statements

I study whether people are still deciding when AI acts for them, and whether the designers who build with AI are still deciding the design. My research investigates how to design with and for AI interaction that supports the autonomy of users and workers, ethically and productively.

“How to ethically and productively design with AI” means studying how to use AI conscientiously to design interactive systems, so that the designer shapes the system in ways that follow from value-based deliberation rather than whatever the tool makes easy. “How to design AI” means studying the concrete design decisions that go into building data-driven interactive systems and their interfaces, by understanding both the explicit handcrafting that goes into them and the automated optimization processes that shape them.

All my work is motivated by the question of how to develop a society that nourishes AI’s capabilities while not just respecting but also supporting human autonomy in a deeply relational world. Autonomy underlies my care for privacy [1, 2] and my push to preserve design intent when designing with AI [3, 4]. My framework for designing AI for human autonomy, under submission at CHI 2027 [5], responds to the long-standing negative impact of technologies on autonomy (dark patterns, which I have measured with users [6] and with designers [7]) and to new challenges, such as the sycophancy that reinforcement learning from human feedback produces [8]. It is equally motivated by the opportunities AI opens: helping people become more autonomous through bright patterns that prompt reflection [9], and enabling new capacities. And it untangles the complex context that distinguishes a design that undermines someone’s autonomy from one that appropriately limits it.

The same question applies to the people who build these systems. Lessig argued that code is law: architecture regulates behavior most thoroughly because it is the least visible of the forces that regulate us, and the individual at the center of law, markets, norms, and architecture is a “pathetic dot,” moved by forces it may take for its own choice [10]. Designers today are moved by the metrics they are handed, the deadlines they face, the norms of their teams, and the AI tools that now sit in every step of their work [4, 11]. Through research and teaching, I want to produce visionary, conscientious designers rather than pathetic ones shaped by external forces.

I have published 8 full research papers and 11 workshop papers and extended abstracts (13 as first or co-first author) at CHI, CSCW, FAccT, DIS, AutomotiveUI, UIST, IEEE ITSC, and Zeitschrift für Psychologie, and have three further first-author papers under review at CHI 2027 [4, 5, 20]. My work has received a Best Paper Honorable Mention at CHI 2024 [15] and the Best Late-Breaking Work Award at CHIWORK 2026 [7], and has been supported by a Digital Life Initiative Doctoral Fellowship, a Siegel Public Interest Technology Impact Fellowship, and a Public Interest Responsible Technology Fund grant. It also reaches beyond academia: brightpatterns.org, UEEQ.org, and a public interactive autonomy rubric put my instruments in practitioners’ hands.

A. Foundation: Autonomy by Design

Should this feature ask the user, persuade the user, restrict the user, or simply act? Teams building AI answer that question weekly and settle it by instinct or by what the law requires. In Autonomy by Design [5], I took a first set of guidelines from the literature through twelve recorded versions in 24 co-design interviews with 8 philosophers of autonomy, 13 AI designers from 13 products, and 3 legal scholars. The framework evaluates one feature on three criteria (authentic values, the value–action link, and the user as informed decider) at four levels: undermining, appropriately limiting, respecting, and supporting. It unifies major tensions between philosophical traditions in a pluralistic framework. These distinctions become more urgent as models increasingly purport to interpret users’ “intent” without distinguishing needs from momentary wants. Practitioners building AI systems for mental health, creativity support, hiring, and work automation used the framework to score features, understand their implications, and change them. The framework is published as an interactive rubric at human-ai-autonomy-rubric.surge.sh, with participant evidence for every co-design decision.

B. End-User Autonomy: Resisting Manipulation and Protecting Information

My first thread gives users, designers, and regulators instruments for recognizing manipulation, and gives designers an antonym to design toward. I defined bright patterns (persuasive designs that serve users’ goals and wellbeing over their momentary desires and business objectives) and started brightpatterns.org [9]. With my mentee Doris Rhomberg, I extended the User Experience Questionnaire with autonomy-focused items to distinguish appropriate from indecent persuasion [13]. In Deception Is Ugly [6], 126 social media users rated 15 dark patterns: perceived deception co-occurred with unattractiveness at r = .93, while the standard UEQ-S flagged one of the 15 patterns as problematic; our autonomy items flagged six as deceptive and eight as pressuring — indicating that user experience metrics need to extend beyond aesthetics and pragmatic qualities to capture manipulation.

The choice of instrument also changes what designers decide. In Making Indecent Persuasion Visible [7, 20], 141 UX professionals who saw our persuasion-focused data rejected manipulative interfaces at nearly twice the baseline rate (56.2% vs. 30.0%), and their 1,313 written justifications shifted from business and aesthetic framings to manipulation, trust, and autonomy. Importantly, their confidence stayed equally high without data: designers were not cautious; they were confidently uncalibrated.

My privacy research is motivated by autonomy over information, which I study through the lens of contextual integrity [16]. In Privacy of Groups in Dense Street Imagery [1], we ran a penetration test on 25 million dashcam images from New York City and showed that blurring faces and license plates does not prevent inference of sensitive group affiliations; we built a typology of identifiable groups and recommendations for researchers using such data. In Regaining Trust [2], a transparent walkthrough increased trust and reduced perceived creepiness of camera-based in-car health monitoring without lengthening onboarding. Under a standing IRB protocol with Wendy Ju, I am currently conducting field studies on privacy notices for autonomous robots in New York City, comparing privacy expectations across demographics and jurisdictions.

C. Designer Autonomy: From Pathetic to Conscientious Designers

My second thread studies the people who build technology while AI takes part in their work. In “I Think We Did This” [4], I interviewed eighteen interaction design students with their repositories and chat logs open: the artifact walk-through, because the record can contradict the self-report where AI’s influence would otherwise go undetected. Design intent eroded along two paths: shared accounts and AI-merged work left no record of who decided what, and the tool offered directions so readily that they became the design. Students pushed back through design-intent cosmetics (redrawing a generated diagram by hand, restoring the feeling of authorship without genuine design intent) and through design-intent recapture, in which they repaired, steered, or guarded the idea. We observe this not as a short-term phenomenon but as a longitudinal trend across four course offerings, in which grades increasingly top out as quality proxies no longer discriminate once AI fills them. Across 203 repositories and 23,065 commits from three cohorts, AI-use disclosure rose from 0% to 66% while attribution of what the tool contributed stayed rare [11]: disclosure norms built for episodic acts cannot capture ambient co-creation. While AI cannot yet do meaningful design, with AI it becomes harder to see what design has meaning.

The forces on designers also live in infrastructure. In My Precious Crash Data [14], twelve employees of autonomous-vehicle companies revealed two previously unknown barriers to sharing safety-critical data: datasets embed costly safety knowledge, so sharing is political even within a company, and safety knowledge is treated as a private competitive advantage rather than a public good. To initiate industry–academia discussion around these pressures, I co-organized Bridge Over Troubled Water at CHI 2026 on aligning commercial incentives with ethical design [17], Interrogating GenAI Augmentation at CHIWORK 2026 toward a professional code of conduct for AI-augmented work [18], and “Nobody Did This” at CSCW 2026 on contribution and accountability in agent-mediated collaboration [19].

Together, my work develops an account of human autonomy as a value that must increasingly be designed for actively rather than merely respected, as AI and its incentive structures mediate our digital and physical lives.

D. Future Directions

1. Autonomy audits for AI products. Whether a product changes after a team has applied our framework is an open question I plan to answer through longitudinal field studies with product teams. The framework’s structure also invites AI-assisted pre-screening of features against the rubric, which I want to develop and validate against expert judgments. Working with additional legal scholars, I want to map the framework onto already-binding regulations to identify the minimum level of autonomy that the law currently requires. I also plan to return to the driver–vehicle handover with the framework in hand, examining whether automation-level transitions meet the framework’s criteria for appropriately limiting, rather than undermining, driver control.

2. Fine-tuned nudging: targeted manipulation. With Helen Nissenbaum, I have begun studying the minimum signals an AI system needs to send to change what a person does. Frontier AI companies present themselves as benevolent societal actors; we investigate how far a commercially or politically motivated adversary could get with personal data and customized open models, as dark-pattern tactics combined with AI personalization can be targeted to individual vulnerabilities [21].

3. How AI-assisted design practices harm vulnerable populations. My work shows that designers can unwittingly lose control over their design when using AI. An open question is whether these practices translate to real harms downstream. Designers increasingly replace real user research with AI-generated personas and synthetic user data. These proxies encode majority norms and demographic assumptions, potentially erasing the people least represented in training data (users with non-standard abilities, lower digital literacy, or marginal social positions) before any product decision is made. I want to study whether this is empirically the case, and whether AI-assisted design can alternatively be structured to surface vulnerable users’ needs more reliably than individual designers alone.

4. Infrastructure for conscientious design. I will carry lessons from our workshops with UX practitioners and HCI researchers into developing a code of conduct for AI use in design: not just disclosure that AI was used, but specification of how it was used and what steps ensured it did not shape design thinking in unintended ways. I will also continue building ethical user experience metrics, including a validated persuasion scale as part of a modular UEQ extension. A parallel infrastructure thread extends the AV safety-data work [14] toward shared data infrastructure for the industry: the organizational and competitive barriers identified in that paper point to the need for neutral intermediaries and governance structures, a conversation I began in Bridge Over Troubled Water [17].

5. Values in vibe coding. A project nearing completion studies what value defaults vibe coding tools encode. Leaked system prompts reveal that some tools instruct the model to prioritize accessible and ethical design, while others push for engagement-maximizing or visually showy output [23]. Beyond explicit prompts, many defaults likely emerge from training data: given that much of the commercial web employs deceptive design, LLMs may have absorbed these patterns [12]. We are investigating whether this is empirically the case and what it means for design practice.

References

1. Matt Franchi*, Hauke Sandhaus*, Madiha Zahrah Choksi, Severin Engelmann, Wendy Ju, and Helen Nissenbaum. 2025. Privacy of Groups in Dense Street Imagery. In FAccT ‘25, 2874–2891. https://doi.org/10.1145/3715275.3732185 (*equal contribution)

2. Hauke Sandhaus, Madiha Zahrah Choksi, and Wendy Ju. 2024. Regaining Trust: Impact of Transparent User Interface Design on Acceptance of Camera-Based In-Car Health Monitoring Systems. In Adjunct Proceedings of AutomotiveUI ‘24, 203–208. https://doi.org/10.1145/3641308.3685048

3. Hauke Sandhaus, Qiuquan Gu, Maria Teresa Parreira, and Wendy Ju. 2025. Co-Designing with Transformers: Unpacking the Complex Role of GenAI in Interactive System Design Education. In DIS ‘25, 1228–1243. https://doi.org/10.1145/3715336.3735805

4. Hauke Sandhaus and Wendy Ju. “I Think We Did This”: Erosion and Reassertion of Students’ Design Intention in AI-Saturated Work. Under review at CHI 2027.

5. Hauke Sandhaus, Jen Semler, Helen Nissenbaum, and Qian Yang. Autonomy by Design: Co-Designing a Framework to Evaluate Whether AI Interfaces Support User Autonomy. Under review at CHI 2027.

6. Hauke Sandhaus and Doris Maria Rhomberg. 2026. Deception Is Ugly: Linking Aesthetic Judgment to Perceived Manipulation in Dark Patterns. Zeitschrift für Psychologie. https://doi.org/10.1027/2151-2604/a000635

7. Hauke Sandhaus, Doris Maria Rhomberg, and Helen Nissenbaum. 2026. Making Indecent Persuasion Visible: How Evaluation Metrics Shape UX Designers’ Ethical Reasoning. In CHIWORK ‘26 Late-Breaking Work. https://doi.org/10.1145/3805029.3818279 Best Late-Breaking Work Award.

8. Mrinank Sharma, Meg Tong, Tomasz Korbak, et al. 2024. Towards Understanding Sycophancy in Language Models. ICLR 2024.

9. Hauke Sandhaus. 2023. Promoting Bright Patterns. In CHI ‘23 Workshop: Designing Technology and Policy Simultaneously. arXiv:2304.01157

10. Lawrence Lessig. 2006. Code: Version 2.0. Basic Books.

11. Maria Teresa Parreira, Pranav Prabhat Sinha, Hauke Sandhaus, and Wendy Ju. 2026. Visible Adoption, Untracked Contribution: GitHub Evidence of the Accountability Gap Across Three Cohorts of an HCI Prototyping Course. In CHIWORK ‘26 Adjunct Proceedings. arXiv:2606.14054

12. Veronika Krauß, Mark McGill, Thomas Kosch, Yolanda Thiel, Dominik Schön, and Jan Gugenheimer. 2025. “Create a Fear of Missing Out”: ChatGPT Implements Unsolicited Deceptive Designs in Generated Websites Without Warning. In CHI ‘25. https://doi.org/10.1145/3706598.3713083

13. Doris Maria Rhomberg* and Hauke Sandhaus*. 2024. Towards Quantifying Ethical User Experience: Evaluating User Perceptions of Dark Patterns in Social Media. In CHI ‘24 Workshop: Mobilizing Research and Regulatory Action on Dark Patterns and Deceptive Design Practices. CEUR-WS Vol. 3720. (*equal contribution)

14. Hauke Sandhaus, Angel Hsing-Chi Hwang, Wendy Ju, and Qian Yang. 2025. My Precious Crash Data: Barriers and Opportunities in Encouraging Autonomous Driving Companies to Share Safety-Critical Data. Proceedings of the ACM on Human-Computer Interaction 9, 7 (CSCW), Article 348. https://doi.org/10.1145/3757493

15. Emma Harvey, Hauke Sandhaus, Abigail Z. Jacobs, Emanuel Moss, and Mona Sloane. 2024. The Cadaver in the Machine: The Social Practices of Measurement and Validation in Motion Capture Technology. In CHI ‘24, Article 768. https://doi.org/10.1145/3613904.3642004 Best Paper Honorable Mention.

16. Helen Nissenbaum. 2010. Privacy in Context: Technology, Policy, and the Integrity of Social Life. Stanford University Press.

17. Thomas Eßmeyer, Johanna Gunawan, Colin M. Gray, Lorena Sánchez Chamorro, Alberto Monge Roffarello, Katie Seaborn, Hauke Sandhaus, Kai Lukoff, Gian-Luca Savino, Shruthi Sai Chivukula, and Satoshi Nakamura. 2026. Bridge Over Troubled Water: Aligning Commercial Incentives With Ethical Design Practice To Combat Deceptive Patterns. In Extended Abstracts of CHI ‘26. https://doi.org/10.1145/3772363.3778758

18. Hauke Sandhaus, Kashif Imteyaz, Mohammed Almutairi, Pooja Prajod, Divya Ramesh, Saiph Savage, Qian Yang, and Michael Muller. 2026. Interrogating GenAI Augmentation for CHIworkers: Strategies for Professional Autonomy and Accountability. In CHIWORK ‘26 (Workshop). https://doi.org/10.1145/3805029.3818271

19. Kashif Imteyaz, Mohammad Rashidujjaman Rifat, Divya Ramesh, Steven R. Rick, Simo Hosio, Hauke Sandhaus, Advait Sarkar, Christoph Riedl, and Saiph Savage. 2026. “Nobody Did This”: Contribution, Originality, and Accountability in Agent-Mediated Collaboration. In Companion Publication of CSCW ‘26. arXiv:2607.26387

20. Hauke Sandhaus, Doris Maria Rhomberg, and Helen Nissenbaum. Persuading Metrics: How User Evaluation Data Shapes Professional Designers’ Willingness to Reject Dark Patterns. Under review at CHI 2027.

21. Thomas Kosch, Veronika Krauß, Christopher Katins, Dominik Schön, Mark McGill, and Jan Gugenheimer. 2026. The AI Accomplice: Exploring Generative Artificial Intelligence in Facilitating and Amplifying Deceptive Designs. In Extended Abstracts of CHI ‘26, Article 971. https://doi.org/10.1145/3772363.3778770

22. Martin Schrepp. 2021. Measuring User Experience with Modular Questionnaires. In International Conference on Advanced Computer Science and Information Systems (ICACSIS). https://doi.org/10.1109/ICACSIS53237.2021.9631321

23. asgeirtj. 2026. System prompts leaks. GitHub repository. https://github.com/asgeirtj/system_prompts_leaks (accessed February 2026)

24. Cornell Tech Vibe Coding Summer 2026. Ethical Vibe Coding for High School Students. https://cornell-tech-vibe-coding-summer-2026.github.io/website/ (accessed October 2026)

I teach the way I research: for autonomy. My research shows how designers become pathetic designers, moved by tools, deadlines and norms they take for their own choices. My teaching aims to build the opposite: visionary, conscientious designers who can articulate what they intend, steer AI toward it, and defend it against external influences, such as metrics that nudge them.

A. Teaching Philosophy

I was trained in Human-Computer Interaction myself (B.Sc. Creative Technology, University of Twente; M.Sc. Human-Computer Interaction, Bauhaus-Universität Weimar), and I believe the challenges of the world increasingly require generalist, boundary-crossing thinkers. This is true not only in business, where AI-improved knowledge work makes one-person startups increasingly common, but also in research and development, where AI as a general-purpose technology is creeping into every domain. My underlying teaching outcome is to enable young designers, developers and researchers to still think slow and deep while working faster, and to educate the next generation of conscientious AI co-designers and co-thinkers.

This matters because the old signs of good design work no longer show whether the work was done. In my study of eighteen students in an interactive device design course, the tool sat in every step, the two-week deadline rewarded finished work, and unacknowledged use became the norm. For example, one team redrew a model’s diagram by hand “to reflect effort” while the design stayed the model’s. Design intent survived only where a student brought a conviction firm enough to state and then steered the tool toward it. No student can set that norm alone, so the course has to. This means three things in practice.

First, intent before generation. Students record what they intend, what they know and what they do not know before they open an AI tool, so the record of their design predates anything the model produced. Second, process over polish. In my classes I rely on fast iterative loops with user testing between them, and each submission names an intent the model did not supply, an alternative the student set aside, and the reason a decision went the way it did; I grade that reasoning rather than the finish of the artifact. Third, boundaries and exits as competences. Students decide in advance what the tool may touch and when they will stop using it, and show afterwards where they did. I teach with AI rather than against it: students learn the tools deeply, and learn when not to use them.

B. Teaching Experience

Ethical Vibe Coding (TECHIE 1121, Cornell Tech, Summer 2026). I designed and head-instructed this course for high school students in Cornell Tech’s Summer Innovation Intensives. Except for Git familiarity, no programming experience is required. I teach how to use AI ethically, how to build ethical things with AI, how to push back on AI harms, and how to use AI for ethics. Students build real apps with AI coding tools, learn how these tools work, probe bias and vulnerabilities in generated output, and apply the Values at Play framework to their own projects. In the end-of-program survey (16 respondents), 88% of students said they learned “a lot” or “a great deal”, students rated the content 4.56 out of 5 for how interesting it was, and they rated the program 4.38 out of 5 for making them more interested in studying technology and engineering. For example, one student named the best part of the program as “learning how to vibe code because I believe it will be an essential tool in the future.” Cornell Tech asked to teach the course again as a “showstopper” class.

Developing and Designing Interactive Devices (INFO 5345 / CS 5424 / ECE 5413, Cornell Tech, Fall 2023 and Fall 2025). As head TA for Wendy Ju’s master’s prototyping course, I revised all lab scripts for updated hardware, set up the class infrastructure, managed grading and gave a guest lecture. I worked with many teams on their coding and design problems as they built a working physical–digital prototype every two weeks. In 2025 I updated the course for AI co-design of technology and built WendyTA, an LLM teaching assistant that supports students’ projects while prompting reflection rather than supplying answers. The course has produced three workshop papers and two full papers, one published at DIS 2025 and one under review at CHI 2027, and many of its students asked to continue working with me afterwards, including through Specialization Projects (INFO 5920).

Privacy in the Digital Age (CS 5436 / INFO 5303, Cornell Tech, Fall 2024). As head TA for Helen Nissenbaum and Vitaly Shmatikov, I taught the section on usable privacy: when privacy settings are needed, how companies can design them to be usable, and which tricks, such as dark patterns, companies have used to make people share more data than they want, which shows that the “privacy paradox” is largely a product of design. I also created an in-class participatory exercise on evaluating privacy policies. Nissenbaum and Shmatikov nominated me for the best teaching assistant award for this course.

HCI and Design (INFO 6410 / CS 5682, Cornell Tech, Fall 2026). In a part-time teaching assistantship with Nicki Dell and Thijs Roumen, I am updating the course for AI co-design and have given full guest lectures.

Conscientious Tech Design Workshop (Siegel PiTech PhD Impact Fellows, Summers 2025 and 2026). With Helen Nissenbaum I co-designed an eight-week workshop that operationalizes Values at Play into whiteboard activities and assignments, so fellows discover, implement and verify values in their own public-interest projects. The 2026 edition was funded by a Public Interest Responsible Technology Fund teaching grant.

Further teaching. I taught Human-Computer Interaction with Gilly Leshed, prepared materials for Deep Learning (CS 5787) so that students learn rather than skate by with AI, including final-exam questions that resist AI-assisted answers, and served as a teaching assistant for User Experience and User Research (INFO 5305), Computer-Mediated Communication (INFO 4450, Spring 2022) and Specialization Projects (INFO 5920). I began teaching in 2014, with Arduino programming for secondary-school pupils.

C. Mentorship

I have mentored master’s students, visiting researchers and high school students on research in ethical design and human–AI interaction. Mentees have co-authored papers at DIS, CHIWORK and in Zeitschrift für Psychologie (including the CHIWORK 2026 Best Late-Breaking Work) and have gone on to design roles at TikTok and Drei Österreich. Students from my courses regularly ask to keep working with me, several of them through Specialization Projects. As in my classroom teaching, my goal is for students to take ownership of their projects; my mentoring statement describes how.

D. Teaching Interests and Goals

I am well equipped to teach undergraduate and graduate courses in human-computer interaction, usable privacy, interaction design, user research and evaluation, and AI ethics. In addition, I am eager to develop three courses:

1. Ethical Design in the Age of AI (undergraduate or graduate). Students study how AI systems can undermine, limit, respect or support user autonomy, examine real products through the Autonomy by Design framework, and propose redesigns. The course bridges UX research, philosophy of autonomy and technology policy, and trains designers to understand the human consequences of technical choices.

2. Ethical Vibe Coding (introductory, any discipline). Students build working applications with AI coding tools, learn how the tools work, probe bias, manipulation and accessibility in generated output, and apply Values at Play to their own projects. No programming experience is required. The course welcomes students from across disciplines into critical engagement with technology.

3. Designing for Human Autonomy (graduate seminar). Students evaluate real AI features and agents with the Autonomy by Design framework, read the philosophical and legal traditions behind it, and redesign a feature from undermining to supporting. The seminar connects values-centered design research to the specific challenges of AI-mediated interaction.

These courses bridge technical courses on machine learning and systems with broader courses on technology and society. I welcome co-teaching with colleagues in computer science, information science and design, since my courses are where their students encounter users, ethics and consequences. My goal is for students to be able to think through the influences affecting their work and design thinking at the speed AI now allows.

Diversity, equity and belonging. The communities most affected by manipulative and autonomy-undermining AI are often the least represented in the classrooms where these systems are designed. Through the Siegel PiTech fellowship I brought ethical design methods to fellows working with under-resourced civic organizations, including YAI; through Ethical Vibe Coding I taught high school students — many from groups underrepresented in computing — to build and audit with AI responsibly. I will continue recruiting and mentoring students from underrepresented groups, designing courses that engage affected communities as participants rather than objects of study, and contributing to efforts to make HCI and design more equitable.

I believe the role of a mentor is to help students understand what research is, and then to let them decide whether research is what they want to pursue. That means guiding mentees toward becoming independent researchers with their own skills, motivations and methods, and teaching them how to find their niche and their fit within the research community.

How I mentor

Four strategies set my mentees up for success.

• I have them carry out the research steps themselves, because motivation and goal setting have to come from within. I rarely assign concrete tasks; tasks and capacity are for each person to set. Instead I ask mentees to take ownership of their projects, define their research questions and build their own action plan. I often ask what they think the right thing to do is, because I want them to develop confidence in their own intuition.

• I ask them to position every contribution against the literature, because academic success rests on being able to situate one’s work, build on others’ knowledge and reflect on it. This builds the self-reliance to navigate research problems without me.

• I give honest feedback, even when it hurts, and I give it early. A mentee who hears a hard assessment from me first will not hear it for the first time from a reviewer.

• I teach research skills and processes in one-to-one sessions, I am available for help at any time, and I listen. Academic life is complicated, and part of my job is to be a supportive presence while mentees find their way through it.

What mentees learn

• How to do literature research, and the diversity of methods and contributions common in my area.

• The difference between seeking knowledge and developing a product.

• How to develop a meaningful research question.

• How to choose the method that best answers that question.

• How to balance rigorous and creative research.

What I expect

Open communication, and a willingness to engage in thoughtful discussion about progress and challenges. I expect mentees to tell me what is not working, in the research and in our working relationship, as early as they can.

Evidence

My mentees’ work speaks for the approach. Doris Rhomberg joined me as a visiting researcher from TU Wien and co-authored three papers with me, including a journal article in Zeitschrift für Psychologie and the Best Late-Breaking Work at CHIWORK 2026; she now designs for Drei Österreich in Vienna. Qiuquan Gu co-authored our DIS 2025 paper and is now a product designer on TikTok’s AI social team. Pranav Sinha, a master’s student at Columbia, co-authored a CHIWORK 2026 workshop paper on AI accountability across three course cohorts. Two further student-led projects, a reflective vibe-coding sandbox and an audit of dark patterns in vibe-coded interfaces, were accepted to the DPCHI 2026 workshop. I currently mentor Irene Wu (Cornell) on bias and inclusive design in AI tools, and Hayatt Khan (Mumbai) on tools that help people see through manipulative design and AI-generated noise.

Beyond individual mentees, I co-run an eight-week workshop for PhD Impact Fellows on value-centered design, and I worked in industry before my PhD, including three years at Volkswagen Group, so I can advise students bound for product teams as well as for academia. A student should leave able to state what they intend, defend it against a reviewer, and ship it.