Notes
Research, experiments, and reflections on AI and design, published along with the process of making them.
Designers' work and careers (17)
How AI changes designers' roles, pay, and careers.
-
What Can Designers Do, and Where Can They Add Value, as Personal Data Breaches Continue? (Industry Sources)
2026-10-24
This note examines, through industry sources, what designers can do about continuing personal data breaches and where they can add value. Sources include Japan's Personal Information Protection Commission, Consumer Affairs Agency, and Digital Agency; the FTC, NIST, EDPB, ICO, and NCSC; coordinated sweeps by consumer protection and privacy enforcement authorities; published figures from Verizon, IBM, Tokyo Shoko Research, JNSA, Pew, and JIPDEC; and first-hand accounts from Microsoft, Google, GitHub, Mercari, Nulab, and GOV.UK. The causes of breaches look different depending on the source. Among incidents disclosed by listed Japanese companies, malware infection and unauthorized access accounted for 64.4%, and misdisplay and misdirected email for 20.5%. Among breach reports filed under Japan's privacy law, 76.9% involved paper only and 93.6% involved 1,000 people or fewer, and documents handed to the wrong patient at hospitals and pharmacies and misdirected credit cards were common. People were involved in 62% of breaches, but the leading initial access vector was exploitation of vulnerabilities (31%), so interface design reaches only part of the breach problem. Within that part, regulators and practitioners have moved from designs that ask users to be careful to designs that make errors hard to make. Japan's Digital Agency states that combining a password with a one-time password cannot stop real-time relay phishing, and NIST prohibits password composition rules and requires that password managers and pasting be allowed. Practitioners who deployed passkeys described where to prompt (fewer than 1% of enrollments came from the settings page, and 97% came from prompt pages), what words to use, and how users recover lost credentials as design issues. Consent and cancellation screens have become objects of regulation: in a sweep by 27 consumer authorities, 75.70% of 642 traders used at least one dark pattern, and the FTC settlement with Amazon required a clear button for declining Prime and cancellation as easy as enrollment. In Japan, a Consumer Affairs Agency study group proposed considering comprehensive rules on dark patterns. Designers' added value appears in the work that regulators now request by naming designers (involvement at the design stage, default settings, interdisciplinary teams) and in reductions in support requests counted by practitioners (2FA-related tickets fell by one-third after GitHub made 2FA mandatory). No data were found showing that salaries or job postings for designers with security or privacy expertise have grown. The academic counterpart is designer-security-privacy-value-literature.
-
What Can Designers Do, and Where Can They Add Value, as Personal Data Breaches Continue? (Academic Research)
2026-10-24
This note examines what designers can do about continuing personal data breaches and where they can add value, drawing on 36 works from usable security and privacy research (SOUPS, CHI, USENIX Security, IEEE S&P, PoPETs, CSCW) and research on design ethics. The starting point is research that rejected the view of users as the weak link. Adams and Sasse argued that security mechanisms are designed and breached by people, so human factors must be considered in their design; Herley argued that users who reject security advice are rational in terms of costs and benefits; and Cranor wrote that designers should keep humans out of the loop whenever possible and, where they cannot, build systems that support humans. In interface research, a redesign of Chrome's SSL warning did not reach its goal of improving comprehension, but a design that visually promoted the safe choice led about 30% more users to remain safe. On consent screens, removing the reject button from the first page increased consent by 22 to 23 percentage points, and highlighting the decline option changed decisions substantially, with the effect persisting even on later neutral banners. A crawl of about 11,000 shopping websites found 1,818 dark pattern instances, and 22 third parties offered dark patterns as turnkey solutions. Research on designers themselves found that designers recognize that they could become complicit in dark patterns, are constrained by commercial pressure and the limited scope of their projects, give sympathetic explanations for other designers' dark patterns, and crossed even their own ethical boundaries in fictional commissions. In product teams building smart home cameras, UX was used for privacy innovation but not for security, where established methods were demanded. Designers' added value can be found in choosing, on the basis of evidence, designs that support people where humans cannot be removed from the loop; in turning legal requirements into criteria that can be checked on screens; and in making values visible within organizations. No study was found that measured whether designers' involvement reduces breaches, and no study focused on designers in Japan. The industry counterpart is designer-security-privacy-value-industry.
-
What Should Juniors Do When Generative AI Takes Away the Entry-Level Work? (Industry Sources)
2026-10-11
This note examines, through industry sources, the problem that generative AI takes over simple tasks so that juniors are asked to handle difficult work without the entry-level practice that used to precede it. Sources include primary documents from the UK government, the New York and St. Louis Federal Reserve Banks, the ECB, and JILPT; published figures from Indeed, IPA, the Society of Authors, and others; and first-hand accounts from design and advertising organizations and AI companies. The first half of the premise, that junior postings have fallen while senior postings have risen, is observed in several data sources. At Indeed, entry-level postings fell 7.5% and senior postings rose 14.7% year over year in May 2026; in the UK, entry-level hiring of graphic designers fell 28% year over year; and staff aged 25 and under at UK advertising agencies fell 19.2%. That generative AI caused the decline, however, has not been established. Indeed's entry-level postings have been falling since 2022, the New York Fed reports that the decline in postings for AI-exposed occupations began before ChatGPT and shows no clear additional break afterwards, and neither the UK government nor the ECB attributes the decline to AI. The behaviors that industry sources support for juniors are using AI in ways that keep one's own understanding (in an experiment with 52 mostly junior software engineers, the AI group averaged 50% on a comprehension quiz against 67% for the hand-coding group, and the gap was smaller for those who generated code and then worked to understand it), seeking questions and critique actively (an AI company's internal study found that juniors came to senior staff with questions less often), showing judgment and thinking process, becoming the people who teach the organization what they learn about AI, and choosing entry points that offer training or apprenticeships. All of these rest on first-hand accounts or small studies, and no industry study compared the effects of these behaviors. The academic counterpart is junior-skill-formation-genai-literature.
-
What Should Juniors Do When Generative AI Takes Away the Entry-Level Work? (Academic Research)
2026-10-11
This note examines, through labor economics, research on expertise and apprenticeship, randomized experiments on generative AI and learning, and studies of creative work, the problem that generative AI takes over simple tasks so that juniors are asked to handle difficult work without the entry-level practice that used to precede it. The shrinking of junior employment is observed: in US payroll records, employment of 22- to 25-year-olds in AI-exposed occupations is 19% below the path of their peers, and at firms that adopted generative AI, junior employment fell through slower hiring. When remote work is estimated jointly, however, the generative AI coefficient shrinks, and in Denmark no effect on earnings or hours was found, so attribution of the decline is contested. On what the groundwork did, deliberate practice explains less than 1% of performance differences in professions, so a shortage of practice volume is not the whole problem. What may be lost is the route of learning by doing peripheral work alongside experts; one theory shows that automating entry-level tasks can separate novices from experts even without reducing employment. Randomized experiments on generative AI and learning split by how AI is used. High school students who practiced with an answer-giving AI scored 17% lower once it was removed, whereas asking AI for explanations, and trying first before receiving explanations, kept or improved understanding. From this, the note lists six behaviors for juniors: try first and then ask AI for explanations; use AI for understanding rather than handing it the generation; keep time to check oneself without AI; seek feedback actively; practice the judgment of ideation and selection; and choose work where experts are within reach. Most experiments involve students and programmers, and no study was found that followed the effects of these behaviors in creative juniors over time. The industry counterpart is junior-skill-formation-genai-industry.
-
Do Decision-Makers Without Taste Stop Commissioning Creators? Checking the Problem of Image-Generating AI Through Three Links (Commissioning, Habituation, and the Creator Base) and Relating It to Research on AI Slop
2026-10-11
A post on X by nakamachi_keiji, a manga artist, on October 9, 2026 argued that the problem with image-generating AI is not that AI makes impressive work and takes creators' jobs. Instead, people who lack taste but hold decision-making authority believe that what they had AI make is tasteful and stop commissioning creators, society becomes accustomed to low quality, and, as a result, the base of creators narrows and the culture of making declines. This note checks this causal chain against 39 academic sources and research reports in the psychology of aesthetic judgment, labor economics, HCI, and design history. The entry and part of the exit of the chain have been measured, but the middle, "taste," has not. On the volume of commissions, job posts related to image creation fell by 17% on a global freelancing platform after image-generating AI appeared (Demirci et al. 2025), whereas on a Chinese commission platform for anime-style images, the average price fell by 64% and the number of orders rose by 121%, with the increase coming from cheap personal orders (Zhang et al. 2026). Workers with strong past records were not protected from the decline (Hui et al. 2024), and on Pixiv, uploads by illustrators who did not use AI fell by 10.1% (Kim et al. 2026, not peer reviewed). The mechanism of mistaken belief is plausible. People identified AI-generated poems with only 46.6% accuracy (Porter and Machery 2024), preferred unlabeled AI-generated paintings, and this preference was unrelated to their experience in art (van Hees et al. 2025). People value what they built themselves about as highly as experts' work (Norton et al. 2012) and overestimate their own performance when using AI (Fernandes et al. 2026). However, people with art education could also barely tell AI paintings apart (54.9% accuracy, Huang et al. 2026), so the problem lies less in decision-makers' lack of taste than in the fact that AI images are hard for anyone to identify. No study was found that measured habituation to low quality with AI images; what has been measured is that works become more similar to one another (Zhou and Lee 2024, Doshi and Hauser 2024). A mechanism by which repeated viewing shapes preference exists (Cutting 2003), and in recorded music, the quality of new works did not fall when the cost of making them fell (Waldfogel 2012). The creator base has not been measured among entry-level creators; in other fields, employment of young workers declined (Brynjolfsson et al. 2026, not peer reviewed) and newer users left first (Burtch et al. 2024). Read against research on AI slop, the post's scenario is the commissioning version of the definition that slop is generated content shared with others without review (Willison 2024), and it corresponds to the lemons mechanism, in which good products withdraw when buyers cannot judge quality (Akerlof 1970). At the same time, judging quality by "taste" takes the same form as the 1852 Chamber of Horrors and the 1939 critique of kitsch, and it inherits the criticism about whose standard defines quality.
-
Can the Forward Deployed Designer Stand as a Role?
2026-08-04
An examination that measures the responsibilities of FDEs (Forward Deployed Engineers), strategy consultants, and designers through job postings and primary discourse, then re-situates Nielsen's proposed FDD (Forward Deployed Designer) on top of those differences. The FDD's job content largely overlaps with existing service-design skills; what is new is the previously uncombined pairing of that work with the FDE-style placement (vendor-side employment, responsibility for running artifacts, and a feedback loop into the product). A vacancy exists in the workplace division of labor, but as a job title the market has not yet materialized: the two postings we could confirm have both closed. The watershed for its emergence is whether this placement settles as an equal division of labor rather than subordination to the FDE.
-
Design and the Professions: A Literature Map of Design's Position Seen Through the Theory of Professions
2026-07-19
Is design a profession in the same sense as medicine or law? This note maps that question neutrally against the sociological theory of professions. It builds on Larson's (1977) account of professionalization as market closure, Abbott's (1988) theory of jurisdiction and inter-professional competition, and Freidson's (2001) professionalism as a third logic, and it positions Schön's (1983) reflective practice, Cross's (2001) disciplining of design, Buchanan's (1992) wicked problems, Star & Griesemer's (1989) boundary objects, and Fischer & Scharff's (2000) meta-design. Design carries practical knowledge and a disciplinary basis, yet it has reached the present without the institutional conditions of entry licensure and jurisdictional monopoly. 8 references.
-
From Maker to Editor: A Structural Analysis of the Designer Role Transition in the Age of AI
2026-07-09
A cross-sectional analysis of the phenomenon in which the designer's role shifts from maker to editor/curator, drawing on 19 academic sources and 15 industry sources. Three structural findings: (1) the transition is empirically confirmed, with 71% spending more time on evaluation/curation than original production (Rivera & Russi 2026, 217 practitioners across 43 countries); (2) new judgment typologies have emerged alongside the transition (agency allocation judgment, trustworthiness judgment); (3) behind the efficiency narrative, 'AI management labor' has surfaced as a new form of cognitive burden. Industry data reveal a structural divergence between 90% adoption and only 10% approval.
-
Field Deploy Engineers and Designers: How Field Knowledge Shapes Design Judgment
2026-07-09
Discusses how the professional experience of a field deploy engineer providing global on-site support for superconducting and cryogenic systems functions as an epistemological foundation for design, through career transitions into service designer, business designer, and design consultant. The trajectory in which the object of deployment abstracts from physical systems -> software -> learning experiences -> design methodology can be read as a concrete instance of Schon's reflective practice and Suchman's situated action. As designers' work shifts from 'making' to 'judging' in the AI era, the note shows through one practitioner's career that the sense of field deployment (the judgment to make a designed solution function in the context of the field) becomes a core capability of design.
-
The Economics of AI and Design — Reading the Structural Transformation of Design Labor Through Six Economic Theories
2026-06-29
A map note that cross-analyzes AI's impact on the design labor market through six theories: the task-based model (Acemoglu & Restrepo), job polarization (Autor), the economics of information goods (Shapiro & Varian), the economics of the creative economy (Caves / Throsby), platform economics (Rochet & Tirole), and Baumol's cost disease. It identifies the structural changes in design that each theory illuminates and contrasts them with empirical data (BLS statistics, DID analyses of the freelance market).
-
Structural Factors behind Designer Pay Stagnation — Academic Review (2026)
2026-06-28
Organizes the structural reasons why designer compensation stagnates relative to engineers, PMs, and data scientists, drawing on 34 peer-reviewed papers. Six structural factors: (A) incomplete professionalization (Abbott's jurisdiction theory / Larson's market closure); (B) the wage penalty on cultural creativity (Hwang 2014: technical CCI > cultural CCI); (C) lower earnings toward Throsby's 'inner circle' (Been et al. 2023: full Dutch administrative data); (D) normalized precarity (second-job rate 2x / 'wageless life' / flexploitation); (E) the commodification of design thinking eroding jurisdiction (Johansson-Sköldberg 2013); (F) AI deskilling diminishing bargaining power (Shukla CHI 2025 / Jiang CHI 2026: 54% report reduced influence).
-
Structural Factors behind Designer Pay Stagnation — Industry Trends (2026)
2026-06-28
Examines the structural reasons why designer compensation stagnates relative to engineers, PMs, and data scientists, drawing on BLS public statistics (T1: 18 items), industry surveys (T2: 18 items), and practitioner testimony (T3: 19 items) totaling 55 sources. Graphic designer median $61,300 is 46% of SWE's $133,080. Ten-year growth: GD +2%, SWE +15%, DS +34%. Structural factors: oversupply (bootcamp graduates 6,500-6,900/year), disproportionate layoffs (1.7x expected), measurement difficulty, C-suite absence, platform race to the bottom. McKinsey Design Index shows correlation with enterprise value but makes zero mention of designer compensation.
-
Designer Careers That Will Generate Value Over the Next Five Years: Industry Trends (2026)
2026-06-28
An analysis of designer career paths likely to generate value in the AI era, drawing on 47 sources across job market and compensation data (BLS/Glassdoor/LinkedIn), consulting surveys (McKinsey/BCG/WEF), vendor trends (Figma/Vercel), and design leader perspectives (Maeda/Zhuo/Nielsen/Spool). Growth areas: AI Experience Designer, Design Engineer, Content Designer. Contracting areas: Graphic Designer, UX Researcher (down 89% from peak hiring).
-
Designer Careers That Will Generate Value over the Next Five Years — Academic Review (2026)
2026-06-28
Synthesizes how designer careers will transform in the AI era from 37 peer-reviewed papers. Identifies five structural changes: (1) roles shift from execution to direction and curation, (2) GenAI is a double-edged sword that both expands creativity and increases fixation, (3) de-skilling risk is particularly pronounced for juniors, (4) design judgment and handling uncertainty emerge as core competencies, (5) displacement effects are empirically confirmed in the freelance market.
-
The Democratization-Recommodification Paradox: How AI Turns 'Anyone Can Do It' into 'No One Can Charge for It'
2026-06-09
When AI democratizes exploration and production (expanding the base of participants), the relative scarcity that mid-tier practitioners relied upon as their basis for billing evaporates. Scarcity migrates to 'the handful at the top whom AI cannot replicate,' reinforcing rents (enclosure and symbolic capital) exclusively at the upper tier. Because democratization lowers barriers only at the bottom without disturbing the closure mechanisms at the top, inequality widens asymmetrically. This is the paradox in which democratization destroys the livelihoods of its own beneficiaries.
-
AI Commoditization Resilience of Design Billing Models — Output-Based Billing Is Most Vulnerable, and Retainer Stickiness Does Not Mean High Margins
2026-06-09
AI resilience of billing models ranks 'outcome-based > time/human-hour > output (deliverable)' (F and C diverge on relative vulnerability of human-hour vs. output billing). The moment a deliverable is defined as a thing, its price slides toward AI's marginal cost. Retainer arrangements are hard to replace, but that means 'hard to fire,' not 'billable at a premium' --- switching cost ≠ gross margin. Profitability requires vertical specialization with named referrals plus offloading operations to cheaper headcount/AI (i.e., headcount reduction as the funding source). The 'gatekeeper business' splits into upper tier (policy design = high-value engagements) and lower tier (inspection = floor pricing) and collapses.
-
The Role of Designers in the AI Era — Industry x Academia Roundtable (Business-Only Revision)
2026-06-07
A joint discussion record in which three industry personas — design firm / strategy consulting design lead / independent design office — argue exclusively from three perspectives: profitability, billability, and AI efficiency gains (zero academic references). Contrasts the industry's cost-benefit calculus with the academic's theoretical examination.