What Has Generative AI Changed About Making for Oneself? A Map of 59 Sources

An integrated summary that maps, through a lightweight scoping of 59 sources, how generative AI (2023–2026) has changed making that starts from one’s own need and serves oneself. The sister note end-user-development-literature maps EUC/EUD through 152 works along the axes of history, social demands, and limits; end-user-development-novice-learning-literature maps 185 works along learning outcomes and motive. The latter’s third collection gathered the lineage of the make-for-your-own-need motive (von Hippel, OSS motivation, personal informatics, makers), yet none of the combined 337 works studied the making-for-oneself quadrant × generative AI. This note fills only that intersection. Below, works from the existing corpora are cited as E and N numbers; works from this corpus as G numbers. Adjacent notes: empirical evidence on learning harms of generative AI is in ai-cognitive-offloading-learning, the industry discourse on vibe coding in vibe-coding-design-production, the maker-to-editor turn in maker-to-editor-paradigm, and agentic experience design in agentic-experience-literature. Collection lineage: source-researcher (profile: scholarly), four parallel tracks → paper-screening-agent. Protocol: .claude/rules/collection-protocol.md (zero fabrication, provenance tracking). The internal ledger with provenance is source/review/genai-personal-software/papers.md.

Survey Metadata

Collection date: 2026-08-14 / Count: 59 (67 candidates in the exploration stage, 4 excluded, 4 within-candidate duplicate merges)
Period: 2023–2026 (late 2022 was allowed where contextually necessary, but all adopted works are from 2023 onward)
Collection design: Four parallel tracks: making personal software with LLMs (16), self-data and generative AI (20), self-made assistive technology and hobbyist creation (14), and limits specific to this quadrant (17). Every track was instructed to include critical and negative evidence, and the fourth track was instructed to check how the four gaps identified by the prior reviews (no delayed tests, unmeasured attachment to self-made artifacts, untracked private tools, no motive comparison) have begun to be addressed in the generative AI era.
Exclusions: 4 works. One duplicate with the existing corpus (Sarkar 2023 = E57), one clinician-provided intervention, one work confined to workplace context, one confined to civic context.
Bibliographic machine verification: The 27 works with DOIs were queried against the Crossref REST API; first-author surname and title matched for all 27. Zero fabricated DOIs. The remaining 32 are arXiv preprints whose primary pages were individually reached. In the re-check of 2026-08-14, three of them (G24, G30, G33) turned out to have peer-reviewed versions in print, and their records were replaced accordingly.
Limits of coverage: Collection ran with the WebSearch session budget exhausted and OpenAlex returning HTTP 429 on every query even via the polite pool, so exploration relied on the arXiv and Crossref APIs. This corpus cannot claim comprehensiveness. Venues with thin arXiv posting practices (IMWUT, CSCW, JMIR-family) are especially likely to be under-covered.
Skew of review status: 32 of 59 works are unreviewed preprints (non-peer). Research at this intersection is concentrated in 2025–2026, and peer-reviewed publication has not caught up. All numbers reported here were confirmed directly in abstracts; the two works whose abstracts were unreachable carry no numbers.

TL;DR

Research at the intersection has begun to exist, but it is young. The year distribution is 2 works from 2023, 13 from 2024, 17 from 2025, and 27 from 2026; 32 are unreviewed preprints. Making-for-oneself × generative AI is a research frontier that stood up from 2025 onward, and the layer of settled, citable findings is still thin.
Evidence that the entry barrier has dropped is converging across four areas. Smart-home automation (G09, G10, G11), personal automation (G04), self-made assistive technology by blind users (G35, G36), and creation by people with no programming experience (G44, G45) all show, in prototype evaluations, that people can build their own tools in natural language. Almost all of it, however, is immediate evaluation.
What happens after the making is barely measured. Two eight-week longitudinal studies exist (journaling, G16; sleep support, G22) plus one three-week self-tracking deployment (G05, with the researcher as the sole participant); no study has followed the maintenance, abandonment, or persistence of self-made tools beyond a few months.
Critical evidence concentrates in the disability domain and is absent from the hobbyist domain. In real use by visually impaired people, studies report a divergence between users’ understanding of system limits and actual performance (G40) and a 22.2% error-answer rate in a diary study (G41), whereas hobbyist creation (3D printing, shader art, creative coding) yields only positive demonstrations, with no study found that examines failure or limits.
Of the four prior gaps, only attachment measurement has started to move. The IKEA effect for AI co-created text has begun to be demonstrated in RCTs (G50, G51; willingness to pay roughly doubled). Delayed-test longitudinal tracking, tracking of private personal tools, and motive comparison (for oneself vs. for work) remain unfilled in the generative AI literature, as confirmed by a systematic review’s own admission (G56) and a framework paper’s explicit naming (G53).

P1. Making Personal Software with LLMs (14 works)

Empirical work in this track concentrates on smart-home automation and personal workflow automation. Representative studies are Home Assistant rule generation (G09, higher acceptance than a conventional UI with N=56), converting loose goal utterances into automations (G10), natural-language configuration of complex conditions (G11, 91.7% intent-match rate), and discovery of repetitive browser patterns (G04); all show that people can reshape their own environments in natural language. System-side proposals arrive in the same period: an infrastructure that hands workflow orchestration to LLMs (G13), and a proposal to let LLMs build the IoT platform itself (G14).

The same studies also report failure. G10 shows creative reasoning by LLMs while stating that “consistent failure patterns” undermine practicality, and G09 records syntax errors and malformed messages as open issues. Doubt has also begun to turn toward what gets built: a 2026 study interrogates the design homogenization produced by web vibe coding (G06). Lowering the entry barrier and keeping a made thing usable are, at this point, separate questions.

Conceptually, malleable software (software users can reshape while using it) is settling in as the scholarly term (G07, G08). By contrast, no 2023–2026 scholarly work found in this search uses “home-cooked apps” or “situated software” in its title or abstract. The Shirky-derived vocabulary has not been absorbed into the peer-reviewed sphere; malleable software functions as its successor term.

Another observation: no empirical study was found of non-programmers building custom GPTs or personal agents for their private lives. Hits were confined to educational contexts (teachers building custom GPTs for students) or GPT Store policy audits; primary research on private use remains a blank.

P2. Self-Data and Generative AI (19 works)

This is the thickest cluster of the four, but it has a structural feature. Nearly all 19 works are user evaluations of systems built by researchers; not one study examines people configuring or building AI for themselves. LLM-integrated journaling with behavioral sensing (G15, G16), LLM interpretation of wearable data (G20, G23, G24), a physical-activity coach (G21), and sleep support (G22) are all researcher-provided systems. Conversational interventions to raise happiness (G17), an anatomy of personal health agent design (G25), note-taking support for personal knowledge management (G29), self-reflection support combined with a human coach (G31), and goal-setting support for goal progress (G32) share the same structure. The core of making for oneself (that the maker is oneself) has not yet been studied in this cluster.

On measurement, two eight-week longitudinal studies exist (G16 with +6% self-reflection; G22 with improved sleep duration). These run longer than P1, but both cover use within the study period; whether participants kept using the systems afterward was not tracked.

Critical findings appear here too. ChatGPT’s diabetes self-management advice can give generic guidance without necessary verification and thereby produce dangerous advice (G28). AI-driven self-tracking may erode users’ agency and independent thinking (G19). Narrative co-construction with LLMs carries a risk of narrative deference (uncritically accepting the self-story the AI tells) (G18). Older adults with chronic conditions value ownership of their own health narratives and approach AI with cautious optimism (G26).

P3. Self-Made Assistive Technology and Hobbyist Creation (12 works)

The disability domain holds the most mature research cluster in this quadrant. A system that lets blind users build custom visual access programs in three modalities (blocks, natural language, demonstration) (G35) and the research agenda behind it (G36) put making-from-one’s-own-need at the center of generative-AI-era assistive technology. Co-making of physical DIY assistive technology (G34) and personal 3D printing without visual dependency (G39) report both the potential of AI support and quadrant-specific barriers: insufficient spatial assistance and constrained dimensional control. There is also work where the maker is not the disabled person themselves: an exploration of occupational therapists using generative AI in DIY assistive technology design (G37), and a training report that sees the future of assistive technology in generative AI (G42) — the latter is not peer-reviewed research.

The strength of this domain is the thickness of its critical evidence. Visually impaired users tend to blame themselves for errors made by large vision-language models, and there is a wide divergence between their understanding of system limits and actual performance (G40). A two-week diary study measured a 22.2% error-answer rate and a 10.8% refusal rate even as users rated the tool “trustworthy” (G41). The figure of a person who cannot verify using a tool that requires verification is the sharpest instance of this quadrant’s limit problem (P4).

Hobbyist creation, by contrast, is thin. Home-user 3D model generation (G43), shader art by non-programmers (G45), and creative coding (G44) center on positive demonstrations; no peer-reviewed study examining hobbyists’ failures, abandonment, or limits was found in this search. Peer-reviewed work on leisure personal projects with Arduino or Raspberry Pi × generative AI was likewise zero.

P4. Limits Specific to This Quadrant (14 works)

Research on the limits of self-made artifacts splits into two levels.

The first level asks whether the maker can verify what they made. A survey of 162 people across experience levels found a recognition-behavior gap: all groups recognized the risks of AI-generated code, while verification and debugging ability depended on experience (G47). In a controlled experiment with novices, 77% of the unrestricted-AI group failed a maintenance task once AI was withdrawn (39% in the scaffolded-AI group; G46). An exploration of how novices experience vibe coding (G48) belongs to this same level. From the verification side there are proposals to build formal verification into vibe coding (G49) and a protocol to quantify cognitive offloading (G59), but both remain at the proposal stage. That measurement, however, was taken about 30 minutes later in the same session; it is not long-term retention.

The second level asks whether the made thing persists. Typologies of technical debt in LLM-assisted development have begun to form through a multivocal review (G55), and repository data yielded debt patterns of test postponement, incomplete adaptation, and insufficient understanding (G57). Yet the 87-study systematic review itself states that how these conflicting effects shape long-term maintainability and evolvability “remains unclear” (G56). The trust-calibration framework likewise names the over-time nature of trust calibration as an open problem (G53). The limits literature itself acknowledges the long-term blank.

What Happened to the Prior Gap List in the Generative AI Era?

We re-examined the sister note’s gap list (gaps 1, 12, 13, and 14 of end-user-development-novice-learning-literature) against these 59 works.

Gap 1 (almost no delayed-test measurement of retention): still unfilled. The closest generative-AI-era study is G46 (a maintenance task with AI withdrawn), but its measurement comes about 30 minutes later. The longest tracking is G05’s three weeks of self-use, with the researcher as the single participant. G56 and G53 state the long-term unknowns explicitly, so this gap persists in a form the research community itself acknowledges.

Gap 12 (no empirical measurement of attachment to self-made artifacts): partially in motion. The IKEA effect for AI co-created non-physical artifacts has begun to be measured with RCTs (G50, G51). In the follow-up experiment, the ChatGPT co-creation group perceived twice the effort of controls and showed roughly double the willingness to pay. The objects, however, were texts such as marketing documents; no study has yet measured attachment to self-made software built with AI. The journal’s peer-review practices are also unverified, so this finding cannot yet be cited as settled.

Gap 13 (no empirical tracking of private self-use tools): still unfilled. The abandonment and maintenance evidence (G57, G58) all targets public repositories; no study follows the fate of tools individuals made privately for themselves. The closest, G05, covers three weeks of the researcher’s own environment and does not measure the survival of ordinary users’ private tools.

Gap 14 (no motive comparison between making for oneself and making as assigned): only a byproduct signal. As a byproduct of its experience-level design, G47 reports a motive split: non-developers cite accessibility (being able to make what they could not), professionals cite work contexts. This suggests that the origin of the motive divides practice, but no study in the generative AI era was designed with the for-oneself vs. for-work comparison as its main question.

That three of the four gaps remain unfilled does not move the review paper’s contribution. If anything, generative AI raises the stakes of gaps 1 and 13. As the barrier to making falls and the population of privately made, unshared artifacts grows, the fact that nobody measures their verification, maintenance, and abandonment matters more.

Gaps Newly Confirmed by This Corpus

No empirical study of people assembling AI for themselves. All 19 works in P2 evaluate researcher-built systems. Primary research on self-building custom GPTs or personal agents for private use was zero within the search range.
No critical examination in the hobbyist domain. While negative findings accumulate in the disability domain, hobbyist and maker use of generative AI consists of positive demonstrations only.
No study centered on the privacy of self-use tools. Privacy-preserving designs exist (G27), but no empirical study of the risks of feeding one’s own data to a self-made AI tool was found.
The “software for one” vocabulary has not entered scholarship. Along with home-cooked apps and situated software, practitioner terms remain unadopted in the peer-reviewed sphere; only malleable software bridges the two. The asymmetry between scholarly and practitioner discourse mirrors the pattern the sister note observed in the Japanese-language sphere.
No generative-AI-era study of abandonment. Studies of abandonment rates for things built by vibe coding returned zero across multiple queries.

Implications for the Review Paper

Connecting the existing 337 works with these 59 fixes the review paper’s position. For forty years EUD research took people who make for work as its main subject (the E corpus), learning outcomes were measured in educational contexts (the N corpus), and the make-for-oneself motive existed only as a pre-generative-AI lineage (von Hippel, OSS, personal informatics). The generative-AI-era frontier began testing this motive only from 2025, concentrates on the entrance (can people make things), and barely measures the exit (what happens to what they made). A design that measures how long privately made, unverified, unmaintained AI-built tools survive remains, even after this re-check of the gaps, a step no one has taken.

Unverified Items

All 59 items were re-checked on 2026-08-14 (details in the ledger source/review/genai-personal-software/papers.md). Every item carrying a DOI was queried against the Crossref REST API and OpenAlex, and for the preprints the existence of a peer-reviewed version was checked against the arXiv journal-ref field and the publisher’s primary page.

The re-check settled the following.

G24: published as Nature Communications 17, art.1143 on January 12, 2026 — no longer merely accepted
G30: the venue is confirmed as CHI 2026 rather than inferred from the DOI prefix; pages 1–23
G33: the ACM title of record is “The Future of Cognitive Personal Informatics,” included in the CHI 2026 Extended Abstracts
G18: the forthcoming book is Werkheiser, I. & Butler, M. (eds.) Finding Our Place in the Digital World: Philosophical Essays on Technology, Phenomenology, and the Environment (Springer, forthcoming)
G02: confirmed as pages 89–95 of VL/HCC 2024 (held in Liverpool)
G10: confirmed as pages 1–38 of Proc. ACM IMWUT 8(1)
G36: the abstract states “Through my dissertation,” which makes it an overview of doctoral research rather than a report of a single empirical study
G45: the publisher-side record is settled (LNCS, pages 192–206), and the evaluation combines structured user studies with qualitative feedback
G50, G51: confirmed as Marketing Science & Inspirations 20(3), 2–6 and 20(4), 2–11
G52: the abstract was retrieved. It is an incentivized, domain-independent, interactive behavioral experiment measuring overreliance on AI advice
G54: the abstract was retrieved. It is a scale-development study running exploratory and confirmatory factor analysis across multiple samples of university students, with the scale explaining 53.965% of total variance

The items that remain unverified are as follows.

G01: the SpringerLink original is behind an authentication gate, and the number of reviewed papers is absent from the abstract as well [unverified]
G02, G03, G10, G12: the number of participants in the evaluation study is absent from both the abstract and the bibliographic databases [unverified]
G03: the method is settled as a case study of non-programmers building web applications. Whether its account of motivation applies to the self-use context needs checking against the full text [to confirm]
G38: a researcher-led deployment, so the extent of first-person making needs checking [to confirm]
G45: the publisher-side abstract could not be reached; the evaluation format was confirmed through the same authors’ arXiv version [unverified]
G50, G51: the journal is indexed in ERIH PLUS, EconBiz, EBSCOhost, and Ulrichsweb, but its review procedure is not stated on the official page [unverified]
Overall: 32 of the 59 works are unreviewed preprints. Peer-reviewed versions were confirmed for three of them (G24, G30, G33); for the rest, all that was confirmed is that the arXiv journal-ref field is empty

References

All URLs verified reachable on 2026-08-14. “(preprint)” marks unreviewed works.

P1. Making Personal Software with LLMs

G01. Esposito, A., Calvano, M., Curci, A., Desolda, G., Lanzilotti, R., Lorusso, C., & Piccinno, A. (2023). End-User Development for Artificial Intelligence: A Systematic Literature Review. Proc. IS-EUD 2023, LNCS 13917, 19–34. https://doi.org/10.1007/978-3-031-34433-6_2
G02. Ge, Y., Dai, Y., Shan, R., Li, K., Hu, Y., & Sun, X. (2024). Cocobo: Exploring Large Language Models as the Engine for End-User Robot Programming. Proc. 2024 IEEE Symposium on Visual Languages and Human-Centric Computing (VL/HCC), Liverpool, 89–95. https://doi.org/10.1109/vl/hcc60511.2024.00020
G03. Weber, I. (2025). Feasibility of AI-Assisted Programming for End-User Development. arXiv:2512.05666 (preprint). https://arxiv.org/abs/2512.05666
G04. Jiang, S., & Nam, D. (2026). Motif: Discovering and Automating Personal Web Workflows. arXiv:2607.10531 (preprint). https://arxiv.org/abs/2607.10531
G05. Wang, Z., Hu, E., Rucker, M., & Barnes, L. E. (2026). PSI: Shared State as the Missing Layer for Coherent AI-Generated Instruments in Personal AI Agents. arXiv:2604.08529 (preprint). https://arxiv.org/abs/2604.08529
G06. Shin, D., Gao, A., Pang, R. Y., Lee, J., Reinecke, K., & Tseng, E. (2026). Interrogating Design Homogenization in Web Vibe Coding. arXiv:2603.13036 (preprint). https://arxiv.org/abs/2603.13036
G07. Cifliku, B. (2026). Hidden Technical Debt in Generative (GenUI) and Malleable User Interfaces. arXiv:2604.16354 (preprint). https://arxiv.org/abs/2604.16354
G08. Min, B., Jiang, P., Huang, Z., & Xia, H. (2026). Gradual Generation of User Interfaces as a Design Method for Malleable Software. arXiv:2601.17975 (preprint). https://arxiv.org/abs/2601.17975
G09. Giudici, M., Sironi, A., Villa, I., Scherini, S., & Garzotto, F. (2025). Generating HomeAssistant Automations Using an LLM-based Chatbot. arXiv:2505.02802 (preprint). https://arxiv.org/abs/2505.02802
G10. King, E., Yu, H., Lee, S., & Julien, C. (2024). Sasha: Creative Goal-Oriented Reasoning in Smart Homes with Large Language Models. Proc. ACM IMWUT, 8(1). https://doi.org/10.1145/3643505
G11. Shi, Y., Liu, X., Yu, C., Yang, T., Gao, C., Liang, C., & Shi, Y. (2024). Bridging the Gap between Natural User Expression with Complex Automation Programming in Smart Homes. arXiv:2408.12687 (preprint). https://arxiv.org/abs/2408.12687
G12. Carcangiu, A., Manca, M., Mereu, J., Santoro, C., Simeoli, L., & Spano, L. D. (2025). Tell-XR: Conversational End-User Development of XR Automations. arXiv:2504.09104 (preprint). https://arxiv.org/abs/2504.09104
G13. Fan, S., Cong, X., Fu, Y., Zhang, Z., Zhang, S., Liu, Y., Wu, Y., Lin, Y., Liu, Z., & Sun, M. (2024). WorkflowLLM: Enhancing Workflow Orchestration Capability of Large Language Models. arXiv:2411.05451 (preprint). https://arxiv.org/abs/2411.05451
G14. Cheng, Y., Xu, M., Zhang, Y., Li, K., Wang, R., & Yang, L. (2024). AutoIoT: Automated IoT Platform Using Large Language Models. arXiv:2411.10665 (preprint). https://arxiv.org/abs/2411.10665

P2. Self-Data and Generative AI

G15. Nepal, S., Pillai, A., Campbell, W., Massachi, T., et al. (2024). Contextual AI Journaling: Integrating LLM and Time Series Behavioral Sensing Technology to Promote Self-Reflection and Well-being using the MindScape App. arXiv:2404.00487 (preprint). https://arxiv.org/abs/2404.00487
G16. Nepal, S., Pillai, A., Campbell, W., Massachi, T., et al. (2024). MindScape Study: Integrating LLM and Behavioral Sensing for Personalized AI-Driven Journaling Experiences. arXiv:2409.09570 (preprint). https://arxiv.org/abs/2409.09570
G17. Heffner, J., Qin, C., Chadwick, M., Summerfield, C., Kurth-Nelson, Z., & Rutledge, R. B. (2025). Increasing Happiness through Conversations with Artificial Intelligence. arXiv:2504.02091 (preprint). https://arxiv.org/abs/2504.02091
G18. Osler, L. (2025). Knowing Oneself with and through AI: From Self-Tracking to Chatbots. In Werkheiser, I. & Butler, M. (eds.) Finding Our Place in the Digital World: Philosophical Essays on Technology, Phenomenology, and the Environment, Springer (forthcoming). arXiv:2512.03682 (preprint). https://arxiv.org/abs/2512.03682
G19. Nolasco, H. R., Vargo, A., & Kise, K. (2025). AI Solutionism and Digital Self-Tracking with Wearables. arXiv:2505.15162 (preprint). https://arxiv.org/abs/2505.15162
G20. Fang, C. M., Danry, V., Whitmore, N., Bao, A., Hutchison, A., Pierce, C., & Maes, P. (2024). PhysioLLM: Supporting Personalized Health Insights with Wearables and Large Language Models. arXiv:2406.19283 (preprint). https://arxiv.org/abs/2406.19283
G21. Jörke, M., Sapkota, S., Warkenthien, L., Vainio, N., Schmiedmayer, P., Brunskill, E., & Landay, J. A. (2025). GPTCoach: Towards LLM-Based Physical Activity Coaching. Proc. CHI 2025. https://doi.org/10.1145/3706598.3713819
G22. Wang, X., Griffith, J., Adler, D. A., Castillo, J., Choudhury, T., & Wang, F. (2025). Exploring Personalized Health Support through Data-Driven, Theory-Guided LLMs: A Case Study in Sleep Health. Proc. CHI 2025. https://doi.org/10.1145/3706598.3713852
G23. Cosentino, J., Belyaeva, A., Liu, X., Furlotte, N. A., et al. (2024). Towards a Personal Health Large Language Model. arXiv:2406.06474 (preprint). https://arxiv.org/abs/2406.06474
G24. Merrill, M. A., Paruchuri, A., Rezaei, N., Kovacs, G., Perez, J., et al. (2025). Transforming Wearable Data into Personal Health Insights using Large Language Model Agents. Nature Communications, 17, art.1143 (published January 12, 2026). https://doi.org/10.1038/s41467-025-67922-y
G25. Heydari, A. A., Gu, K., Srinivas, V., Yu, H., Zhang, Z., et al. (2025). The Anatomy of a Personal Health Agent. arXiv:2508.20148 (preprint). https://arxiv.org/abs/2508.20148
G26. Dai, D., Pakianathan, P. V. S., Treff, G., Sareban, M., Smeddinck, J. D., & Kuoppamäki, S. (2026). Exploring Self-Tracking Practices of Older Adults with CVD to Inform the Design of LLM-Enabled Health Data Sensemaking. arXiv:2603.23733 (preprint). https://arxiv.org/abs/2603.23733
G27. Cui, Y., Emami, A., Prioleau, T., & Singh, N. (2026). If Only My CGM Could Speak: A Privacy-Preserving Agent for Question Answering over Continuous Glucose Data. ACL Findings 2026 (accepted). https://arxiv.org/abs/2604.17133
G28. Hussain, W., & Grundy, J. (2025). Advice for Diabetes Self-Management by ChatGPT Models: Challenges and Recommendations. arXiv:2501.07931 (preprint). https://arxiv.org/abs/2501.07931
G29. Wisoff, J., Tang, Y., Fang, Z., Guzman, J., Wang, Y., & Yu, A. (2025). NoteBar: An AI-Assisted Note-Taking System for Personal Knowledge Management. arXiv:2509.03610 (preprint). https://arxiv.org/abs/2509.03610
G30. Abbas, A., Wohn, C., Jagtap, A., Rho, E. H., Kim, Y.-H., & Lee, S. W. (2026). “Having Lunch Now”: Understanding How Users Engage with a Proactive Agent for Daily Planning and Self-Reflection. Proc. CHI 2026, 1–23. https://doi.org/10.1145/3772318.3790957
G31. Arakawa, R., & Yakura, H. (2024). Coaching Copilot: Blended Form of an LLM-Powered Chatbot and a Human Coach to Effectively Support Self-Reflection for Leadership Growth. Proc. CUI ‘24. https://doi.org/10.1145/3640794.3665549
G32. Schimpf, M., Voigt, J., & Bohné, T. (2026). AI-Assisted Goal Setting Improves Goal Progress Through Social Accountability. arXiv:2603.17887 (preprint). https://arxiv.org/abs/2603.17887
G33. Schneegass, C., Chiossi, F., Cox, A. L., Dritsa, D., Mitrevska, T., Rainey, S., & Wilson, M. L. (2026). The Future of Cognitive Personal Informatics. Extended Abstracts of CHI 2026. https://doi.org/10.1145/3772363.3778687

P3. Self-Made Assistive Technology and Hobbyist Creation

G34. Kosa, S., Lee, B., Li, F. M., Mondal, M., Zhao, Y., & He, L. (2026). Not Seeing the Whole Picture: Challenges and Opportunities in Using AI for Co-Making Physical, DIY-AT for People with Visual Impairments. Proc. CHI 2026. https://doi.org/10.1145/3772318.3791815
G35. Herskovitz, J., Xu, A., Alharbi, R., & Guo, A. (2024). ProgramAlly: Creating Custom Visual Access Programs via Multi-Modal End-User Programming. Proc. UIST ‘24. https://doi.org/10.1145/3654777.3676391
G36. Herskovitz, J. (2024). DIY Assistive Software: End-User Programming for Personalized Assistive Technology. ACM SIGACCESS Accessibility and Computing. https://doi.org/10.1145/3654768.3654772
G37. Li, J., & Aflatoony, L. (2024). Exploring the Potential of Generative AI in DIY Assistive Technology Design by Occupational Therapists. Proc. ASSETS ‘24. https://doi.org/10.1145/3663548.3688506
G38. Jahangir, R., & Ishii, K. (2026). The Accessibility Capability Boundary: Operational Limits and Expansion Potential of AI-Generated Browser-Native Accessibility Systems. arXiv:2605.19638 (preprint). https://arxiv.org/abs/2605.19638
G39. Pan, Y., Liu, S., Lin, J., Shen, J., Wei, W., Zhang, L., Jiang, Y., Shen, Z., & Tang, X. (2026). Personal 3D Printing without Visual Dependency in the Generative AI Era: Opportunities and Challenges. Proc. DIS ‘26 Companion. https://doi.org/10.1145/3802974.3809447
G40. Tang, X., Fang, C., Wang, R., Sun, Y., & Chen, L. (2025). “This is My Fault”, Really? Understanding Blind and Low-Vision People’s Perception of Hallucination in Large Vision Language Models. Proc. UIST ‘25. https://doi.org/10.1145/3746059.3747597
G41. Gonzalez Penuela, R., Jung, C., Lin, K., Hu, Y., & Azenkot, S. (2026). How Multimodal Large Language Models Support Access to Visual Information: A Diary Study With Blind and Low Vision People. Proc. CHI 2026. https://doi.org/10.1145/3772318.3793266
G42. Payne, A. (2025). The Rise of the Robots: Is Generative AI the Future of Assistive Technology? University of Dundee (training report). https://doi.org/10.20933/100001442
G43. Danry, V., Guzelis, O., Huang, H., Gershenfeld, N., & Maes, P. (2025). From Words to Worlds: Exploring Generative 3D Models in Design and Fabrication. 3D Printing and Additive Manufacturing. https://doi.org/10.1089/3dp.2023.0309
G44. Angert, T., Suzara, M., Han, J., Pondoc, C. L., & Subramonyam, H. (2023). Spellburst: A Node-based Interface for Exploratory Creative Coding with Natural Language Prompts. Proc. UIST ‘23. https://doi.org/10.1145/3586183.3606719
G45. Yuksel, K., & Sawaf, H. (2026). AI Co-Artist: A LLM-Powered Framework for Interactive GLSL Shader Animation Evolution. EvoMUSART 2026, LNCS. https://doi.org/10.1007/978-3-032-24350-8_13

P4. Limits Specific to This Quadrant

G46. Sankaranarayanan, S. (2026). Mitigating “Epistemic Debt” in Generative AI-Scaffolded Novice Programming using Metacognitive Scripts. arXiv:2602.20206 (preprint). https://arxiv.org/abs/2602.20206
G47. Fawzy, A., Tahir, A., & Blincoe, K. (2026). From Prompting to Verification: How Experience Shapes Vibe Coding Practices. arXiv:2605.24521 (preprint). https://arxiv.org/abs/2605.24521
G48. Gama, K., Calegario, F., Jackson, V., Nolte, A., Morais, L. A., & Garcia, V. (2026). “Can you feel the vibes?”: An Exploration of Novice Programmer Engagement with Vibe Coding. Proc. ICSE 2026. https://doi.org/10.1145/3786580.3786992
G49. Mitchell, J., & Shaaban, Y. (2025). Position: Vibe Coding Needs Vibe Reasoning: Improving Vibe Coding with Formal Verification. ACM SIGPLAN LMPL ‘25. https://doi.org/10.1145/3759425.3763390
G50. Czuprák, N., & Németh, R. (2025). The IKEA Effect in Human-AI Collaboration: Does the Effect Exist for Non-Physical Products? Part I. Marketing Science & Inspirations, 20(3), 2–6. https://doi.org/10.46286/msi.2025.20.3.1
G51. Czuprák, N., & Németh, R. (2025). The IKEA Effect in Human-AI Collaboration: Does the Effect Exist for Non-Physical Products? Part II. Marketing Science & Inspirations, 20(4), 2–11. https://doi.org/10.46286/msi.2025.20.4.1
G52. Klingbeil, A., Grützner, C., & Schreck, P. (2024). Trust and Reliance on AI — An Experimental Study on the Extent and Costs of Overreliance on AI. Computers in Human Behavior, 160, 108352. https://doi.org/10.1016/j.chb.2024.108352
G53. Araujo, T. (2026). Unpacking the Dynamics of Generative AI Use in Our Daily Lives: Towards an Integrative Trust Calibration Framework. AI & Society, 40. https://doi.org/10.1007/s00146-026-03279-0
G54. Gümüş, M. M. (2026). Developing a Scale for Generative AI Overreliance and Investigating Its Associations with Critical Thinking and Creativity. SSRN (preprint). https://doi.org/10.2139/ssrn.6891552
G55. Ehsani, R., Rawal, S., Cai, Y., & Chatterjee, P. (2026). Faster Code, Deeper Debt? A Multivocal Literature Review on Technical Debt and Its Early Signs in LLM-Assisted Software Development. arXiv:2606.14796 (preprint). https://arxiv.org/abs/2606.14796
G56. Matias, B. C., Freire, S., Freitas, J., Fronchetti, F., Damevski, K., & Spinola, R. (2026). A Survey on Large Language Model Impact on Software Evolvability and Maintainability: the Good, the Bad, the Ugly, and the Remedy. arXiv:2601.20879 (preprint). https://arxiv.org/abs/2601.20879
G57. Al Mujahid, A., & Imran, M. M. (2026). “TODO: Fix the Mess Gemini Created”: Towards Understanding GenAI-Induced Self-Admitted Technical Debt. arXiv:2601.07786 (preprint). https://arxiv.org/abs/2601.07786
G58. Selvanayagam, N., Ghaleb, T. A., & Abdellatif, M. (2026). Self-Admitted Technical Debt in LLM Software. arXiv:2601.06266 (preprint). https://arxiv.org/abs/2601.06266
G59. Aiersilan, A. (2026). The Vibe-Check Protocol: Quantifying Cognitive Offloading. arXiv:2601.02410 (preprint). https://arxiv.org/abs/2601.02410