding a qualifying phrase.
  4. [Section I] Minor typo: 'CANARYuses' should be 'CANARY uses' (the same spacing issue appears elsewhere in the text).

Circularity Check

1 steps flagged · score 6.0 of 10

Issues-identified metric is generated by the Help Ladder itself, making a key outcome self-referential.

  1. self definitional [Section VII-B1, paragraph following Table I]
    "CANARY also supported the identification of more issues compared to the baseline. In the baseline condition, the number of issues identified closely matched the number of help sessions, suggesting that each help interaction typically focused on a single visible problem. In contrast, CANARY surfaced more issues than the number of help sessions because the Help Ladder decomposed programming obstacles into multiple layers."

    In the CANARY condition, 'issues identified' is the count of issues the system itself decomposes and surfaces via the Help Ladder. The independent variable (the ladder's decomposition) directly produces the dependent variable (number of issues identified). Thus the claim that the Help Ladder supports the identification of more issues is true by construction, not an empirical finding. The baseline count is human-identified issues, so the comparison conflates system-generated output with human observation.

full rationale

The paper's core claims of more help sessions, faster time-to-intervention, and lower help-initiation effort are measured from recorded interactions and surveys, which are not directly generated by the system and therefore carry independent empirical content. However, the 'issues identified' metric in the CANARY condition is literally the system's own decomposition output, so this part of the effectiveness claim reduces to the intervention by definition. Additionally, the evaluation bundles the progressive ladder with multiple-choice suggestions and code variants, so the ladder's specific ordering is not isolated from the AI-generated fixes, though that is a confounding issue rather than a circular reduction. One clear self-definitional step is present, warranting a partial circularity score.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

Canary's claims are empirical, not derivational. The only numeric parameters are system thresholds chosen by hand; none are fit to the study outcome data, though they shape the intervention. The load-bearing assumptions are the unvalidated four-level classification, the LLM's classification reliability, and the expertise proxy. No new unobservable entities are posited; Canary and the Help Ladder are concrete system artifacts directly instantiated in the described UI and evaluation.

free parameters (4)
  • Repeated failure detection threshold = 3 identical errors in a 2-minute window
    Section VI-B: used to infer a student cannot resolve an issue independently; chosen by hand with no sensitivity analysis.
  • Time-on-task stuck threshold = 7 minutes on the same function
    Section VI-B: after 7 minutes the student is flagged as a potential helpee; arbitrary and uncalibrated.
  • Semantic comment analysis interval = 1 minute
    Section VI-B: comments are scanned once per minute for confusion expressions; chosen by hand.
  • Natural breakpoint inactivity threshold = 30 seconds of no keyboard input
    Section VI-B: help opportunities are surfaced at workflow breakpoints; arbitrary.
assumptions (5)
  • domain assumption Scaffolding theory and the foot-in-the-door principle transfer from tutoring psychology to synchronous, multi-user code-editing contexts.
    Section VI-C borrows scaffolding and foot-in-the-door literature to justify the Ladder; the size of the transfer is assumed, not measured.
  • ad hoc to paper The four-level ordering (structural < runtime < local logic < requirement) monotonically increases the contextual understanding a helper needs.
    Section VI-C: categories are 'derived from formative study code artifacts'; no validation of ordering, no inter-rater reliability.
  • domain assumption The LLM can reliably classify issues into these four levels and generate correct, educationally appropriate suggestions.
    Section VI-C and interface description; no accuracy, error analysis, or prompt details are given.
  • ad hoc to paper A helper's expertise can be approximated by the set of concepts they have previously implemented in their personal editor.
    Section VI-C1: expertise-adaptive questioning uses this proxy; the proxy is unvalidated and the adaptation effects are never isolated.
  • domain assumption Within-subject counterbalancing of conditions and tasks removes learning, order, and fatigue effects in 20-minute tasks within a 1.5-hour session.
    Section VII-A: one 5-minute warm-up per condition; no explicit check for carryover or fatigue effects.

how reviews work

0 comments
Cite this review

Pith. "Pith review of The Help Ladder: Skill-Adaptive Peer Scaffolding for Real-Time Collaborative Programming." pith.science (2026). https://pith.science/paper/XHB7O2VM

@misc{pith2026260723031,
  author       = {Pith},
  title        = {Pith review of: The Help Ladder: Skill-Adaptive Peer Scaffolding for Real-Time Collaborative Programming},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XHB7O2VM}},
  note         = {Machine review of arXiv:2607.23031}
}
read the original abstract

Collaborative programming is a widely adopted classroom activity to encourage peer scaffolding, yet real-time collaboration often breaks down into parallel individual work with minimal interaction. Our formative studies reveal that even when students want to collaborate, they are held back by the effort required to understand a teammate's entire problem at once. We present Canary, a system that supports peer scaffolding by breaking down programming obstacles into smaller steps tailored to a student's skill level. Canary alerts potential helpers to specific places where they can start, using AI to turn complex problems into a step-by-step ladder that starts with easy fixes before moving toward harder logic. By providing this gradual ramp-up, Canary enables students to make quick contributions and progressively work toward solving their peers' problems. Our evaluation shows that this staged approach makes helping feel less overwhelming, leading to more frequent and effective collaboration among students.

Figures

Figures reproduced from arXiv: 2607.23031 by the authors.

Figure 1
Figure 1. We present CANARY, a collaborative programming environment that supports peer scaffolding through progressive contextual guidance. When a student encounters difficulty during real-time collaborative programming, CANARY decomposes the obstacle into a layered Help Ladder consisting of structural, runtime, local logic, and requirement-level issues. This allows helpers to enter through lightweight, low-context contribut… view at source ↗
Figure 2
Figure 2. Needfinding procedure consisting of Phase 1 (collaborative program [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. User interface of the early version of the system used in the preliminary [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Interface of CANARY. (A) Flag for Help button for explicitly requesting assistance from peers. (B) Current user indicator displaying the user’s profile picture, name, and participant ID. (C) Shared session timer showing the remaining activity time. (D) Tabs for navigat…
Figure 5
Figure 5. Figure 5: Example of a helper’s view displaying issue highlighting and scaffolding interactions across different levels of the Help Ladder for the function [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: Example of low-context peer scaffolding through a multiple-choice [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: Comparison of collaboration-specific workload ratings [46] between [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

46 extracted references · 2 canonical work pages

  1. [1]

    Teaching programming in higher education: a bibliometric analysis of trends, technologies, and pedagogical approaches,

    M. Vinueza-Morales, J. Rodas-Silva, C. Vidal-Silva, J. C ´ordova- Mor´an, and E. Cevallos-Ay ´on, “Teaching programming in higher education: a bibliometric analysis of trends, technologies, and pedagogical approaches,”Frontiers in Education, vol. V olume 10 - 2025, 2025. [Online]. Available: https://www.frontiersin.org/journals/ education/articles/10.3389...

    arXiv 2025

  2. [2]

    Self- efficacy and behavior patterns of learners using a real-time collaboration system developed for group programming,

    T.-C. Hsu, H. Abelson, E. Patton, S.-C. Chen, and H.-N. Chang, “Self- efficacy and behavior patterns of learners using a real-time collaboration system developed for group programming,”International journal of computer-supported collaborative learning, vol. 16, no. 4, pp. 559–582, 2021

    2021

  3. [3]

    Currents trends in use of collaborative learning in teaching of robotics and pro- gramming – a systematic review of literature,

    J. Moraes de Carvalho and J. Francisco de Magalh ˜aes Netto, “Currents trends in use of collaborative learning in teaching of robotics and pro- gramming – a systematic review of literature,” in2020 IEEE Frontiers in Education Conference (FIE), 2020, pp. 1–8

    2020

  4. [4]

    Peer instruction: Ten years of experience and results,

    C. H. Crouch and E. Mazur, “Peer instruction: Ten years of experience and results,”American journal of physics, vol. 69, no. 9, pp. 970–977, 2001

    2001

  5. [5]

    Real-time collaborative programming in undergraduate education: A comprehensive empirical analysis of its impact on knowledge, behaviors, and attitudes,

    D. Sun and F. Xu, “Real-time collaborative programming in undergraduate education: A comprehensive empirical analysis of its impact on knowledge, behaviors, and attitudes,”Journal of Educational Computing Research, vol. 63, no. 1, pp. 33–63, 2025. [Online]. Available: https://doi.org/10.1177/07356331241295739

    work page doi:10.1177/07356331241295739 2025

  6. [6]

    The impact of group discussion and formation on student performance: An experience report in a large CS1 course,

    T. Wu, X. Tang, S. Wong, X. Chen, C. A. Shaffer, and Y . Chen, “The impact of group discussion and formation on student performance: An experience report in a large CS1 course,” inProceedings of the 56th ACM Technical Symposium on Computer Science Education, Vol. 1, 2025, pp. 1260–1266

    2025

  7. [7]

    Self-selected groups vs. random groups: An analysis of student engagement, achievement, and preferences in collaborative learning,

    A. S. D. Martha, S. Widowati, A. Aurelvia, S. H. Salma, and M. D. Adani, “Self-selected groups vs. random groups: An analysis of student engagement, achievement, and preferences in collaborative learning,” inCollaboration Technologies and Social Computing, I.-A. Chounta, H. Egi, A. Nugraha, H. B. Santoso, T. Inoue, and T. A. Salim, Eds. Cham: Springer Nat...

    2026

  8. [8]

    DBox: Scaf- folding algorithmic programming learning through learner-LLM co- decomposition,

    S. Ma, J. Wang, Y . Zhang, X. Ma, and A. Y . Wang, “DBox: Scaf- folding algorithmic programming learning through learner-LLM co- decomposition,” inProceedings of the 2025 CHI Conference on Human Factors in Computing Systems, 2025

    2025

Show all 46 references
  1. [9]

    Transforming procedural instructions into in-situ augmented reality guides with instructar,

    P. Keelawat and R. Suzuki, “Transforming procedural instructions into in-situ augmented reality guides with instructar,” inAdjunct Proceedings of the 37th Annual ACM Symposium on User Interface Software and Technology, ser. UIST Adjunct ’24. New York, NY , USA: Association for...

    2024

  2. [10]

    Computer-supported collabora- tive learning in programming education: A systematic literature review,

    L. Silva, A. J. Mendes, and A. Gomes, “Computer-supported collabora- tive learning in programming education: A systematic literature review,” in2020 IEEE Global Engineering Education Conference (EDUCON), 2020, pp. 1086–1095

    2020

  3. [11]

    Process oriented guided inquiry learning (pogil) for computer science,

    C. Kussmaul, “Process oriented guided inquiry learning (pogil) for computer science,” inProceedings of the 43rd ACM Technical Symposium on Computer Science Education, ser. SIGCSE ’12. New York, NY , USA: Association for Computing Machinery, 2012, p. 373–378. [Online]. Availabl...

    2012

  4. [12]

    Factors influencing the social help-seeking behavior of introductory programming students in a competitive university environment,

    A. K. Cohen, A. Oleson, and A. J. Ko, “Factors influencing the social help-seeking behavior of introductory programming students in a competitive university environment,”ACM Transactions on Computing Education, vol. 24, no. 1, 2024

    2024

  5. [13]

    Unproductive help-seeking in programming: What it is and how to address it,

    S. Marwan, A. Dombe, and T. W. Price, “Unproductive help-seeking in programming: What it is and how to address it,” inProceedings of the 2020 ACM Conference on Innovation and Technology in Computer Science Education, 2020, pp. 54–60

    2020

  6. [14]

    Structuring collaboration in programming through personal-spaces,

    D. Hayatpur, T. Helfenbaum, H. Xia, W. Stuerzlinger, and P. Gries, “Structuring collaboration in programming through personal-spaces,” in Extended Abstracts of the 2023 CHI Conference on Human Factors in Computing Systems, 2023

    2023

  7. [15]

    Improving communication between pair programmers using shared gaze awareness,

    S. D’Angelo and A. Begel, “Improving communication between pair programmers using shared gaze awareness,” inProceedings of the 2017 CHI Conference on Human Factors in Computing Systems, 2017

    2017

  8. [16]

    VizGroup: An AI-assisted event-driven system for collaborative programming learning analytics,

    X. Tang, S. Wong, K. Pu, X. Chen, Y . Yang, and Y . Chen, “VizGroup: An AI-assisted event-driven system for collaborative programming learning analytics,” inProceedings of the 37th Annual ACM Symposium on User Interface Software and Technology, 2024

    2024

  9. [17]

    Dynamite: Real-time debriefing slide authoring through ai-enhanced multimodal interaction,

    P. Keelawat, D. Barron, K. Narasimhan, D. Manesh, X. Tang, X. Chen, S. W. Lee, and Y . Chen, “Dynamite: Real-time debriefing slide authoring through ai-enhanced multimodal interaction,” in2025 IEEE Symposium on Visual Languages and Human-Centric Computing (VL/HCC), 2025, pp. 139–151

    2025

  10. [18]

    Workspace awareness for groupware,

    C. Gutwin and S. Greenberg, “Workspace awareness for groupware,” inConference Companion on Human Factors in Computing Systems, ser. CHI ’96. New York, NY , USA: Association for Computing Machinery, 1996, p. 208–209. [Online]. Available: https://doi.org/10. 1145/257089.257284

    1996

  11. [19]

    FASTDash: A visual dashboard for fostering awareness in software teams,

    J. T. Biehl, M. Czerwinski, G. Smith, and G. G. Robertson, “FASTDash: A visual dashboard for fostering awareness in software teams,” in Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, 2007, pp. 1313–1322

    2007

  12. [20]

    Understanding real-time collaborative programming: A study of Visual Studio Live Share,

    X. Tan, X. Lv, J. Jiang, and L. Zhang, “Understanding real-time collaborative programming: A study of Visual Studio Live Share,”ACM Transactions on Software Engineering and Methodology, vol. 33, no. 4, 2024

    2024

  13. [21]

    Understanding chatbot-mediated task management,

    C. Toxtli, A. Monroy-Hern ´andez, and J. Cranshaw, “Understanding chatbot-mediated task management,” inProceedings of the 2018 CHI Conference on Human Factors in Computing Systems, 2018, pp. 1–6

    2018

  14. [22]

    CHOIR: A chatbot-mediated organizational memory leveraging communication in university research labs,

    S. Lee, A. Abbas, Y . Chen, Y .-H. Kim, and S. W. Lee, “CHOIR: A chatbot-mediated organizational memory leveraging communication in university research labs,” arXiv preprint arXiv:2509.20512, 2025

    2025

  15. [23]

    Orchestration scripts: A system for encoding an organization’s ways of working to support situated work,

    K. Garg, D. Gergle, and H. Zhang, “Orchestration scripts: A system for encoding an organization’s ways of working to support situated work,” inProceedings of the 2023 CHI Conference on Human Factors in Computing Systems, 2023

    2023

  16. [24]

    Exploring real-time col- laborative heart rate displays for cycling partners,

    M. Agharazidermani, L. Lu, and K. E. Boyer, “Exploring real-time col- laborative heart rate displays for cycling partners,” inProceedings of the 25th International Conference on Mobile Human-Computer Interaction Companion, 2023

    2023

  17. [25]

    Sharing facial cues at different target positions to support virtual collaboration,

    T. Teo, A. Jing, C. Zhang, M. Frederick, H. Shin, Y . Lee, Y .-H. Gil, M. Billinghurst, and G. A. Lee, “Sharing facial cues at different target positions to support virtual collaboration,” inProceedings of the Extended Abstracts of the CHI Conference on Human Factors in Comput...

    2025

  18. [26]

    Exploring a real-time feedback display of non-verbal cues in online work meetings to support self-presentation,

    K. Chow, R. Rutishauser, A. N. Meyer, J. McGrenere, and T. Fritz, “Exploring a real-time feedback display of non-verbal cues in online work meetings to support self-presentation,”Proceedings of the ACM on Human-Computer Interaction, vol. 9, no. CSCW071, 2025

    2025

  19. [27]

    LADICA: A large shared display interface for generative AI cognitive assistance in co- located team collaboration,

    Z. Zhang, W. Peng, X. Chen, L. Cao, and T. J.-J. Li, “LADICA: A large shared display interface for generative AI cognitive assistance in co- located team collaboration,” inProceedings of the 2025 CHI Conference on Human Factors in Computing Systems, 2025

    2025

  20. [28]

    It’s Weird That it Knows What I Want: Usability and interactions with Copilot for novice programmers,

    J. Prather, B. N. Reeves, P. Denny, B. A. Becker, J. Leinonen, A. Luxton- Reilly, G. Powell, J. Finnie-Ansley, and E. A. Santos, “It’s Weird That it Knows What I Want: Usability and interactions with Copilot for novice programmers,”ACM Transactions on Computer-Human Interactio...

    2023

  21. [29]

    CodeAid: Evaluating a classroom deployment of an LLM-based programming assistant that balances student and educator needs,

    M. Kazemitabaar, R. Ye, X. Wang, A. Z. Henley, P. Denny, M. Craig, and T. Grossman, “CodeAid: Evaluating a classroom deployment of an LLM-based programming assistant that balances student and educator needs,” inProceedings of the 2024 CHI Conference on Human Factors in Computi...

    2024

  22. [30]

    Ivie: Lightweight anchored explanations of just-generated code,

    L. Yan, A. Hwang, Z. Wu, and A. Head, “Ivie: Lightweight anchored explanations of just-generated code,” inProceedings of the 2024 CHI Conference on Human Factors in Computing Systems, 2024

    2024

  23. [31]

    CoPrompt: Sup- porting prompt sharing and referring in collaborative natural language programming,

    L. Feng, R. Yen, Y . You, M. Fan, J. Zhao, and Z. Lu, “CoPrompt: Sup- porting prompt sharing and referring in collaborative natural language programming,” inProceedings of the 2024 CHI Conference on Human Factors in Computing Systems, 2024

    2024

  24. [32]

    Codeon: On-demand software development assistance,

    Y . Chen, S. W. Lee, Y . Xie, Y . Yang, W. S. Lasecki, and S. Oney, “Codeon: On-demand software development assistance,” inProceedings of the 2017 CHI Conference on Human Factors in Computing Systems, 2017, pp. 6220–6231

    2017

  25. [33]

    Is AI the better programming partner? Human-Human pair programming vs. Human-AI pAIr programming,

    Q. Ma, T. Wu, and K. Koedinger, “Is AI the better programming partner? Human-Human pair programming vs. Human-AI pAIr programming,” arXiv:2306.05153 [cs.HC], 2023

    2023 arXiv

  26. [34]

    Trade-offs for substi- tuting a human with an agent in a pair programming context: The good, the bad, and the ugly,

    S. K. Kuttal, B. Ong, K. Kwasny, and P. Robe, “Trade-offs for substi- tuting a human with an agent in a pair programming context: The good, the bad, and the ugly,” inProceedings of the 2021 CHI Conference on Human Factors in Computing Systems, 2021

    2021

  27. [35]

    Human-human-AI triadic programming: Uncovering the role of AI agent and the value of human partner in collaborative learning,

    T. Daryanto, X. Ding, K. Ping, L. T. Wilhelm, Y . Chen, C. Brown, and E. H. Rho, “Human-human-AI triadic programming: Uncovering the role of AI agent and the value of human partner in collaborative learning,” arXiv:2601.12134 [cs.HC], 2026

    2026

  28. [36]

    Assistance or disruption? Exploring and evaluating the design and trade-offs of proactive AI programming support,

    K. Pu, D. Lazaro, I. Arawjo, H. Xia, Z. Xiao, T. Grossman, and Y . Chen, “Assistance or disruption? Exploring and evaluating the design and trade-offs of proactive AI programming support,” inProceedings of the 2025 CHI Conference on Human Factors in Computing Systems, 2025

    2025

  29. [37]

    The impact of generative AI on critical thinking: Self-reported reductions in cognitive effort and confidence effects from a survey of knowledge workers,

    H.-P. Lee, A. Sarkar, L. Tankelevitch, I. Drosos, S. Rintel, R. Banks, and N. Wilson, “The impact of generative AI on critical thinking: Self-reported reductions in cognitive effort and confidence effects from a survey of knowledge workers,” inProceedings of the 2025 CHI Confe...

    2025

  30. [38]

    The role of tutoring in problem solving,

    D. Wood, J. S. Bruner, and G. Ross, “The role of tutoring in problem solving,”Journal of child psychology and psychiatry, vol. 17, no. 2, pp. 89–100, 1976

    1976

  31. [39]

    Scaffolding in teacher– student interaction: A decade of research,

    J. Van de Pol, M. V olman, and J. Beishuizen, “Scaffolding in teacher– student interaction: A decade of research,”Educational psychology review, vol. 22, no. 3, pp. 271–296, 2010

    2010

  32. [40]

    Compliance without pressure: the foot- in-the-door technique

    J. L. Freedman and S. C. Fraser, “Compliance without pressure: the foot- in-the-door technique.”Journal of personality and social psychology, vol. 4, no. 2, p. 195, 1966

    1966

  33. [41]

    Debugging with an ai tutor: Investigating novice help-seeking behaviors and perceived learning,

    S. Yang, H. Zhao, Y . Xu, K. Brennan, and B. Schneider, “Debugging with an ai tutor: Investigating novice help-seeking behaviors and perceived learning,” inProceedings of the 2024 ACM Conference on International Computing Education Research - Volume 1, ser. ICER ’24. New York,...

    2024

  34. [42]

    Employing subgoals in computer programming education,

    L. E. Margulieux, R. Catrambone, and M. Guzdial, “Employing subgoals in computer programming education,”Computer Science Education, vol. 26, no. 1, pp. 44–67, 2016. [Online]. Available: https://doi.org/10.1080/08993408.2016.1144429

    2016

  35. [43]

    Variolite: Supporting exploratory programming by data scientists,

    M. B. Kery, A. Horvath, and B. Myers, “Variolite: Supporting exploratory programming by data scientists,” inProceedings of the 2017 CHI Conference on Human Factors in Computing Systems, ser. CHI ’17. New York, NY , USA: Association for Computing Machinery, 2017, p. 1265–1276. ...

    2017

  36. [44]

    skwiki: a multimedia sketching system for collaborative creativity,

    Z. Zhao, S. K. Badam, S. Chandrasegaran, D. G. Park, N. L. Elmqvist, L. Kisselburgh, and K. Ramani, “skwiki: a multimedia sketching system for collaborative creativity,” inProceedings of the SIGCHI Conference on Human Factors in Computing Systems, ser. CHI ’14. New York, NY , ...

    2014

  37. [45]

    Studying collaborative interactive machine teaching in image classification,

    B. Mohammadzadeh, J. Franc ¸oise, M. Gouiff `es, and B. Caramiaux, “Studying collaborative interactive machine teaching in image classification,” inProceedings of the 29th International Conference on Intelligent User Interfaces, ser. IUI ’24. New York, NY , USA: Association fo...

    2024

  38. [46]

    Development of the team workload questionnaire (twlq),

    J. Sellers, W. S. Helton, K. N ¨aswall, G. J. Funke, and B. A. Knott, “Development of the team workload questionnaire (twlq),” Proceedings of the Human Factors and Ergonomics Society Annual Meeting, vol. 58, no. 1, pp. 989–993, 2014. [Online]. Available: https://doi.org/10.117...

    2014 doi

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.