What is AI doing to designers, and what are designers doing to AI?
We surveyed 217 design practitioners across 43 countries to find out. They told us how AI is reshaping their work and their thinking. They also told us how they, in turn, are choosing what to call AI, when to follow it, and when to override it. This dashboard summarizes the patterns and the voices. No prior context required. Browse any tab in any order.
Three lenses.
Nine concepts.
One survey.
Most surveys ask one type of question about AI adoption: how often, how many tools, what for. We used three academic lenses to ask the deeper questions. How AI is changing what designers do, how they think, and what design means as a profession.
Single-lens studies miss what AI is actually doing to design.
"What can a tri-disciplinary scaffold see that one discipline alone can't?"
HCI alone captures interaction patterns but misses identity and meaning. Design Studies addresses practice transformation but lacks instruments for cross-sample comparison. STS provides critical, situated analysis but rarely operationalizes into measurable patterns. No existing study integrates all three. We built one, and made it bilingual to test whether the patterns hold across linguistic communities.
A note on method. This survey combined two ways of listening. The quantitative half asked structured questions about frequency, tools, role labels, and Likert scales, mapping the shape of AI integration across 217 practitioners. The qualitative half opened space for designers to answer in their own words: what their work has become, what AI feels like to work with, what researchers are missing. The structured questions tell us where designers sit. The open answers tell us what designers say. The most interesting findings live in the gap between the two.
From theoretical lenses to survey items.
"How does an Actor-Network theory concept become a survey question without losing what makes it valuable?"
Each lens contributes three literature-derived concepts, nine in total. Design Studies and HCI offered validated constructs adaptable to structured items. STS presented a challenge: Actor-Network Theory was designed for ethnographic study, not surveys. We constructed three custom items targeting perception (agency attribution), behavior (inscription response), and outcome (network reconfiguration). This perception-behavior-outcome triad approximates ethnographic richness within survey constraints.
A candid note on the STS items. Unlike the Design Studies and HCI concepts, which draw on validated scales (Bhargava's three-level model, CAILS/CAIMS), all three STS items are custom-built. No validated ANT survey instruments exist. We use STS as an interpretive lens for structured data, not as a full ethnographic ANT study. The three items are deliberately designed to capture perception, behavior, and outcome as partial compensation for the absence of psychometric validation.
Adoption is universal.
Language about it is not.
The survey returns one big finding: three lenses converge on the same gradient. Deeper AI integration changes practice, cognition, and meaning together. But how practitioners name what they do with AI lags behind what they actually do, a gap with consequences for governance, training, and professional identity.
Four numbers that frame the rest.
Every other finding is a sub-question of these four.
Nine measurements.
Three lenses.
One convergent finding.
"If you had to read AI's effect on design practice through one chart, what would it say?"
Each column below is a single empirical measurement from the survey. Three measurements per lens. Each cell shows how that measurement changes as practitioners move from Routine users (Level I) to Explorer users (Level II) to Integrated users (Level III). The pattern across all nine columns tells one story: integration depth predicts everything.
From production to governance.
"What are designers actually doing differently, and what does it cost them?"
Designers are spending less time making and more time evaluating, curating, and directing AI-generated outputs. The shift is not subtle: 71% of the full sample report more time evaluating than producing, and at Level III, that climbs to 91%, with 45% reporting it as the dominant pattern (more than 2× the rate at Level I).
The deeper the integration, the bigger the shift. Level III users are 2.5× as likely to report a major evaluation shift compared to Level I. The act of "designing" is being reorganized around AI's outputs.
The labeling gap. 60% call AI an "assistant", but 36% of Level III users name it a "collaborator," 4× the rate among Level I users. Practitioners upgrade the label only when behavior leaves no alternative.
AI adoption is not a line.
It's a landscape.
When you measure what designers do with AI separately from what they call it, distinct positions emerge. The assumed "destination" (the Aligned Governor) is only 10% of practitioners. The path through transformation isn't a single line. It's at least two routes shaped by cultural and linguistic context.
Six positions in the landscape.
"What happens to AI transformation if you stop assuming everyone is moving in the same direction?"
Each archetype combines a behavioral integration position (Design Studies × HCI: how integrated, what interaction style) with an agency attribution position (STS: tool / assistant vs. collaborator / unpredictable). The 2×2 below shows where each archetype lives. Cells with multiple cards have multiple distinct sub-positions.
Only 3 respondents fit the strict pattern (n=3 of 177, 1.7%), too small to characterize as an archetype, but worth noting: this is the rarest stance in the sample.
Key insight. The Sophisticated Instrumentalist has higher metacognition (3.27) and lower anxiety (2.62) than the Aligned Governor (2.72 / 3.17). The person who refuses to call AI a "collaborator" is, on average, the most competent and least worried on the team. The mismatch between what they do and what they say isn't a deficit to fix. It's a discriminating signal worth listening to.
Half the sample doesn't fit a clean archetype.
The five archetypes account for 89 of 177 respondents (50%). The remaining 88 (50%) sit in what we call the "mostly aligned middle", a position where behavior and language move in the same direction but neither extreme emerges. This is not a residual bucket. It's the largest cluster in the sample, and it has methodological consequences.
Designers whose three-axis position is internally consistent but doesn't reach the discriminating extremes that define an archetype. They do as much with AI as their integration level predicts. They describe it the way their behavior predicts. The story isn't with them. It's with the half who break the pattern.
Your dashboard probably shows AI frequency. That tells you almost nothing about archetype. The Sophisticated Instrumentalist and the Aligned Governor both use AI daily, but they describe what they do in opposite ways. Without measuring agency attribution, you can't see the difference.
If your most integrated team members refuse to call AI a "collaborator," that isn't them being behind the curve. They may be reading the situation more accurately than the framework allows. Treat it as data.
Half your team isn't an archetype. They're moving steadily across all three axes together. Transformation programs designed for the Aligned Governor will miss them. They don't need re-orientation. They need scaffolding for the path they're already on.
They do more with AI.
They just won't call it that.
Latin American designers, predominantly Colombian (n=82, 90% Colombian), outperform global peers on every behavioral measure of AI integration. But they decline to call what they do "collaboration." This behavioral-linguistic mismatch is the paper's most distinctive finding.
Latin America arrives at the same destination, just from further away.
"Why does the Latin American agency-attribution gap collapse with seniority?"
At every seniority level, Latin American designers attribute less agency to AI than their global peers, but the gap collapses with experience. From 40 points at junior level, to 16 points at mid-career, to 3 points at senior. This is not a fixed cultural position. It's a developmental trajectory. The pedagogical question for design education in the region: can we compress the time-to-arrival?
Developmental, not fixed. Senior Latin American designers reach near-parity (within 3 points) of global peers on naming AI a "collaborator" or "unpredictable participant." The story isn't that Latin Americans don't see AI as a collaborator. They reach that recognition through a different route, and they get there. The 40-point junior gap signals where the curriculum effort needs to land.
What they do with AI
vs. what they say AI is.
"What happens when the same group leads on every behavioral measure but trails on every linguistic one?"
Latin American mid-career designers (n=54) outperform their global peers (n=49) on every behavioral indicator. But they decline to call what they do "collaboration." This is the paper's central finding: behavioral integration without linguistic agency.
Behavioral integration without linguistic agency. Latin American mid-career designers do more with AI, use more tools, and follow its surprises more often, but they decline to name what they do as collaboration. The same behavioral reality is narrated through fundamentally different discursive registers. The most consequential finding of this study sits in the gap between a bar chart and a vocabulary.
The Spanish corpus surfaces concepts the English data has no equivalent for.
When Latin American practitioners describe their AI relationship, they reach for words that don't appear at comparable rates anywhere else. The vocabulary is a finding.
36% of Spanish speakers express employment concern (vs. 8% in English). ~12% invoke humanidad or esencia, naming an irreducible human core of design. The Spanish corpus contains the words "amor propio" (professional self-worth), "afectaciones cognitivas" (cognitive harm), and "tiempo lento" (slow time), none of which have direct English-corpus analogues.
The same survey.
Two vocabularies.
Two relationships.
Running the survey bilingually wasn't a translation exercise. It was a methodological choice. Each language corpus was coded independently before cross-language comparison, to prevent English-language frameworks from colonizing Spanish-language meanings. What emerged: two communities, naming the same AI relationship through fundamentally different discursive registers.
Same behavior. Different language.
"What does it mean when one community calls AI a 'partner' and the other calls it a 'tool that helps'?"
English-language respondents reach for relational vocabulary, partner, companion, collaborator, thought partner, at five times the rate Spanish-language respondents do. Spanish speakers describe the same behavioral reality but stay closer to instrumental vocabulary, herramienta, asistente, dupla, even when the underlying interaction is collaborative. Whether English vocabulary describes or constitutes the relationship is exactly the question this study leaves open.
~35% of English speakers use relational vocabulary ("partner," "companion," "collaborator," "thought partner") in open-ended responses.
~6.5% of Spanish speakers use equivalent relational vocabulary at comparable usage levels, five times less than English. 36% express employment concern (vs. 8% in English). The Spanish corpus carries an emotional register the English corpus does not.
"Assistant" means different things in different places.
"Why does the same word, Assistant, show up across all regions while meaning something completely different in each?"
The bilingual coding revealed that calling AI an "assistant" is the dominant frame across all regions, but the mechanism behind that framing varies by region. The qualitative data reveals at least three distinct logics behind the same label. This matters for transformation strategy: training programs designed for one form of instrumentalism will not work for the others.
Clients use AI to bypass designers, arriving with AI-generated mockups and asking the designer only to "polish" them. Calling AI a "collaborator" would misrepresent who actually holds power. The instrumental framing is accurate to the economic reality, not a misdescription of it.
Senior designers see AI commodifying decades of accumulated expertise. Naming AI as "assistant" is a jurisdictional claim, a way of preserving professional territory. "I remain in control" is what the language is doing, not what the work is doing.
AI is infrastructure for speed. No identity threat, no displacement pressure, just workflow optimization. The instrumental framing reflects a relationship that is genuinely instrumental: the practitioner is unbothered, the tool is useful, the work continues.
The implication. Training programs that work for pragmatic European instrumentalists will fail for structurally displaced Latin American designers. Transformation strategies designed for defensive North American seniors will be irrelevant to Latin American juniors who are already more integrated than their Anglo counterparts. The behavioral-linguistic mismatch is not one phenomenon. It is at least three, each requiring a different response.
Concepts the Spanish corpus surfaces with no English equivalent.
Three terms that appeared repeatedly in Spanish responses, with no direct conceptual analogue in the English corpus. These aren't translation failures. They are findings about what each language community can articulate.
An irreducible human core of design that AI cannot replicate. ~12% of Spanish responses invoke it. The English corpus contains no comparable concept, practitioners who write in English do not name what AI cannot reach.
Why it matters: Suggests Spanish-language design discourse maintains an essentialist frame for human design value that English-language discourse has abandoned.
Professional self-worth. Distinct from "self-esteem" or "professional pride." Names a specific erosion: not job loss, but the dissolution of belief in one's own work. Surfaces only in Spanish, anomaly ES-56.
Why it matters: The English-language discourse frames the AI threat as displacement (jobs). The Spanish-language discourse adds a second register: devaluation (self-worth).
Slow time. Names the cognitive incubation that AI's acceleration removes. "Ese tiempo lento hacía parte esencial del proceso", that slow time was an essential part of the process. Anomaly ES-3.
Why it matters: The closest English-corpus concept is "thinking-time bottleneck" (anomaly EN-87), but it lacks the temporal-aesthetic dimension tiempo lento carries.
We asked. This is what they said.
"Is there something about how AI is affecting your work that researchers aren't paying enough attention to?"
The open-ended question that produced the most distinctive corpus across both languages. Three signals dominate.
Practitioners report fatigue from constant evaluation, a kind of governance overhead that didn't exist before. "I think more, but I create less." Researchers measure adoption; respondents are asking us to measure the cost of evaluation.
Multiple respondents flag that AI use is no longer optional, companies, clients, and tools assume integration. "La uso por orden directa de mis jefes." Mandate, not adoption, is the right frame for a portion of the workforce.
Multiple senior practitioners worry that AI is eating the work that produces designers, early-career skill-building tasks now automated. The pipeline question: where will senior designers come from in 2035?
Where the patterns break.
Anomaly analysis was a third analytical step alongside deductive and inductive coding, surfacing voices that contradicted the convergent finding, opened new conceptual territory, or named experiences the framework didn't anticipate. 29 anomalies (16 EN + 13 ES) sit here as both data and provocation.
Four ways a voice can sit outside the patterns.
Each anomaly carries one or more tags. Multi-tagged anomalies are the most generative. They break the pattern in more than one direction.
Directly inverts a finding from the convergent dataset, e.g., AI causes burnout instead of relieving it.
Names a phenomenon no survey item or framework concept anticipated, e.g., environmental guilt.
Voices a position held by very few respondents, singular but consequential, e.g., AI as management weapon.
Articulates a phenomenon visible in the structured data, but with conceptual depth the survey couldn't capture, e.g., "tiempo lento."
Filter, search, expand.
Filter by language and tag, or search by text. Click any card to expand the full quote, respondent metadata, and why-it-matters analysis.
Not the headline. The seams.
These 29 voices are not the headline. The headline lives in the other tabs, the convergence, the archetypes, the bilingual gap. This tab holds the part of the data that resisted categorization: practitioners who said something the framework couldn't absorb, or said it in a way that broke our coding. We've kept them visible, in their own words, because aggregating them away would have been the more comfortable choice, and the wrong one.
Read individually, no single anomaly proves anything. Read together, they do something more interesting: they mark where the next questions live. The Spanish-language concept of humanidad as something AI threatens. The senior practitioner who says AI shifted design "from artifact creation to decision architecture." The structural displacement that turns "collaborator" into a misnomer because clients arrive with AI-generated mockups and ask the designer only to polish them. These are not patterns we found. They are seams in the framework, places where another study could begin.
They are also small. n=29 of 217. Early signals, not findings. Present today.
These voices will anchor Stage 2 of this research, provotyping and focus groups designed to test whether what reads here as anomaly today reads as signal tomorrow. We invite the design and research community to take these seams seriously, as starting points for your own work, prompts for your own teams, and inputs into the conversations we should all be having about what AI is doing to design practice and how designers, in turn, are reshaping AI.
If you read nothing else,
read this.
A synthesis of what the survey says, taken together. One paragraph that connects the dots across the previous tabs. Five short cards, one for each kind of person who might be reading this. And one closing question to think with.
What the study is actually saying.
If you've read the previous tabs in any order, you've encountered a convergence, a paradox, a landscape, a vocabulary gap, and 29 dissenting voices. Here is the through-line.
AI is moving design from making to governing, and that transition is real, measurable, and consistent across three disciplinary lenses. But how designers describe what they're doing lags behind what they're doing, and the gap is patterned: the most integrated practitioners are often the most reluctant to call AI a "collaborator." This is not a maturity deficit. It's a discriminating professional stance, one that shows up most sharply in Latin America, surfaces concepts the English-speaking world has no words for, and breaks in 29 specific places where the framework couldn't follow. The gap between what designers do and what they say AI is is where the next decade of design practice will be negotiated.
Five audiences. One thing each.
Not a list of conclusions, a list of redirects. Find the card that fits you. One observation worth keeping, and one provocation worth sitting with.
If you use AI daily, follow its surprises, and still call it a "tool", you're in the largest behaviorally-integrated archetype in our sample. You're not behind. You're navigating something the vocabulary hasn't caught up to.
Pay attention next time you describe your AI workflow. The word you reach for tells you something about the relationship you're willing to claim, and the responsibility you're willing to share.
Your dashboard probably tracks AI frequency. Half your team shares the same frequency level but lives in opposite archetypes. The Sophisticated Instrumentalist and the Aligned Governor look identical in usage data, and require completely different conversations.
If you ran the three-axis survey on your own team, where would they cluster? And what would the silent-navigator majority need that no transformation program is currently giving them?
The behavioral-linguistic mismatch is invisible to any single lens. The Spanish-only concepts (humanidad, amor propio, tiempo lento) are invisible to any English-only corpus. The bilingual plus tri-lens architecture isn't a methodological flourish. It's the condition for seeing what's actually happening.
Which of your current findings would survive a structured cross-language re-analysis, and which are artifacts of working in one language at a time?
The agency-attribution gap collapses with seniority (40 → 16 → 3 points). This is a pedagogical opportunity, not a cultural deficit. Latin American senior designers reach near-parity on naming AI as a collaborator. The question is what closes that gap, and whether curriculum can compress it.
If your students are graduating into a profession where directing AI is more of the job than making things, what should the studio look like? What's the first assignment that doesn't assume making as the primary act?
71% of designers spend more time evaluating AI outputs than producing original work. The dominant failure mode in our open-text data isn't bad output. It's governance overhead: the cognitive cost of deciding what to keep. Your tool's productivity gains may be invisible to users carrying that load.
What would it look like to design for the cost of evaluation, not just the speed of generation? What evaluation rituals could your tool embed, and which should it stay out of the way of?
If you remember one thing, let it be this.
When the most competent practitioner on your team refuses to call AI a "collaborator," what are they protecting, and is the framework asking them to give it up?
The survey.
The authors.
The caveats.
Everything you need to evaluate, replicate, or critique this study. Survey questions mapped to lenses and concepts. Methodological commitments and known limitations. Authors, reference, acknowledgments, and the data integrity firewall that separates this study's real respondents from any synthetic or analytical proxies.
23 questions. Nine derived from the concept framework.
The full instrument comprised 23 items: a demographic block, two AI-behavior items (frequency & tools), nine concept-derived items mapped to the three lenses, four open-ended items for experiential depth, and a follow-up item for Stage 2 recruitment. Open-ended prompts functioned as a falsifiability mechanism, allowing emergence beyond the framework. Deployed bilingually via LinkedIn (~10,000 connections), 217 completions over a two-week field period.
| Block | Lens | Concept · Item |
|---|---|---|
| Demographics | Context | Country · gender · education · profession · industry · work context · primary work language · design area · years of experience |
| Behavior | AI usage | AI frequency: Daily / Several times a week / Few times a month / Rarely / Never · AI tools used: Conversational / Image generation / Code / Workflow / Custom-enterprise / Embedded |
| Concept 1 | Design Studies | Three-level usage model: Routine / Explorer / Integrated (Bhargava & Gopal, 2022) |
| Concept 2 | Design Studies | Production-to-evaluation shift: 5-point evaluation/creation balance (Simkute et al., 2024) |
| Concept 3 | Design Studies | Values reconfiguration: Likert: AI is changing what counts as good design |
| Concept 4 | HCI | Interaction style: Directive / Iterative / Thinking partner / No consistent approach (CAILS scale) |
| Concept 5 | HCI | Metacognitive awareness: Likert: I can predict the quality of AI output (CAIMS) |
| Concept 6 | HCI | Trust-autonomy balance: Likert items: skills anxiety, task division, director-curator identity |
| Concept 7 | STS | Agency attribution: AI as Tool / Assistant / Collaborator / Unpredictable participant (custom, Latour-inspired) |
| Concept 8 | STS | Inscription response: When AI surprises you: modify / explore / reject / use as starting point (custom, Akrich-inspired) |
| Concept 9 | STS | Network reconfiguration: Has work been divided / discussed / valued differently? (custom, Law & Varanasi 2025) |
| Open | Qualitative | Practice change · AI relationship · Meaning of design · What researchers are missing · Stage 2 follow-up |
What this study can claim, and can't.
Five commitments and limitations that shape every finding in this dashboard.
Bilingual coding protocol
Each language corpus was coded independently before cross-language comparison. No translation occurred prior to coding. This prevents English-language frameworks from colonizing Spanish-language meanings, and is itself part of the substantive finding.
Seniority-geography confound
Latin American respondents skew younger than Anglo-Western respondents. Raw geographic comparisons are unreliable. All cross-region comparisons in this dashboard are seniority-controlled (Junior / Mid-career / Senior).
STS as interpretive lens
No validated ANT survey instruments exist. Our three STS items are custom-built, capturing perception, behavior, and outcome as partial compensation for the absence of psychometric validation. We use STS interpretively, not ethnographically.
Recursive entanglement
AI (Claude, Anthropic) was used to assist in coding survey data about AI's impact on design practice. This recursive entanglement is acknowledged as a methodological limitation. Authors maintained final coding decisions; AI assisted with pattern surfacing.
Latin American composition
Latin America (n=82) is 90% Colombian (n=74); the remaining 8 respondents come from six other countries. The non-Colombian subsample is too small to test independently with confidence, but it broadly echoes the Colombian sample's discursive patterns, naming AI as "assistant" rather than "collaborator," lower agency attribution than Rest of World, suggesting the linguistic finding extends beyond a single national sample even where behavioral measures are underpowered to confirm it.
Field period & distribution
Survey fielded between February 17 and March 2, 2026, distributed entirely through LinkedIn, across the authors' professional networks (~10,000 connections) and substantially amplified by a repost from Jakob Nielsen that extended reach beyond the original network. Median completion time: ~10 minutes. Voluntary, anonymous, no incentive offered.
🔥 Data integrity firewall. The empirical findings shown across this dashboard come exclusively from 217 real survey respondents across 43 countries. Synthetic personas (n=75) generated for instrument testing are blocked from the analytical dataset. Exploratory analyses developed for a planned cross-sector organizational survey (Stage 2) are not represented here, they are kept methodologically separate. Every percentage in this dashboard traces back to survey_clean.json.
Who answered the survey.
A snapshot of the 217 design practitioners whose responses underlie every chart in this dashboard. Latin America (n=82) is the largest regional cluster and is predominantly Colombian (n=74), surfaced openly here so readers can weight every regional comparison appropriately.
Credits and how to refer to this work.
Authors
Dr. Jaime Rivera
Universidad Nacional de Colombia / IIT Institute of Design
PhD Design · UX research & strategy · Designer-researcher entanglement
Marianna Russi MDes(c)
Universidad Nacional de Colombia
Theoretical framework · Design Studies lens · Stage 2 thesis on design judgment (criterio)
Reference
Rivera, J. & Russi, M. (2026). AI and the situated emerging professional in design practice: An exploratory study through three disciplinary lenses. Base Diseño e Innovación, 10(13).
Status: Academic paper under review (second round, pending approval). Special issue on Design and AI · Universidad del Desarrollo · NC State University · UAM Azcapotzalco.
Quantitative + qualitative · Tri-lens · n=217 · 43 countries · Bilingual
Provocative prototypes surfacing assumptions about design judgment (criterio), the candidate concept emergent from Stage 1 anomalies.
Building on Stage 2 findings, situated dialogue across language communities and seniority levels.
This study exists because they trusted that the questions were worth answering.
Special thanks
Special thanks to Jakob Nielsen, whose generous repost amplified the survey to his global network and substantially expanded our reach beyond what we could have achieved alone. We're grateful for his support and for the broader Nielsen Norman Group community that engaged with this work.
To the respondents and the network
We thank the 217 designers across 43 countries who took 10 minutes from their working week to think publicly with us. We also thank the colleagues, peers, and strangers on LinkedIn who shared the survey with their networks. This study exists because they trusted that the questions were worth answering.