Pith. sign in

REVIEW 3 major objections 3 minor 3 cited by

Frontend Diffusion: Empowering Self-Representation of Junior Researchers and Designers Through Multi-agent System

T0 review · 3 major / 3 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read Frontend Diffusion claims that a sketch-to-website multi-agent pipeline lets junior researchers and designers present themselves online, with users experiencing the AI as an enhancer rather than a replacement.

desk verdict A useful small HCI study with an overclaimed headline: the capability-enhancer theme is elicited by a leading question, so treat it as perceived benefit; the multi-agent system and alignment theme are the real contribution. read the letter →

arxiv 2502.03788 v2 pith:73HWTCRB submitted 2025-02-06 cs.HC

Zijian Ding , Qinshi Zhang , Mohan Chi , Ziyi Wang This is my paper · ORCID
classification cs.HC
keywords codegenerationmulti-agentsystemshuman-AIcollaborationself-representationfrontenddevelopmentuserstudythematicanalysisgenerativeAI
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

Frontend Diffusion is a multi-agent system that turns a hand-drawn layout sketch and a short text prompt into a working personal website, and the paper claims this can help junior researchers and designers represent themselves online without first becoming frontend developers. In a qualitative study with 13 participants from research and design backgrounds, users described the AI as a capability enhancer rather than a replacement, saying it lifted the technical burden of repetitive coding and freed them for content, ideation, and self-reflection. The study also found that effective co-creation requires alignment in both directions: the system must guide newcomers and refine prompts, and users need fine-grained control over individual parts of the generated page. If these findings hold, agentic design tools offer a low-barrier route to professional self-presentation for early-career scholars and creative practitioners, and multi-agent systems should be designed with visible, addressable roles rather than a single black-box generation step.

What carries the argument

The load-bearing mechanism is the three-agent refinement loop with a shared-memory state. The Design Agent first turns the user's SVG sketch into a structured Product Requirements Document (PRD), injecting image-search keywords and retrieved image URLs; the Code Agent then renders the PRD and the user's prompt into an initial website; the Critic Agent reviews each version, proposes improvements in layout, accessibility, and performance, and triggers another generation round, defaulting to four iterations with user-visible version branching. The PRD is the intermediate artifact that carries the user's intent from sketch to code, and the critic loop is what moves the page from first draft to refined site.

What would settle it

Have a new group of participants use Frontend Diffusion for a session and then complete a short AI-free task, such as changing the color scheme and adding a section directly in the generated HTML/CSS; compare their speed and success rate with a control group that only watched a template tutorial. If the reported sense of enhancement does not translate to measurable performance gains, the paper's headline theme is not supported as a claim about capability.

Watch

Extended reading notes

Core claim

On its own terms, the paper establishes an end-to-end collaboration prototype and a qualitative account of how users experience it. Three cooperating agents—a Design Agent that converts the sketch into a structured Product Requirements Document with image choices, a Code Agent that renders that document into runnable HTML/CSS/JavaScript, and a Critic Agent that reviews the code and drives up to four refinement iterations—turn a rough drawing into a near-final personal website. The user interviews yielded two themes: the tool is experienced as a human-capability enhancer, because it removes pain points and opens room for reflection and career exploration, and human-AI alignment must be bidirectional, because novices need onboarding and prompt guidance while users want fine-grained section-level control and the ability to catch and reuse the AI's unexpected creative output.

Load-bearing premise

The study assumes that what 13 participants said in a 45-minute session about enhancement versus replacement reflects an actual increase in their abilities, rather than their reaction to the question's wording.

Editorial extensions

If this is right

  • Junior researchers and designers with no web-development background can produce a personal website from a rough drawing and a one-line prompt, because the critic loop carries out the finishing work automatically.
  • Users in the study reported feeling more capable, not displaced: the tool took over repetitive implementation, leaving them to concentrate on content, narrative, and self-reflection.
  • Effective human-AI co-creation requires alignment in both directions: onboarding support and prompt refinement from the AI, plus fine-grained control, version remixing, and a way to express dynamic behavior from the user.
  • The same three-agent pattern can be pointed at career development, for example an AI that role-plays a recruiter or consultant and advises on the look, feel, and content of a portfolio.
  • For multi-agent design systems to support this alignment, each agent's role should be visible and separately addressable, so a user can ask the Design Agent for variations or the Critic Agent for another review round.

Reading between the lines

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

  • Beyond the study, the enhancer-not-replacement finding rests on what 13 users said immediately after a single session; a testable extension is to measure whether repeated use actually transfers frontend skill or self-efficacy, for example with an AI-free editing task before and after use.
  • The sketch-to-PRD-to-code pipeline is not limited to websites: the same intent-carrying intermediate format could generate CVs, slide decks, or portfolios, and the critic loop could be evaluated against single-shot generation on accessibility or code-quality metrics.
  • Bidirectional alignment, taken seriously, implies per-agent controls in the interface, letting users intervene at design, implementation, or critique stages rather than at the whole pipeline only; the paper sketches this direction but does not implement or test it.
  • Because the interview question directly asked whether the tool 'enhances or replaces,' a framing-controlled replication would show whether the headline theme survives less leading wording.
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 / 3 minor

Summary. The paper introduces Frontend Diffusion, an open-source multi-agent system with Design, Code, and Critic agents that converts a user-drawn layout sketch and a textual prompt into iteratively refined HTML/CSS/JavaScript code. The authors report a qualitative user study with 13 junior researchers and designers in which participants created personal websites and then took part in semi-structured interviews. From thematic analysis of the interviews, the paper identifies two themes: AI as a human capability enhancer rather than a replacement, and bidirectional human-AI alignment. The paper argues that such a system can support self-representation by reducing technical burden and freeing users for higher-level content and career-reflection tasks, and it discusses future directions including AI career advising, hierarchical prompting, and local-model solutions for privacy.

Significance. If the claims are supported, the work would provide a useful design exploration of multi-agent code generation for non-expert users and a concrete open-source artifact (the GitHub repository) plus a named qualitative analysis method. The paper's main value at present is as a design study of perceived benefits and as a source of design implications, especially the bidirectional-alignment subthemes such as onboarding support, prompt guidance, fine-grained control, and version remixing. However, the central empirical claim—that the user study 'shows AI as a human capability enhancer rather than a replacement'—is not established by the reported data, since the evidence is self-reported perception elicited by a leading interview question, with no baseline, no pre/post capability measure, and no objective evaluation of the generated websites or code. The reported themes are plausible and internally coherent, but the abstract and results sections overstate their status as findings about actual capability enhancement.

major comments (3)
  1. [Abstract; Section IV-B; Section V (Theme 1)] The central claim that the study 'shows AI as a human capability enhancer rather than a replacement' is not supported by the evidence as presented. The main theme was elicited by the direct, binary interview question in Section IV-B ('Do you think this tool enhances human abilities or replaces them?'), which presupposes the dichotomy and invites a pro-tool response in a researcher-moderated session. The supporting quotes in Table II and Section V are largely hypothetical or future-oriented: P2 says 'if it can generate HTML... it would enhance my productivity,' P3 speculates that 'everyone might want one,' and P13 describes frontend development as generally frustrating rather than reporting a measured change during the session. No pre/post measure of capability, no baseline condition, and no objective assessment of skill or task performance is reported. The Limitations section in Section VI-C acknowledges the narrow sample and image mismatch but does not acknowledge this self-report/behavior gap. I recommend replacing 'shows' with language about perceived benefits and explicitly reframing Theme 1 as 'participants perceived AI as a capability enhancer,' unless behavioral evidence is added.
  2. [Section III; Section V; Table II] The paper describes the system as producing 'refined website code' and the Critic Agent as improving layout, accessibility, and performance, but no objective evaluation of the generated websites is provided. The Results contain no metrics for code validity, visual fidelity to the user's sketch, accessibility conformance, page render success, or the number of iterations needed to reach an acceptable version. The only evidence for quality is participants' self-report, and some of the quotes in Table II refer to imagined or future use cases rather than the actual outputs produced in the study. Adding even minimal objective checks—such as HTML validity, successful rendering of all generated pages, a comparison of the final output to the sketch, or expert ratings of a sample of outputs—would materially strengthen the capability-enhancer claim; without such checks, the conclusions should be limited to user experience and perceived benefit.
  3. [Section IV; Section VI-C] The study has no comparison condition, so it cannot attribute the reported benefits to the specific multi-agent architecture. Participants' comments (e.g., P7's comparison with Wix and P13's general difficulty with frontend development) could apply to any sketch- or prompt-based website generator. Since the paper's contribution is the multi-agent system and its 'bidirectional human-AI alignment' findings, a comparison against a single-agent baseline, a template-based tool, or at least a within-subjects manipulation of the Critic Agent's iterations would be needed to support the design-implication claims as evidence rather than as suggestive feedback. This gap should be acknowledged in Section VI-C and the claims adjusted accordingly.
minor comments (3)
  1. [Section V] The sentence introducing the two themes reads 'Theme 1 - AI as A Human Capability Enhancer and Bidirectional Human–AI Alignment,' which makes it sound as though there is only one theme; the paper should label the second theme explicitly (e.g., 'Theme 2: Bidirectional Human-AI Alignment') in the same sentence.
  2. [Section VI-A] The Discussion uses strong language such as 'The results demonstrate' and 'AI reduces both cognitive load and technical barriers'; given the qualitative, perception-based evidence, this should be tempered to 'suggest' or 'participants reported.'
  3. [Section VI-C] The Limitations section should also mention the leading nature of the interview question about enhancement versus replacement and the absence of behavioral outcome measures, since these directly qualify the headline finding.

Circularity Check

1 steps flagged · score 4.0 of 10

The 'AI as a capability enhancer' theme is partly a restatement of the interview question that directly asks participants to choose between enhancement and replacement; the rest of the paper is self-contained.

  1. self definitional [Abstract; Section IV-B 'Post-Study Interview'; Section V 'Results' (Theme 1)]
    "Abstract: 'A user study with 13 junior researchers and designers shows AI as a human capability enhancer rather than a replacement.' Section IV-B interview outline: 'Do you think this tool enhances human abilities or replaces them? Why?'"

    The headline finding is the affirmative answer to the exact question posed in the interview script. Participants were asked to choose between 'enhances human abilities' and 'replaces them,' and the abstract then reports the study 'shows AI as a human capability enhancer rather than a replacement.' The theme label is thus built into the data-collection instrument: the category being 'discovered' was explicitly offered to participants in a researcher-moderated session, with no baseline, pre/post measure, or behavioral test of actual capability enhancement.

full rationale

The paper's central contribution is a multi-agent system plus a qualitative user study. The system design itself (Design Agent, Code Agent, Critic Agent) is described concretely and is not circular. The main circularity concern is limited to Theme 1: the abstract's claim that the study 'shows AI as a human capability enhancer rather than a replacement' closely mirrors the interview question 'Do you think this tool enhances human abilities or replaces them?' This is a partial self-definitional issue because the result category was explicitly named in the prompt, and the study lacks any behavioral evidence of capability change; the Limitations section acknowledges sample narrowness and image mismatch but does not acknowledge this self-report/behavior gap. Theme 2 (Bidirectional Human-AI Alignment) is supported by more open-ended quotes and specific design suggestions, so it has independent content. The only self-citation in the paper is reference [27], used for a future-work suggestion about coordination among agents; it is not load-bearing for any central claim. No uniqueness-theorem importation, ansatz smuggling, or renaming of known results is present. The score reflects one partially construction-built finding, while acknowledging that the system and most qualitative themes retain independent value.

Assumptions & free parameters 1 free parameters · 4 assumptions · 0 invented entities

The paper introduces no fitted parameters and no invented physical entities. The central claims rest on domain assumptions about LLM competence, sample representativeness, and the validity of self-reported interview data. These assumptions are acknowledged in the paper's limitations but are not independently verified.

free parameters (1)
  • max_critic_iterations = 4
    Default number of critic-agent refinement loops (Section III). Chosen by hand as a system setting; not fitted to data and not central to the qualitative themes.
assumptions (4)
  • domain assumption The selected LLM (Claude-3.5-Sonnet) reliably converts sketches and prompts into runnable, refined website code through the three-agent loop.
    Section III describes the system workflow but reports no objective code-quality or visual-fidelity evaluation.
  • domain assumption A 13-participant convenience sample of junior researchers and designers is sufficient to reveal general themes about AI-supported self-representation.
    Section IV-A and Section VI-C acknowledge the narrow sample; no non-academic professionals were included.
  • domain assumption Interview self-reports accurately capture whether the tool enhances rather than replaces human capability.
    Section V Theme 1 is built from quotes in response to a direct interview question in Section IV-B; no behavioral pre/post measures are reported.
  • domain assumption Thematic analysis by the authors is a valid and reliable interpretation of the transcripts.
    Section IV-C follows Braun and Clarke but shows no inter-rater reliability or audit trail beyond analytic memos.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Frontend Diffusion: Empowering Self-Representation of Junior Researchers and Designers Through Multi-agent System." pith.science (2026). https://pith.science/paper/73HWTCRB

@misc{pith2026250203788,
  author       = {Pith},
  title        = {Pith review of: Frontend Diffusion: Empowering Self-Representation of Junior Researchers and Designers Through Multi-agent System},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/73HWTCRB}},
  note         = {Machine review of arXiv:2502.03788}
}
read the original abstract

With the continuous development of generative AI's logical reasoning abilities, AI's growing code-generation potential poses challenges for both technical and creative professionals. But how can these advances be directed toward empowering junior researchers and designers who often require additional help to build and express their professional and personal identities? We introduce Frontend Diffusion, a multi-agent coding system transforming user-drawn layouts and textual prompts into refined website code, thereby supporting self-representation goals. A user study with 13 junior researchers and designers shows AI as a human capability enhancer rather than a replacement, and highlights the importance of bidirectional human-AI alignment. We then discuss future work such as leveraging AI for career development and fostering bidirectional human-AI alignment of multi-agent systems.

Figures

Figures reproduced from arXiv: 2502.03788 by the authors.

Figure 1
Figure 1. Left: Website generation workflow: (a) user inputs prompt; (b) user draws layout of the website in sketch; (c) the system generates the first website; [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 3 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. P2P: Automated Paper-to-Poster Generation and Fine-Grained Benchmark

    cs.CL 2025-05 conditional novelty 6.0 of 10

    P2P is a multi-agent framework that automatically generates HTML-rendered academic posters from papers, backed by a 30k instruction dataset and a 121-pair evaluation benchmark.

  2. Reverse Browser: Vector-Image-to-Code Generator

    cs.SE 2025-09 conditional novelty 5.0 of 10

    An open-weights system that turns vector images of web designs into HTML/CSS, with new datasets and a multi-scale pixel metric, though accuracy remains below production quality.

  3. Vibe Coding vs. Agentic Coding: Fundamentals and Practical Implications of Agentic AI

    cs.SE 2025-05 conditional novelty 3.0 of 10

    A qualitative taxonomy positions vibe coding and agentic coding as complementary paradigms rather than rivals in AI-assisted software development.

Reference graph

Works this paper leans on

27 extracted references · 15 canonical work pages · cited by 3 Pith papers

  1. [1]

    Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning,

    D. Guo, D. Yang, H. Zhang, J. Song, R. Zhang, R. Xu, Q. Zhu, S. Ma, P. Wang, X. Bi et al. , “Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning,” arXiv preprint arXiv:2501.12948 , 2025

    arXiv 2025

  2. [2]

    OpenAI o3-mini System Card,

    OpenAI, “OpenAI o3-mini System Card,” Dec. 2025. [Online]. Available: https://cdn.openai.com/o3-mini-system-card.pdf

    work page 2025

  3. [3]

    Co-Writing Screenplays and Theatre Scripts with Language Models: Evaluation by Industry Professionals,

    P. Mirowski, K. W. Mathewson, J. Pittman, and R. Evans, “Co-Writing Screenplays and Theatre Scripts with Language Models: Evaluation by Industry Professionals,” in Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems , ser. CHI ’23. New York, NY , USA: Association for Computing Machinery, Apr. 2023, pp. 1–34. [Online]. Available: h...

    arXiv 2023

  4. [4]

    Large-scale Text-to-Image Generation Models for Visual Artists' Creative Works

    H.-K. Ko, G. Park, H. Jeon, J. Jo, J. Kim, and J. Seo, “Large-scale Text-to-Image Generation Models for Visual Artists’ Creative Works,” in Proceedings of the 28th International Conference on Intelligent User Interfaces, Mar. 2023, pp. 919–933, arXiv:2210.08477 [cs]. [Online]. Available: http://arxiv.org/abs/2210.08477

    work page Pith review arXiv 2023

  5. [5]

    AI Art and its Impact on Artists,

    H. H. Jiang, L. Brown, J. Cheng, M. Khan, A. Gupta, D. Workman, A. Hanna, J. Flowers, and T. Gebru, “AI Art and its Impact on Artists,” in Proceedings of the 2023 AAAI/ACM Conference on AI, Ethics, and Society. Montr \’{e}al QC Canada: ACM, Aug. 2023, pp. 363–374. [Online]. Available: https://dl.acm.org/doi/10.1145/3600211.3604681

    work page arXiv 2023

  6. [6]

    The Impact of Generative AI on Artists,

    R. Kawakami and S. Venkatagiri, “The Impact of Generative AI on Artists,” in Creativity and Cognition . Chicago IL USA: ACM, Jun. 2024, pp. 79–82. [Online]. Available: https://dl.acm.org/doi/10.1145/3635636.3664263

    work page arXiv 2024

  7. [7]

    Evaluating Large Language Models Trained on Code,

    M. Chen, J. Tworek, H. Jun, Q. Yuan, H. P. d. O. Pinto, J. Kaplan, H. Edwards, Y . Burda, N. Joseph, G. Brockman, A. Ray, R. Puri, G. Krueger, M. Petrov, H. Khlaaf, G. Sastry, P. Mishkin, B. Chan, S. Gray, N. Ryder, M. Pavlov, A. Power, L. Kaiser, M. Bavarian, C. Winter, P. Tillet, F. P. Such, D. Cummings, M. Plappert, F. Chantzis, E. Barnes, A. Herbert-V...

    arXiv 2021

  8. [8]

    SWE-bench: Can language models resolve real-world github issues?

    C. E. Jimenez, J. Yang, A. Wettig, S. Yao, K. Pei, O. Press, and K. R. Narasimhan, “SWE-bench: Can language models resolve real-world github issues?” in The Twelfth International Conference on Learning Representations , 2024. [Online]. Available: https://openreview.net/forum?id=VTF8yNQM66

    work page 2024

Show all 27 references
  1. [9]

    Design2Code: How Far Are We From Automating Front-End Engineering?

    C. Si, Y . Zhang, Z. Yang, R. Liu, and D. Yang, “Design2Code: How Far Are We From Automating Front-End Engineering?” Mar. 2024, arXiv:2403.03163 [cs]. [Online]. Available: http://arxiv.org/abs/2403.03163

    2024 arXiv

  2. [10]

    OpenHands: An Open Platform for AI Software Developers as Generalist Agents,

    X. Wang, B. Li, Y . Song, F. F. Xu, X. Tang, M. Zhuge, J. Pan, Y . Song, B. Li, J. Singh, H. H. Tran, F. Li, R. Ma, M. Zheng, B. Qian, Y . Shao, N. Muennighoff, Y . Zhang, B. Hui, J. Lin, R. Brennan, H. Peng, H. Ji, and G. Neubig, “OpenHands: An Open Platform for AI Software D...

    2024 arXiv

  3. [11]

    Automatically Generating UI Code from Screenshot: A Divide-and- Conquer-Based Approach,

    Y . Wan, C. Wang, Y . Dong, W. Wang, S. Li, Y . Huo, and M. R. Lyu, “Automatically Generating UI Code from Screenshot: A Divide-and- Conquer-Based Approach,” Oct. 2024, arXiv:2406.16386 [cs]. [Online]. Available: http://arxiv.org/abs/2406.16386

    2024 arXiv

  4. [12]

    UICoder: Finetuning Large Language Models to Generate User Interface Code through Automated Feedback,

    J. Wu, E. Schoop, A. Leung, T. Barik, J. P. Bigham, and J. Nichols, “UICoder: Finetuning Large Language Models to Generate User Interface Code through Automated Feedback,” Jun. 2024, arXiv:2406.07739 [cs]. [Online]. Available: http://arxiv.org/abs/2406.07739

    2024 arXiv

  5. [13]

    FrameKit: A Tool for Authoring Adaptive UIs Using Keyframes,

    J. Wu, K. Todi, J. Chan, B. A. Myers, and B. Lafreniere, “FrameKit: A Tool for Authoring Adaptive UIs Using Keyframes,” in Proceedings of the 29th International Conference on Intelligent User Interfaces . Greenville SC USA: ACM, Mar. 2024, pp. 660–674

    2024

  6. [14]

    PromptInfuser: How Tightly Coupling AI and UI Design Impacts Designers’ Workflows,

    S. Petridis, M. Terry, and C. J. Cai, “PromptInfuser: How Tightly Coupling AI and UI Design Impacts Designers’ Workflows,” in De- signing Interactive Systems Conference . IT University of Copenhagen Denmark: ACM, Jul. 2024, pp. 743–756

    2024

  7. [15]

    DesignAID: Using Generative AI and Semantic Diversity for Design Inspiration,

    A. Cai, S. R. Rick, J. L. Heyman, Y . Zhang, A. Filipowicz, M. Hong, M. Klenk, and T. Malone, “DesignAID: Using Generative AI and Semantic Diversity for Design Inspiration,” in Proceedings of The ACM Collective Intelligence Conference . Delft Netherlands: ACM, Nov. 2023, pp. 1...

    2023

  8. [16]

    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,” Sep. 2024

    2024

  9. [17]

    Towards Generating UI Design Feedback with LLMs,

    P. Duan, J. Warner, and B. Hartmann, “Towards Generating UI Design Feedback with LLMs,” in Adjunct Proceedings of the 36th Annual ACM Symposium on User Interface Software and Technology . San Francisco CA USA: ACM, Oct. 2023, pp. 1–3. [Online]. Available: https://dl.acm.org/do...

    2023

  10. [18]

    Generating Automatic Feedback on UI Mockups with Large Language Models,

    P. Duan, J. Warner, Y . Li, and B. Hartmann, “Generating Automatic Feedback on UI Mockups with Large Language Models,” in Proceedings of the CHI Conference on Human Factors in Computing Systems . Honolulu HI USA: ACM, May 2024, pp. 1–20

    2024

  11. [19]

    UIClip: A Data-driven Model for Assessing User Interface Design,

    J. Wu, Y .-H. Peng, X. Y . A. Li, A. Swearngin, J. P. Bigham, and J. Nichols, “UIClip: A Data-driven Model for Assessing User Interface Design,” in Proceedings of the 37th Annual ACM Symposium on User Interface Software and Technology . Pittsburgh PA USA: ACM, Oct. 2024, pp. 1–16

    2024

  12. [20]

    AI Is Not Enough: A Hybrid Technical Approach to AI Adoption in UI Linting With Heuristics,

    Y . Lu, T. Knearem, S. Dutta, J. Blass, C. Kliman-Silver, and F. Bentley, “AI Is Not Enough: A Hybrid Technical Approach to AI Adoption in UI Linting With Heuristics,” in Extended Abstracts of the CHI Conference on Human Factors in Computing Systems . Honolulu HI USA: ACM, May...

    2024

  13. [21]

    Sketch2Prototype: rapid conceptual design exploration and prototyping with generative AI,

    K. M. Edwards, B. Man, and F. Ahmed, “Sketch2Prototype: rapid conceptual design exploration and prototyping with generative AI,” Proceedings of the Design Society , vol. 4, pp. 1989–1998, May 2024

    1989

  14. [22]

    Evolving Roles and Workflows of Creative Practitioners in the Age of Generative AI,

    S. Palani and G. Ramos, “Evolving Roles and Workflows of Creative Practitioners in the Age of Generative AI,” in Creativity and Cognition . Chicago IL USA: ACM, Jun. 2024, pp. 170–184. [Online]. Available: https://dl.acm.org/doi/10.1145/3635636.3656190

    2024

  15. [23]

    Supermind Ideator: Exploring generative AI to support creative problem-solving,

    S. R. Rick, G. Giacomelli, H. Wen, R. J. Laubacher, N. Taubenslag, J. L. Heyman, M. S. Knicker, Y . Jeddi, H. Maier, S. Dwyer, P. Ragupathy, and T. W. Malone, “Supermind Ideator: Exploring generative AI to support creative problem-solving,” Nov. 2023

    2023

  16. [24]

    Parallel prototyping leads to better design results, more divergence, and increased self-efficacy,

    S. P. Dow, A. Glassco, J. Kass, M. Schwarz, D. L. Schwartz, and S. R. Klemmer, “Parallel prototyping leads to better design results, more divergence, and increased self-efficacy,” ACM Transactions on Computer-Human Interaction, vol. 17, no. 4, pp. 1–24, Dec. 2010

    2010

  17. [25]

    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

    2006

  18. [26]

    Intent Tagging: Exploring Micro-Prompting Interactions for Supporting Granular Human-GenAI Co-Creation Workflows,

    F. Gmeiner, N. Marquardt, M. Bentley, H. Romat, M. Pahud, D. Brown, A. Roseway, N. Martelaro, K. Holstein, K. Hinckley, and N. Riche, “Intent Tagging: Exploring Micro-Prompting Interactions for Supporting Granular Human-GenAI Co-Creation Workflows,” Feb. 2025, arXiv:2502.18737...

    2025 arXiv

  19. [27]

    From MAS to MARS: Coordination Failures and Reasoning Trade-offs in Hierarchical Multi-Agent Robotic Systems within a Healthcare Scenario,

    Y . Bai, Z. Ding, S. Wen, X. Chang, and A. Taylor, “From MAS to MARS: Coordination Failures and Reasoning Trade-offs in Hierarchical Multi-Agent Robotic Systems within a Healthcare Scenario,” Aug. 2025, arXiv:2508.04691 [cs]. [Online]. Available: http://arxiv.org/abs/2508.04691

    2025 arXiv

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.