Pith. sign in

REVIEW 3 major objections 5 minor 54 references

TaskArtisan: Designing Composable Generative Widgets for LLM-Assisted Analysis

T0 review · 3 major / 5 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 TaskArtisan shows that composing reusable generative widgets shifts LLM-assisted analysis effort from repeated prompting to upfront workflow authoring, yielding faster repeated steps and more completions.

desk verdict Useful systems paper, but the headline setup/reuse trade-off is partly confounded by different backend models across conditions. read the letter →

arxiv 2607.17394 v2 pith:XGXWB4QF submitted 2026-07-19 cs.HC

Meng Chen , Amy Pavel This is my paper · ORCID
classification cs.HC
keywords generativeUILLM-assistedanalysiscomposablewidgetsreusetechnologyprobehuman-AIinteractionworkflowsmodularity
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

This paper argues that LLM-generated analysis tools become more useful when built as composable widgets—separating data, analysis logic, and interface—so users can assemble repeatable workflows. In a 12-participant comparison against a chat-based LLM tool, TaskArtisan users spent significantly more time initially authoring a workflow but significantly less time on repeated analysis steps, completed more tasks, and reported lower frustration. The paper thereby establishes a trade-off: generative UIs shift effort from repeated prompting to upfront authoring. It also proposes a provisional design framework with three axes—malleability, specification, and interoperability—to guide future generative UI design for analysis.

What carries the argument

The carrying mechanism is the 'generative widget': an LLM-generated GUI that couples a GUI component (controls and layout) with an LLM component (analysis logic) through a JSON schema, decoupled from the data it processes. Users author widgets from sketches and prompts, bind data shapes (text, image, CSV) by dragging them onto widgets, and compose widgets into pipelines via two patterns: fan-out (a container applies several widgets to the same data) and sequential (arrows pass one widget's output to another). This decoupling of data, analysis, and display is what makes reuse possible—the same widget can bind to new data without rewriting the analysis prompt.

What would settle it

Run the same comparison with tasks where each subsequent step is a novel, non-repeating analysis rather than an exact repeat; if TaskArtisan no longer shows significantly faster repeat-step times, or total completion time becomes worse, the claim that composable widgets reduce repetitive work fails to generalize.

Watch

Extended reading notes

Core claim

The central discovery is that composing generative widgets changes where effort goes in LLM-assisted analysis. TaskArtisan separates each analysis tool into a data shape, an LLM component, a GUI component, and a connecting schema, so a widget can be reused across datasets and chained into pipelines. In a controlled within-subject study with 12 professionals, setup time for a three-step analysis workflow was significantly higher with TaskArtisan (21.1 vs 14.0 minutes) but repeated steps were significantly faster (4.3 vs 7.8 minutes), with more participants completing all steps (10/12 vs 7/12) and significantly lower frustration. The authors take this as evidence that the upfront authoring cos

Load-bearing premise

The reuse benefit depends on analysis work consisting of stable, repeatable pipelines; if analyses evolve between runs, the upfront authoring cost of composable widgets may not be recovered and chat could come out ahead.

Editorial extensions

If this is right

  • If the upfront authoring cost is paid back by reuse, then generative UI tools should invest in widget libraries, templates, and management tools so users do not rebuild workflows from scratch.
  • The significant reduction in repeat-step time implies that organizations with recurring analysis pipelines—same reports on new data—would see productivity gains from composable widgets.
  • Higher completion rates and lower frustration suggest that composable widgets reduce the context-loss and ambiguous-reference errors that plague long chat conversations.
  • The three-axis framework (malleability, specification, interoperability) gives designers a vocabulary for trade-offs; for instance, typed inputs and outputs could reduce the schema-transformation failures observed in the study.
  • The rigidity and authoring burden reported by non-programmers implies that future systems need adaptive specification mechanisms to lower upfront effort while preserving reuse.

Reading between the lines

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

  • A natural extension beyond the paper: the study's repeated steps were exact repeats of a fixed workflow applied to new personas; if real analyses evolve between runs, the advantage may shrink or invert, so a longitudinal study with open-ended tasks would clarify when composable widgets win.
  • The three axes may not be independent: increasing interoperability by connecting widgets likely raises specification demands and can reduce malleability unless schemas are strongly typed; designers may need to treat the triangle as a constrained design space.
  • A testable design extension would be template widgets with adjustable slots instead of fully custom authoring; this could lower the 'think like a software engineer' barrier for non-programmers while retaining reuse benefits.
  • The widget architecture effectively turns analysis workflows into end-user programs; scaffolding such as example-driven or wizard-based creation could be compared against sketch-and-prompt authoring in a follow-up experiment.
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

3 major / 5 minor

Summary. This paper investigates how generative UIs can support LLM-assisted analysis. The authors report a formative interview study with 6 professionals, an analysis of 100 publicly shared Claude Artifacts in the Data Analysis category, and a within-subject controlled study (N=12) comparing a chat baseline (Claude Artifacts) with TaskArtisan, a technology probe for authoring and assembling generative widgets on an infinite canvas. The central empirical claim is that TaskArtisan shifts effort from repeated prompting to upfront workflow authoring: participants spent significantly more time on initial analysis setup (M=21.1 vs 14.0 min) but significantly less time on repeated steps (M=4.3 vs 7.8 min), with higher completion and lower frustration. The paper synthesizes these findings into a provisional three-axis design framework: malleability, specification, and interoperability.

Significance. The work addresses a timely and important question: whether generative UI can move LLM use from ephemeral chat outputs to reusable, composable analysis tools. The contribution is concrete: a working technology probe, a systematic artifact analysis, and a controlled study that (if the quantitative result is trustworthy) demonstrates a clear trade-off. The provision of a design framework with three axes gives the community a vocabulary for comparing generative UI systems. The paper is honest about the provisional status of the framework, and the qualitative findings (e.g., modularity reducing repetitive work, abstraction burden for non-programmers) are well illustrated with participant quotes. The empirical package is transparent about participant demographics and study procedure, and the use of Wilcoxon tests is appropriate for the small sample.

major comments (3)
  1. [§IV.C, §V.A, §VI.A (Table IV)] The central quantitative claim is confounded by the backend model. Section IV.C states that TaskArtisan's LLM components use OpenAI GPT-4o with web-search and code-execution tools, while Section V.A states that the baseline condition used Claude Artifacts with claude-sonnet-4.0; footnote 7 further reports that one baseline participant (P2) used ChatGPT Data Analyst. The observed differences in setup time (21.1 vs 14.0 min) and repeat time (4.3 vs 7.8 min) could therefore reflect model latency or reliability rather than the composable-widget design. The sentence in §V.A claiming that counterbalancing ensures differences reflect interaction designs 'rather than models' is incorrect: counterbalancing only controls order effects, not model differences. To support the headline trade-off, the evaluation should hold the model fixed across conditions or measure/control model-level factors (e.g.,
  2. [§V.A, Fig. 8, §VI.A] The reuse benefit is measured only under exact repetition. Steps 4–5 repeat the same workflow for Students 2 and 3 after the workflow was established for Student 1. The claim that modularity 'reduces repetitive work' is therefore supported only for stable, pre-specified pipelines. The paper's own framework (§VII) emphasizes malleability for evolving analysis needs, and the participants' reports of refining widgets (e.g., P4's missing 'select all' option) show that workflows are not always static. If real analyses evolve, the upfront authoring cost of TaskArtisan may not be amortized, and the advantage over chat could invert. The authors should explicitly bound the reuse claim to fixed/repeatable workflows or provide evidence about how workflows were modified during the repeated steps.
  3. [§VI.A, Table IV] The completion-rate difference (10/12 vs 7/12) is reported without an inferential test, and baseline failures such as 5/12 failing to obtain correct visualizations at Step 2 are attributed to the chat interface. Given the model confound in the first major comment, these rates confound interface with model capability. The authors should report per-condition error counts by model, and ideally exclude or separately analyze the participant who used a different baseline system (P2), or show that the results are robust to that participant.
minor comments (5)
  1. [§V.A, footnote 7] The use of ChatGPT Data Analyst for P2 breaks the within-subject baseline. Even as a footnote, this should be treated as a protocol deviation; a sensitivity analysis without P2 would reassure readers.
  2. [§VI.A, Fig. 9] The NASA-TLX scale is described as '1 = negative, 5 = positive', but the standard NASA-TLX uses higher scores to indicate higher workload. Please clarify how the six subscales were inverted or recoded, and whether the 'performance' subscale was handled separately.
  3. [Table IV] The header 'Setup ∗∗ Repeat∗∗' is ambiguous. State the p-value directly in the caption instead of relying only on asterisks.
  4. [Throughout] The system name is rendered inconsistently as 'TaskArtisan' and 'TASKARTISAN' (e.g., in the abstract and §VI headings). Please use one form consistently.
  5. [§III.B] The artifact analysis reports two researchers independently open-coding 100 artifacts, but no inter-rater reliability statistic (e.g., Cohen's kappa) is reported. Adding this would strengthen the credibility of the taxonomy.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: empirical study with measured between-condition outcomes; framework is a provisional synthesis, not a derivation.

full rationale

This is an empirical HCI systems paper, not a derivation chain. The central claims—longer setup time (M=21.1 vs 14.0 min, p<0.01), faster repeat time (M=4.3 vs 7.8 min, p<0.01), higher completion rate (10/12 vs 7/12), and lower frustration—come from a within-subject user study with measured times and ratings, not from fitting parameters to the same outcomes. No equation defines the results in terms of an input variable, and no fitted quantity is renamed as a prediction. The three-axis design framework in Section VII is explicitly labeled 'provisional' and is synthesized from observed trade-offs, so the framework is an interpretation of the study rather than a claimed prediction derived from independent axioms; this is a normal design-research loop, not circularity. Self-citations are not load-bearing: the only nearby author-overlapping references ([51], [52]) are used to describe GPT-4o generation of student personas and do not justify the central claim. No uniqueness theorem is imported from prior work by the same authors, and no ansatz is smuggled in via self-citation. The backend-model difference between TaskArtisan (GPT-4o) and the baseline (Claude-Sonnet-4.0) is a potential empirical confound affecting attribution of the trade-off to interface design, but it does not make the measured outcome equivalent to the paper's inputs by construction and is therefore outside the scope of circularity analysis. I find no specific reduction of a claimed result to its own inputs or to a self-citation chain.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

The central claims are empirical; no fitted numerical parameters. The main unstated premises are task representativeness, self-report accuracy, and LLM code-generation reliability. No physical or unexplained entities are introduced; 'generative widget' and 'data shape' are software constructs with implementation details.

assumptions (3)
  • domain assumption All participants' self-reports about their professional work and LLM use are accurate.
    Interviews (§III.A) and post-task interviews (§V.B) are analyzed via affinity diagramming; no independent verification of workplace practices.
  • domain assumption LLM code generation is reliable enough that generated widgets implement the user's intended analysis and UI.
    TaskArtisan's entire interaction model depends on LLM-generated GUI code; the study itself found 2/12 incomplete widgets and repeated failures (P11), so this assumption is only partially met (§IV.C, §VI.B).
  • domain assumption The 5-step repeated-student task is representative of real analysis workflows.
    The measured reuse benefit is created by the exact repetition in Steps 4–5 (§V.A, Fig. 8). If real workflows evolve or do not repeat, the trade-off shifts.

how reviews work

0 comments
Cite this review

Pith. "Pith review of TaskArtisan: Designing Composable Generative Widgets for LLM-Assisted Analysis." pith.science (2026). https://pith.science/paper/XGXWB4QF

@misc{pith2026260717394,
  author       = {Pith},
  title        = {Pith review of: TaskArtisan: Designing Composable Generative Widgets for LLM-Assisted Analysis},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XGXWB4QF}},
  note         = {Machine review of arXiv:2607.17394}
}
read the original abstract

People increasingly use chatbots such as ChatGPT for everyday analysis tasks. While chatbots unify many analysis functions (e.g., scripts, visualizations, summaries), long conversations become hard to navigate, making it difficult to revisit prior steps or reuse successful workflows. LLMs now generate high-fidelity GUI code that enables people to create customized analysis tools beyond text. Yet, what new opportunities generative UIs bring to analysis work remain unclear. We interviewed six professionals about analysis with chatbots, analyzed publicly shared LLM-generated GUI tools, and conducted a comparison study (N=12) between a chatbot and TaskArtisan, a technology probe that enables people to create and assemble generative analysis UI widgets for sequential and fan-out composition. We find that GUI improved clarity and visual presentation but also introduced rigidity and additional prompting challenges. We summarize the trade-offs into a provisional design framework (malleability, specification, interoperability) to inform future generative UI in LLM-assisted analysis workflows.

Figures

Figures reproduced from arXiv: 2607.17394 by the authors.

Figure 1
Figure 1. Three-axis design framework for generative UI tools: [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Screen shots and the names of the Claude artifacts [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Alice, a college counselor, uses TASKARTISAN to support student advising. She sketches and describes to (a1) author a Line Chart widget. She can now (a2) drag college ranking data onto the widget to visualize ranking trajectories. She also previously (b1) created a School Recommendation widget to get personalized college suggestion. As Alice and her student discuss options, she (b2) assembles a Radar Chart, Comparis… view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Widget Architecture. A shared schema coordinates the [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Example of a drag-and-drop interaction. (1) Users [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: Reuse of “Japanese Vocabulary Study Cards & Quiz” [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 8
Figure 8. Figure 8: User Study Design. Participants first complete an analysis task (Steps 1–3), then repeat it (Steps 4–5). A. Study Design We designed two realistic decision-making tasks: a career recommendation task (T1) and a college recommendation task (T2), each requiring participan…
Figure 9
Figure 9. Figure 9: Distribution of the NASA-TLX rating scores for the [PITH_FULL_IMAGE:figures/full_fig_p008_9.png]
Figure 10
Figure 10. Figure 10: Subjective participant ratings of the usefulness of [PITH_FULL_IMAGE:figures/full_fig_p009_10.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

54 extracted references · 1 canonical work pages

  1. [1]

    How Do Analysts Understand and Verify AI-Assisted Data Analyses?

    K. Gu, R. Shang, T. Althoff, C. Wang, and S. M. Drucker, “How Do Analysts Understand and Verify AI-Assisted Data Analyses?” inProceedings of the 2024 CHI Conference on Human Factors in Computing Systems, ser. CHI ’24. New York, NY , USA: Association for Computing Machinery, May 2024, pp. 1–22. [Online]. Available: https://dl.acm.org/doi/10.1145/3613904.3642497

    arXiv 2024

  2. [2]

    How do Data Science Workers Collaborate? Roles, Workflows, and Tools,

    A. X. Zhang, M. Muller, and D. Wang, “How do Data Science Workers Collaborate? Roles, Workflows, and Tools,”Proceedings of the ACM on Human-Computer Interaction, vol. 4, no. CSCW1, pp. 1–23, May 2020, publisher: Association for Computing Machinery (ACM). [Online]. Available: https://dl.acm.org/doi/10.1145/3392826

    work page doi:10.1145/3392826 2020

  3. [3]

    Generative AI in Knowledge Work: Design Implications for Data Navigation and Decision-Making,

    B. Yun, D. Feng, A. S. Chen, A. Nikzad, and N. Salehi, “Generative AI in Knowledge Work: Design Implications for Data Navigation and Decision-Making,” inProceedings of the 2025 CHI Conference on Human Factors in Computing Systems, ser. CHI ’25. New York, NY , USA: Association for Computing Machinery, Apr. 2025, pp. 1–19. [Online]. Available: https://doi.o...

    arXiv 2025

  4. [4]

    Marco: Supporting Business Document Workflows via Collection-Centric Information Foraging with Large Language Models,

    R. Fok, N. Lipka, T. Sun, and A. F. Siu, “Marco: Supporting Business Document Workflows via Collection-Centric Information Foraging with Large Language Models,” inProceedings of the 2024 CHI Conference on Human Factors in Computing Systems, ser. CHI ’24. New York, NY , USA: Association for Computing Machinery, May 2024, pp. 1–20. [Online]. Available: http...

    arXiv 2024

  5. [5]

    Passages: Interacting with Text Across Documents|Proceedings of the 2022 CHI Conference on Human Factors in Computing Systems

    “Passages: Interacting with Text Across Documents|Proceedings of the 2022 CHI Conference on Human Factors in Computing Systems.” [Online]. Available: https://dl.acm.org/doi/abs/10.1145/ 3491102.3502052?utm source=chatgpt.com

    arXiv 2022

  6. [6]

    Generative and Malleable User Interfaces with Generative and Evolving Task-Driven Data Model,

    Y . Cao, P. Jiang, and H. Xia, “Generative and Malleable User Interfaces with Generative and Evolving Task-Driven Data Model,” inProceedings of the 2025 CHI Conference on Human Factors in Computing Systems. Yokohama Japan: ACM, Apr. 2025, pp. 1–20. [Online]. Available: https://dl.acm.org/doi/10.1145/3706598.3713285

    arXiv 2025

  7. [7]

    Data analysis with ChatGPT

    “Data analysis with ChatGPT.” [Online]. Available: https://help.openai. com/en/articles/8437071-data-analysis-with-chatgpt

    arXiv

  8. [8]

    Claude 3.7 Sonnet and Claude Code

    “Claude 3.7 Sonnet and Claude Code.” [Online]. Available: https: //www.anthropic.com/news/claude-3-7-sonnet

Show all 54 references
  1. [9]

    Sensecape: Enabling multilevel exploration and sensemaking with large language models,

    S. Suh, B. Min, S. Palani, and H. Xia, “Sensecape: Enabling multilevel exploration and sensemaking with large language models,” inProceedings of the 36th Annual ACM Symposium on User Interface Software and Technology. ACM, pp. 1–18. [Online]. Available: https://dl.acm.org/doi/...

  2. [12]

    NoteEx: Interactive Visual Context Manipulation for LLM-Assisted Exploratory Data Analysis in Computational Notebooks,

    M. H. Payandeh, L.-P. Yuan, and J. Zhao, “NoteEx: Interactive Visual Context Manipulation for LLM-Assisted Exploratory Data Analysis in Computational Notebooks,” Nov. 2025, arXiv:2511.07223 [cs]. [Online]. Available: http://arxiv.org/abs/2511.07223

    2025

  3. [13]

    Exploration and Explanation in Computational Notebooks,

    A. Rule, A. Tabard, and J. D. Hollan, “Exploration and Explanation in Computational Notebooks,” inProceedings of the 2018 CHI Conference on Human Factors in Computing Systems. Montreal QC Canada: ACM, Apr. 2018, pp. 1–12. [Online]. Available: https://dl.acm.org/doi/10.1145/317...

    2018

  4. [14]

    Managing Messes in Computational Notebooks|Proceedings of the 2019 CHI Conference on Human Factors in Computing Systems

    “Managing Messes in Computational Notebooks|Proceedings of the 2019 CHI Conference on Human Factors in Computing Systems.” [Online]. Available: https://dl.acm.org/doi/10.1145/3290605.3300500? utm source=chatgpt.com

    2019

  5. [16]

    Dynamic Prompt Middleware: Contextual Prompt Refinement Controls for Comprehension Tasks,

    I. Drosos, J. Williams, A. Sarkar, N. Wilson, S. Rintel, and P. Panda, “Dynamic Prompt Middleware: Contextual Prompt Refinement Controls for Comprehension Tasks,” inProceedings of the 4th Annual Symposium on Human-Computer Interaction for Work. Amsterdam Netherlands: ACM, Jun....

    2025

  6. [18]

    Replit – Build apps and sites with AI

    “Replit – Build apps and sites with AI.” [Online]. Available: https://replit.com/

  7. [19]

    [Online]

    Cells, generators, and lenses: Design framework for object-oriented interaction with large language models|proceedings of the 36th annual ACM symposium on user interface software and technology. [Online]. Available: https://dl.acm.org/doi/10.1145/3586183.3606833

  8. [20]

    Ai-instruments: Embodying prompts as instruments to abstract & reflect graphical interface commands as general-purpose tools,

    N. Riche, A. Offenwanger, F. Gmeiner, D. Brown, H. Romat, M. Pahud, N. Marquardt, K. Inkpen, and K. Hinckley, “Ai-instruments: Embodying prompts as instruments to abstract & reflect graphical interface commands as general-purpose tools,” inProceedings of the 2025 CHI Conferenc...

    2025

  9. [21]

    Persist: Persistent and Reusable Interactions in Computational Notebooks,

    K. Gadhave, Z. Cutler, and A. Lex, “Persist: Persistent and Reusable Interactions in Computational Notebooks,” 2024

    2024

  10. [22]

    How data science workers work with data: Discovery, capture, curation, design, creation,

    M. Muller, I. Lange, D. Wang, D. Piorkowski, J. Tsay, Q. V . Liao, C. Dugan, and T. Erickson, “How data science workers work with data: Discovery, capture, curation, design, creation,” inProceedings of the 2019 CHI Conference on Human Factors in Computing Systems, ser. CHI ’19...

    2019

  11. [23]

    Patterns of sharing customizable software,

    W. E. Mackay, “Patterns of sharing customizable software,” in Proceedings of the 1990 ACM conference on Computer-supported cooperative work, ser. CSCW ’90. New York, NY , USA: Association for Computing Machinery, Sep. 1990, pp. 209–221. [Online]. Available: https://dl.acm.org/...

    1990

  12. [24]

    Model-View-Controller (MVC) Architecture

    J. Deacon, “Model-View-Controller (MVC) Architecture.”

  13. [25]

    Technology probes: inspiring design for and with families,

    H. Hutchinson, W. Mackay, B. Westerlund, B. B. Bederson, A. Druin, C. Plaisant, M. Beaudouin-Lafon, S. Conversy, H. Evans, H. Hansen, N. Roussel, and B. Eiderb ¨ack, “Technology probes: inspiring design for and with families,” inProceedings of the SIGCHI Conference on Human Fa...

  14. [26]

    Generative Interfaces for Language Models,

    J. Chen, Y . Zhang, Y . Zhang, Y . Shao, and D. Yang, “Generative Interfaces for Language Models,” Aug. 2025, arXiv:2508.19227 [cs]. [Online]. Available: http://arxiv.org/abs/2508.19227

    2025 arXiv

  15. [27]

    Misty: UI prototyping through interactive conceptual blending,

    Y . Lu, A. Leung, A. Swearngin, J. Nichols, and T. Barik, “Misty: UI prototyping through interactive conceptual blending,” inProceedings of the 2025 CHI Conference on Human Factors in Computing Systems, ser. CHI ’25. Association for Computing Machinery, pp. 1–17. [Online]. Ava...

    2025

  16. [28]

    The GenUI study: Exploring the design of generative UI tools to support UX practitioners and beyond,

    X. A. Chen, T. Knearem, and Y . Li, “The GenUI study: Exploring the design of generative UI tools to support UX practitioners and beyond,” inProceedings of the 2025 ACM Designing Interactive Systems Conference. ACM, pp. 1179–1196. [Online]. Available: https://dl.acm.org/doi/10...

    2025

  17. [29]

    Figma Make: Create with AI-Powered Design Tools

    “Figma Make: Create with AI-Powered Design Tools.” [Online]. Available: https://www.figma.com/make/

  18. [30]

    Cursor - The AI Code Editor

    “Cursor - The AI Code Editor.” [Online]. Available: https://cursor.com/ home

  19. [31]

    v0 by Vercel

    “v0 by Vercel.” [Online]. Available: https://v0.app/

  20. [32]

    Introducing A2UI: An open project for agent-driven interfaces- Google Developers Blog

    “Introducing A2UI: An open project for agent-driven interfaces- Google Developers Blog.” [Online]. Available: https://developers.googleblog. com/introducing-a2ui-an-open-project-for-agent-driven-interfaces/

  21. [33]

    Potluck: Dynamic documents as personal software

    “Potluck: Dynamic documents as personal software.” [Online]. Available: https://www.inkandswitch.com/potluck/

  22. [34]

    DynaVis: Dynamically Synthesized UI Widgets for Visualization Editing,

    P. Vaithilingam, E. L. Glassman, J. P. Inala, and C. Wang, “DynaVis: Dynamically Synthesized UI Widgets for Visualization Editing,” in Proceedings of the CHI Conference on Human Factors in Computing Systems. Honolulu HI USA: ACM, May 2024, pp. 1–17. [Online]. Available: https:...

    2024

  23. [35]

    Malleable Overview-Detail Interfaces,

    B. Min, A. Chen, Y . Cao, and H. Xia, “Malleable Overview-Detail Interfaces,” inProceedings of the 2025 CHI Conference on Human Factors in Computing Systems, ser. CHI ’25. New York, NY , USA: Association for Computing Machinery, Apr. 2025, pp. 1–25. [Online]. Available: https:...

    2025

  24. [37]

    Ideasynth: Iterative research idea development through evolving and composing idea facets with literature-grounded feedback,

    K. Pu, K. J. K. Feng, T. Grossman, T. Hope, B. Dalvi Mishra, M. Latzke, J. Bragg, J. C. Chang, and P. Siangliulue, “Ideasynth: Iterative research idea development through evolving and composing idea facets with literature-grounded feedback,” inProceedings of the 2025 CHI Confe...

    2025

  25. [38]

    InsightLens: Augmenting LLM-Powered Data Analysis With Interactive Insight Management and Navigation,

    L. Weng, X. Wang, J. Lu, Y . Feng, Y . Liu, H. Feng, D. Huang, and W. Chen, “InsightLens: Augmenting LLM-Powered Data Analysis With Interactive Insight Management and Navigation,”IEEE Transactions on Visualization and Computer Graphics, vol. 31, no. 6, pp. 3719–3732, Jun. 2025...

    2025

  26. [39]

    Available: https://doi.org/10.1145/3706598.3714057

    [Online]. Available: https://doi.org/10.1145/3706598.3714057

  27. [40]

    ProactiveAgent: Personalized Context-Aware Reminder System,

    Y . Ma and J. Ren, “ProactiveAgent: Personalized Context-Aware Reminder System,” inAdjunct Proceedings of the 36th Annual ACM Symposium on User Interface Software and Technology, ser. UIST ’23 Adjunct. New York, NY , USA: Association for Computing Machinery, Oct. 2023, pp. 1–3...

    2023

  28. [41]

    Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of AI on Knowledge Worker Productivity and Quality,

    F. Dell’Acqua, E. McFowland III, E. R. Mollick, H. Lifshitz- Assaf, K. Kellogg, S. Rajendran, L. Krayer, F. Candelon, and K. R. Lakhani, “Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of AI on Knowledge Worker Productivity and Quality...

    2023

  29. [43]

    User Experience Design Professionals’ Perceptions of Generative Artificial Intelligence|Proceedings of the 2024 CHI Conference on Human Factors in Computing Systems

    “User Experience Design Professionals’ Perceptions of Generative Artificial Intelligence|Proceedings of the 2024 CHI Conference on Human Factors in Computing Systems.” [Online]. Available: https://dl.acm.org/doi/full/10.1145/3613904.3642114

    2024

  30. [44]

    Dirty Data in the Newsroom: Comparing Data Preparation in Journalism and Data Science,

    S. Kasica, C. Berret, and T. Munzner, “Dirty Data in the Newsroom: Comparing Data Preparation in Journalism and Data Science,” in Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems, ser. CHI ’23. New York, NY , USA: Association for Computing Machinery...

    2023

  31. [45]

    CollabCoder: A Lower-barrier, Rigorous Workflow for Inductive Collaborative Qualitative Analysis with Large Language Models,

    J. Gao, Y . Guo, G. Lim, T. Zhang, Z. Zhang, T. J.-J. Li, and S. T. Perrault, “CollabCoder: A Lower-barrier, Rigorous Workflow for Inductive Collaborative Qualitative Analysis with Large Language Models,” inProceedings of the CHI Conference on Human Factors in Computing System...

    2024

  32. [46]

    Wrangler: interactive visual specification of data transformation scripts,

    S. Kandel, A. Paepcke, J. Hellerstein, and J. Heer, “Wrangler: interactive visual specification of data transformation scripts,” in Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, ser. CHI ’11. New York, NY , USA: Association for Computing Machinery...

    2011

  33. [47]

    Using thematic analysis in psychology,

    V . Braun and V . Clarke, “Using thematic analysis in psychology,”Qualitative Research in Psychology, vol. 3, no. 2, pp. 77–101, Jan. 2006, publisher: Routledge eprint: https://doi.org/10.1191/1478088706qp063oa. [Online]. Available: https: //doi.org/10.1191/1478088706qp063oa

    2006 doi

  34. [48]

    Reification, polymorphism and reuse: three principles for designing visual interfaces,

    M. Beaudouin-Lafon and W. E. Mackay, “Reification, polymorphism and reuse: three principles for designing visual interfaces,” inProceedings of the working conference on Advanced visual interfaces. ACM, pp. 102–

  35. [49]

    Instrumental interaction: an interaction model for designing post-WIMP user interfaces,

    M. Beaudouin-Lafon, “Instrumental interaction: an interaction model for designing post-WIMP user interfaces,” inProceedings of the SIGCHI conference on Human Factors in Computing Systems. The Hague The Netherlands: ACM, Apr. 2000, pp. 446–453. [Online]. Available: https://dl.a...

  36. [50]

    Think-Aloud Computing: Supporting Rich and Low-Effort Knowledge Capture,

    R. Krosnick, F. Anderson, J. Matejka, S. Oney, W. S. Lasecki, T. Grossman, and G. Fitzmaurice, “Think-Aloud Computing: Supporting Rich and Low-Effort Knowledge Capture,” inProceedings of the 2021 CHI Conference on Human Factors in Computing Systems, ser. CHI ’21. New York, NY ...

    2021

  37. [51]

    Why am I seeing this: Democratizing End User Auditing for Online Content Recommendations,

    C. Chen, L. Li, L. Cao, Y . Ye, T. Li, Y . Yao, and T. J.-j. Li, “Why am I seeing this: Democratizing End User Auditing for Online Content Recommendations,” Apr. 2025, arXiv:2410.04917 [cs]. [Online]. Available: http://arxiv.org/abs/2410.04917

    2025 arXiv

  38. [52]

    DrawTalking: Building interactive worlds by sketching and speaking,

    K. T. Rosenberg, R. H. Kazi, L.-Y . Wei, H. Xia, and K. Perlin, “DrawTalking: Building interactive worlds by sketching and speaking,” inProceedings of the 37th Annual ACM Symposium on User Interface Software and Technology, ser. UIST ’24. Association for Computing Machinery, p...

  39. [53]

    Nasa-task load index (nasa-tlx); 20 years later,

    S. G. Hart, “Nasa-task load index (nasa-tlx); 20 years later,” inPro- ceedings of the human factors and ergonomics society annual meeting, vol. 50, no. 9. Sage publications Sage CA: Los Angeles, CA, 2006, pp. 904–908

    2006

  40. [54]

    NaturalEdit: Code Modification through Direct Interaction with Adaptive Natural Language Representation,

    N. Tang, D. Meininger, G. Xu, Y . Shi, Y . Huang, C. McMillan, and T. J.-J. Li, “NaturalEdit: Code Modification through Direct Interaction with Adaptive Natural Language Representation,” Apr. 2026, arXiv:2510.04494 [cs.HC]. [Online]. Available: http://arxiv.org/ abs/2510.04494

    2026

  41. [55]

    An Empathy-Based Sandbox Approach to Bridge the Privacy Gap among Attitudes, Goals, Knowledge, and Behaviors,

    C. Chen, W. Li, W. Song, Y . Ye, Y . Yao, and T. J.-J. Li, “An Empathy-Based Sandbox Approach to Bridge the Privacy Gap among Attitudes, Goals, Knowledge, and Behaviors,” inProceedings of the CHI Conference on Human Factors in Computing Systems. Honolulu HI USA: ACM, May 2024,...

    2024

  42. [56]

    SQUIRE: Interactive UI Authoring via Slot QUery Intermediate REpresentations,

    A. Leung, R. Cheng, J. Wu, J. Nichols, and T. Barik, “SQUIRE: Interactive UI Authoring via Slot QUery Intermediate REpresentations,” inProceedings of the 38th Annual ACM Symposium on User Interface Software and Technology. Busan Republic of Korea: ACM, Sep. 2025, pp. 1–17. [On...

    2025 doi

  43. [57]

    Graffiti: Enabling an Ecosystem of Personalized and Interoperable Social Applications|Proceedings of the 38th Annual ACM Symposium on User Interface Software and Technology

    “Graffiti: Enabling an Ecosystem of Personalized and Interoperable Social Applications|Proceedings of the 38th Annual ACM Symposium on User Interface Software and Technology.” [Online]. Available: https://dl.acm.org/doi/10.1145/3746059.3747627

  44. [61]

    StoryEnsemble: Enabling Dynamic Exploration & Iteration in the Design Process with AI and Forward-Backward Propagation,

    S. Suh, M. Lai, K. Pu, S. P. Dow, and T. Grossman, “StoryEnsemble: Enabling Dynamic Exploration & Iteration in the Design Process with AI and Forward-Backward Propagation,” inProceedings of the 38th Annual ACM Symposium on User Interface Software and Technology. Busan Republic...

    2025

  45. [109]

    Available: https://dl.acm.org/doi/10.1145/345513.345267

    [Online]. Available: https://dl.acm.org/doi/10.1145/345513.345267

  46. [2025]

    Available: https://doi.org/10.1145/3706598.3714259

    [Online]. Available: https://doi.org/10.1145/3706598.3714259

Pith tools

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