Design Systems in the AI Era: Competency vs. Craft  🔗 **Source:** [Design Systems in the AI Era: Competency vs. Craft](https://www.youtube.com/watch?v=AbKSa_U-3XI#transcript) > [!note]+ Document Notes > > [!info] Generated analysis — not the author's words. Sections below are machine-written > > from the captured source. Only text inside `> [!quote]` blocks is verbatim from the author. > > Do not quote or attribute any other line to the named author. > > --- > tags: [design-systems, ai-automation, ux-research, wordpress, lean-ux] > author: Michelle Hunt (Lone Rock Point) > source: WordCamp US 2026 (YouTube) > reference: https://www.youtube.com/watch?v=AbKSa_U-3XI > date created: 2026-08-19 > > --- > > # 🏷️ Title: Design Systems in the AI Era: Competency vs. Craft (WordCamp US 2026) > <!-- deep-read:sections start --> ## 🧭 Table of Contents - [📝 Summary](#-summary) - [📖 Main Content](#-main-content) - [🗒️ Notes](#️-notes) - [📚 References](#-references) - [🗣️ Paraphrase](#️-paraphrase) - [📰 Newsletter](#-newsletter) - [✅ Action Items](#-action-items) - [🧪 Hypotheses](#-hypotheses) - [📊 Methodology](#-methodology) - [❓ Questions](#-questions) - [🔮 Future Research](#-future-research) - [💡 Implications](#-implications) - [🧠 Learning Styles Assessment](#-learning-styles-assessment) - [📊 Accessibility Assessment](#-accessibility-assessment) - [🌍 DEIB Assessment](#-deib-assessment) - [✅ Logical Validity](#-logical-validity) - [🛡️ DARVO Analysis](#-darvo-analysis) - [🎯 Conclusion](#-conclusion) ### 🔗 References - [Nielsen Norman Group — Design Systems 101](https://www.nngroup.com/articles/design-systems-101/) - [Kano model (referenced as "Kono model" in the talk)](https://en.wikipedia.org/wiki/Kano_model) - Michelle Hunt's other talks on Lean UX (referenced but not linked in-talk) ## 📝 Summary - **What happened**: At WordCamp US 2026's AI track, Michelle Hunt (director of UX engineering, Lone Rock Point) argued that design systems have spent roughly a decade optimizing "competency markers" — consistency, accessibility, performance, reliability — all of which AI can now measure and assist with. She warned that AI's strength at raising these measurable floors creates a converging risk: as everyone uses AI to hit the same competency bar, output becomes generically "fine but not good." She proposed a second axis, "differentiators" (personality, craft, empathy, respect), which are judged rather than measured and require human taste, and closed with a Lean UX-style continuous-feedback-loop framework (assume/build/measure/learn/adjust) plus a call for human stewardship and accountability as the maintained, ongoing part of a design system that most orgs skip. - **Why it matters**: This reframes the current "AI will commoditize design systems" anxiety into a structural argument — AI raises floors (competency) but cannot by itself raise ceilings (differentiation), because differentiation requires judgment, not measurement. For WordPress specifically, this bears directly on Core's theme.json, block patterns, Playground, and the growing set of AI-assisted authoring features (guidelines, inline suggestions) — all floor-lifting tools that risk homogenizing WordPress sites unless paired with human-owned craft decisions. - **Key stakeholders**: UX/design system teams at agencies and product orgs; WordPress Core Design and Core AI contributors building theme.json/pattern tooling; AI/design tool vendors (Figma, Storybook, "Claude design," code-gen tools); end users experiencing the resulting homogenization; internal content creators/maintainers who consume design systems as an "internal UX." - **Risks and tradeoffs**: The talk is explicit that "fine" is not always bad — commodity experiences (a one-page menu site) don't need differentiation. The risk is applying floor-lifting AI tooling uncritically to products that do need to stand out, and mistaking AI-assisted competency work for the entirety of design system maturity. A second risk, which the talk raises but doesn't fully resolve, is that "judgment" and "stewardship" are asserted as irreducibly human without much argument for why AI evaluation (which she acknowledges exists) can't approximate judgment given enough training signal — she treats this as axiomatic rather than proving it. - **Bottom line**: Design systems need both comprehensive outputs (tokens, components, guides — increasingly AI-assisted) and thoughtful, evidence-based human inputs (stewardship, ownership, continuous research) or they become "just another artifact" that goes obsolete. AI can be a builder; it should not be the decision-maker. ## 📖 Main Content - **Design system definition**: Hunt opens with the Nielsen Norman Group definition — "a set of standards to manage design at scale by reducing redundancy while creating a shared language and visual consistency across different pages and channels" — and uses it as the throughline for the whole talk, returning to the phrase "a set of standards" later to ask *who* sets and evolves those standards. - **Why design systems exist**: For end users, pre-standards experiences suffered from inconsistency, unpredictability (→ lack of trust), and cognitive load (→ disengagement). For content creators/internal users, the problems were redundancy, decision fatigue, and "blank canvas syndrome." Design systems address both audiences — Hunt flags herself as "a big proponent of internal user content creator user experience," treating the internal authoring experience as equally important to the end-user one. - **Experience models she reviews**: (1) a "ceiling/floor" satisfaction range framed against a shifting "expectations" baseline (above = positive, below = negative); (2) the Kano model ("Kono" in the transcript) — must-haves, linear satisfiers/performers, delighters/exciters, and indifferent features, with the key insight that delighters migrate to must-haves over time (cruise control, unlimited talk-and-text); (3) the user journey map, plotting satisfaction across touchpoints to find opportunity gaps. - **Downward forces**: Entropy (lack of decision-making) and reactivity (short-sighted, in-the-moment decisions) erode experience quality over time if nothing actively counters them — this sets up the need for the feedback-loop framework later. - **Competency markers (the floor)**: Consistency, accessibility, performance, reliability. For each, Hunt gives (a) the concept-level tools (tokens, components, theme.json, patterns/blocks, style guides for consistency; semantic structure, contrast, writing level for accessibility; optimized/lazy-loaded/cached code for performance; uptime, error handling, testing/QA for reliability), (b) the current tooling (Figma, Storybook, accessibility checkers, CDNs, core web vitals monitors, unit/UAT testing), and (c) how AI is already assisting each one (AI-generated tokens/patterns/variants and code-gen for consistency; automated accessibility/language-level audits and plain-language remediation explanations for accessibility; auto-optimization and performance audits for performance; AI code review, spec/test generation, automated bug fixing for reliability). - **The core problem with competency-only work**: It only lifts the floor (worst experiences → meeting expectations); the ceiling stays capped by the same short-sighted decision-making. Compounding this, expectations aren't static — "yesterday's delight becomes today's necessity" — so standing still is actually falling behind. - **On AI "slop"**: Hunt distinguishes careless slop (broken, hallucinating, inaccurate — the current failure mode, which she believes is solvable) from a deeper, more permanent risk: even maximally competent, accurate AI output tends toward "generic and hollow and sterile... soulless," converging everyone toward "everything's fine but nothing's good." She's careful to validate that "fine" is legitimate for commodity products (a one-page restaurant site) but argues most of her audience is building things that need to be more than fine. - **Differentiators (the ceiling)**: Personality (voice/tone, humor, warmth, authenticity — evaluated via brand/voice work and social engagement), craft (edge-case handling, empty/error/loading states, micro-interactions, multisensory design quality — evaluated via design critique and qualitative feedback), empathy (contextual sensitivity, anticipating needs, fitting mental models, inclusion — evaluated via user interviews, journey maps, jobs-to-be-done, support tickets), and respect (attention/time value, honesty, consent, autonomy, access — evaluated via dark-pattern audits and trust surveys). The unifying claim: "differentiation is judged not measured," where competency metrics are binary pass/fail and differentiation requires "taste and experience and opinions" — i.e., shared humanity. - **Stewardship and ownership as the missing piece**: Returning to "a design system is a set of standards," Hunt argues most orgs only build the "design repository" half of NN Group's two-part definition (repository + the people managing it) and neglect the human stewardship/ownership half. Her "spicy take": without the people-management half, "you are not building a design system at all... just building another artifact that's going to become obsolete." A system needs both comprehensive outputs (artifacts) and thoughtful inputs (evidence-based updates) to stay relevant, and — her load-bearing claim — "a machine can produce things and a machine can evaluate things, but it can't be responsible." - **The Lean UX-style feedback loop**: An adapted assume → build → measure → learn → adjust cycle (explicitly framed as "a lot like the scientific method"), run continuously through the project and into maintenance, not just at discovery. Assume = hypothesis in the format "we believe [X outcome] will happen for [persona] because we built [feature]... we will know this is true if [measurable outcome]." Build = the smallest testable artifact (prototype, survey, interview, new component/page). Measure = gather the data tied to the "we will know this is true if" clause. Learn = interpret whether/how fast the hypothesis played out and surface unexpected findings. Adjust = revise or generate new hypotheses; the loop repeats indefinitely. - **Operationalizing the loop at her own org**: (1) Ongoing research — a standing baseline of user/business understanding (vs. one-off "reactive research" per project) combined with Lean UX's just-in-time targeted research, likened to a live traffic map vs. a static map you dig out only when lost. (2) Collaborative alignment — recurring client conversations that revisit whether hypotheses still hold, explicitly setting expectations up front that the relationship isn't "deliver and disappear." (3) Measurement infrastructure — quantitative (analytics, A/B tests) and qualitative (direct user feedback) metrics feeding early-stage scorecards/dashboards (she flags this as "alpha," still in progress at Lone Rock Point). (4) Applying the same Lean UX loop internally to the team's own process, not just the client's product. (5) Human accountability — closing point: humans remain the owners/stewards of the system and its decisions; judgment and empathy are the differentiators, tied back to "shared humanity." - **Q&A**: An audience member (agency UX practitioner) asked how to counter AI-driven homogenization. Hunt's answer: AI is "a builder... not a decision maker" — if everything looks the same, it's because teams are letting AI make decisions it shouldn't. She acknowledged the tradeoff explicitly: doing the harder creative work costs more time and money than letting AI produce "fine" output fast, but framed the choice as "we're not competing to be the cheapest... we're trying to be good." - **Uncertainties/contested claims not resolved in the talk**: Hunt doesn't define a measurable boundary between "competency" and "differentiator" work — some items (e.g., accessibility's "writing level/style") arguably straddle both measurable and judged territory. She also doesn't address whether AI evaluation tooling (which she says already assists accessibility/reliability judgment) could, given more data, start approximating "taste" — she treats non-measurability as a stable property rather than testing it. ### 🗒️ Notes #### ⏱️ Opening / framing - Self-intro: Michelle Hunt, director of UX engineering, Lone Rock Point; ~13 years speaking on design/design systems; explicitly non-technical, concept-focused talk in the AI track. - Anchors the whole talk on the NN Group definition: "a set of standards to manage design at scale by reducing redundancy while creating a shared language and visual consistency across different pages and channels." #### ⏱️ Why standards exist - End-user costs of no standards: inconsistency, unpredictability/lack of trust, cognitive load/disengagement. - Internal-user costs: redundancy, decision fatigue, blank canvas syndrome. - Explicit stance: design systems improve UX for both end users and internal/content-creator users. #### ⏱️ Satisfaction models - Ceiling/floor/expectations-baseline model; above expectations = positive, below = negative. - Kano ("Kono") model: must-haves, linear satisfiers, delighters/exciters, indifferent features; delighters migrate to must-haves over time (cruise control, unlimited talk/text examples). - User journey map: touchpoint-by-touchpoint satisfaction, used to find "opportunity gaps." - Downward forces: entropy and reactivity erode experience over time absent active counter-effort. #### ⏱️ Competency markers (floor-lifting, AI-assisted) - Four markers: consistency, accessibility, performance, reliability — each paired with concepts, current tooling, and specific AI/automation assistance (tokens/pattern generation, accessibility/language-level audits with plain-language remediation, performance auto-optimization, AI code review/test generation/bug fixing). - Core limitation: competency work only lifts the floor; the ceiling remains capped by the same short-sighted decision-making. Expectations rise over time, so a static floor eventually becomes a new deficiency. #### ⏱️ The "slop" argument - Distinguishes careless/current-state slop (broken, hallucinating) — solvable — from a deeper convergence risk: even highly competent AI output trends generic/hollow/sterile, producing a "fine but not good" baseline. - Validates that "fine" is acceptable for commodity products; argues most of the room's work needs to be more than fine. #### ⏱️ Differentiators (ceiling-raising, judged not measured) - Personality, craft, empathy, respect — each defined with concrete evaluation tools (brand/voice work, design critique, user interviews/JTBD, dark-pattern audits/trust surveys). - Central distinction: competency = measurable (pass/fail metrics); differentiation = judged (taste, experience, opinion — "connects with our shared humanity"). #### ⏱️ Stewardship, ownership, and the Lean UX loop - Revisits "set of standards" to ask who sets/evolves them; argues most orgs neglect the "people managing it" half of NN Group's repository+people framing. - "Spicy take": a repository without stewardship isn't a design system, just an artifact headed for obsolescence. - Load-bearing claim: machines can produce and evaluate but "can't be responsible" — humans are owners/stewards. - Lean UX loop: assume (hypothesis) → build (smallest testable thing) → measure (data against the hypothesis's success criteria) → learn (interpret) → adjust (revise/re-hypothesize), run continuously, not just at discovery. - Operational mechanisms at Lone Rock Point: ongoing (vs. purely reactive) research; collaborative alignment check-ins with clients; quantitative+qualitative measurement/scorecards (self-described as early/alpha); applying the loop to internal process, not just client deliverables; human accountability as the closing principle. #### ⏱️ Q&A - Question: how to counter AI-driven homogenization at an agency. - Answer: AI is a builder, not a decision-maker; homogenization results from ceding decisions to AI; doing differentiated work costs more time/money but "we're trying to be good," not cheapest. ## 📚 References - [Nielsen Norman Group — Design Systems 101](https://www.nngroup.com/articles/design-systems-101/) - [Kano model overview](https://en.wikipedia.org/wiki/Kano_model) - [Lone Rock Point](https://lonerockpoint.com/) — Michelle Hunt's employer, referenced as the source of the Lean UX-adapted framework - Note: Hunt references "some other talks about lean UX" and closing-slide resources without naming them explicitly on the transcript audio — not independently verifiable from this source alone. ## 🗣️ Paraphrase - Design systems exist to fix the inconsistency, unpredictability, and cognitive load users face without standards — and the redundancy, decision fatigue, and blank-canvas paralysis creators face without them. - Several established models (ceiling/floor vs. expectations, Kano, user journey maps) describe how experience quality is perceived, and all point to the same idea: expectations aren't fixed, they rise, and today's delighter becomes tomorrow's baseline requirement. - Design system work has spent a decade on four "competency markers" — consistency, accessibility, performance, reliability — all of which are measurable and all of which AI now meaningfully assists with, from token generation to automated accessibility audits to AI-driven code review. - The catch: competency work only raises the floor. It doesn't touch the ceiling, which stays limited by the same short-sighted decisions as before — and because expectations keep climbing, standing still on competency alone is a losing position. - Even setting aside today's sloppy, error-prone AI output, there's a subtler long-term risk: maximally competent AI output still tends toward generic, hollow sameness, converging everyone toward "fine but not good." That's fine for genuinely commodity products, but not for anything meant to stand out. - What raises the ceiling instead are "differentiators" — personality, craft, empathy, respect — which can't be reduced to a pass/fail metric; they require human judgment and taste. - Going back to the definition of a design system as "a set of standards," the real gap in most orgs isn't the artifact repository (tokens, components, guides) — it's the neglected other half: who owns and stewards those standards over time. A system without active human stewardship isn't really a design system; it's an artifact heading toward obsolescence. - The proposed fix is a continuous Lean UX-style feedback loop (assume/build/measure/learn/adjust) run throughout a project's life, backed by ongoing (not just reactive) research, recurring client alignment conversations, a real measurement practice, and — non-negotiably — humans remaining the accountable owners and stewards, because machines can produce and evaluate but cannot be responsible. ## 📰 Newsletter Michelle Hunt's WordCamp US 2026 talk lands a useful reframe for anyone nervous about AI flattening WordPress design: AI is very good at lifting the floor (theme.json consistency, accessibility checks, Core Web Vitals, automated QA) and not equipped to raise the ceiling on its own. Her four-and-four framework — competency markers (consistency, accessibility, performance, reliability = measured) versus differentiators (personality, craft, empathy, respect = judged) — maps cleanly onto where WordPress Core AI and Core Performance initiatives are already investing, and doubles as a warning label for anyone tempted to let an agent make the taste calls. Her closing point is the one worth pinning to a Trello card: "a machine can produce things and a machine can evaluate things, but it can't be responsible" — stewardship of a design system, patterns library, or block theme is still a human job. If your team is leaning on AI-assisted pattern generation or Playground-driven prototyping, pair it with an actual continuous-feedback loop (her Lean UX assume/build/measure/learn/adjust cycle) instead of a one-time discovery sprint, or you'll just ship "fine" faster. ## ✅ Action Items #### DevRel - Developer advocacy: build a talk/workshop pairing Hunt's competency-vs-differentiator framework with WordPress Core AI's roadmap, with a measurable takeaway (e.g., "audit your patterns library against both axes"). - Education: workshop on distinguishing measurable design-system QA (contrast checkers, performance budgets) from judged design-system QA (critique sessions, dark-pattern audits). - Developer experience: add a "competency vs. differentiator" self-check to design-system contribution docs/templates. - Developer marketing: content series on AI-assisted consistency tooling (theme.json, patterns) framed honestly as floor-lifting, not craft-replacing. - Community management: survey WordPress design-system maintainers on how much of their process is already AI-assisted vs. human-judged. - Internal/external/cross-project: align Core Design, Core AI, and Core Performance teams on shared language for "measured" vs. "judged" quality gates. #### OSPO - Policy & Governance: define who "owns" (per Hunt's stewardship framing) the WordPress design system's evolving standards — not just who maintains the repo. - License Compliance: n/a — not addressed in source. - Security Management: n/a — not addressed in source. - Software Inventory: track which design-token/pattern generation tools are AI-assisted for provenance/attribution clarity. - Developer Support: guidance on when to trust AI-generated accessibility/performance audits vs. when human review is required. - Community Engagement: fund a working group applying Hunt's continuous-feedback-loop model to WordPress.org's own pattern directory. - Strategic Alignment: adopt "lifting the floor vs. raising the ceiling" as a shared OKR framing for AI tooling investment. #### Community - Maintainers: add a stewardship/ownership checklist to design-system release hygiene (who reviews AI-suggested tokens/patterns before merge). - Contributors: create "first issue" rotations specifically for differentiator-side work (empty states, error states, micro-interactions) versus competency-side AI-assisted tickets. - Sponsors: fund the "ongoing research" practice Hunt describes (standing user-understanding baseline) rather than one-off sponsored research sprints. - Ecosystem: coordinate cross-plugin/theme audits of accessibility and performance competency markers using shared automated tooling. #### WordPress - Core: weigh AI-assisted consistency features (guidelines, inline suggestions) explicitly against the risk of homogenized site output flagged in this talk. - Core Performance: continue AI-assisted performance auditing/auto-optimization; document it as competency-marker work, not a differentiator. - Core AI: build evaluation sets that test for "generic/hollow" output convergence, not just correctness — directly responsive to Hunt's "future-state slop" argument. - **Core Program Team:** facilitate a cross-team retro on where AI tooling is lifting floors vs. where it's being asked (wrongly) to make ceiling/judgment calls. - Mobile: audit app design-system components for competency (accessibility, performance) vs. differentiator (micro-interaction craft) gaps. - Playground: use Playground demos to let contributors *feel* the difference between AI-generated "fine" scaffolding and hand-crafted differentiator work. - Community: build a WordCamp session track pairing this talk with hands-on design-critique practice (a "craft" tool Hunt names explicitly). - Training: develop a learning path on Lean UX-style continuous feedback loops for WordPress product teams. - Marketing: avoid overselling AI design tools as replacing craft/judgment; align messaging with "builder, not decision-maker." - Support: mine support tickets for empathy/craft gaps per Hunt's evaluation method (she names support tickets as an empathy-evaluation tool). - Design: adopt the competency/differentiator split as an explicit design-system documentation section. - Documentation: document which parts of the design system are AI-generated (tokens, patterns) vs. human-curated (voice, critique standards). - Photos: n/a — not addressed in source. - TV: n/a — not addressed in source. - Polyglots: n/a — not addressed in source. - Accessibility: continue expanding automated accessibility checker + AI language-level analysis, explicitly framed by Hunt as a competency-marker success story. - Meta: track AI-assisted contributor tooling as competency infrastructure requiring human stewardship sign-off. - CLI: n/a — not addressed in source. - Hosting: n/a — not addressed in source. - Tide: n/a — not addressed in source. - Openverse: n/a — not addressed in source. - Themes: apply the differentiator framework to theme review criteria (craft/personality), not just competency (performance/accessibility) checks already in place. - Plugins: n/a — not addressed in source. #### The WPOCC - Programs: fund fellowships specifically on design-system stewardship practice (the "people managing it" half Hunt says is usually neglected). - Projects: pilot a continuous-feedback-loop (assume/build/measure/learn/adjust) governance clinic for WordPress product teams. - Funding experiments: n/a — not addressed in source. ### FAIR - Technical independence: n/a — not addressed in source. - Policies and covenants: the stewardship/ownership argument ("a machine can't be responsible") is directly relevant to FAIR's human-accountability principles for AI-assisted governance artifacts — worth citing. - Productized governance artifacts: consider Hunt's "set of standards — who sets them, who evolves them" framing as a template question for FAIR charters. ## 🧪 Hypotheses 1. Design systems that rely primarily on AI-assisted competency tooling (consistency/accessibility/performance/reliability automation) without a parallel human-judgment practice will show measurably declining differentiation (brand distinctiveness, user-perceived "delight") over time even as competency metrics improve. 2. Organizations with an explicit stewardship/ownership role for their design system (a named accountable human or team) will show higher design-system longevity (lower "obsolete and rebuilt" rate) than those without one. 3. Teams practicing continuous (vs. one-time discovery-phase) UX research will detect "delighter → must-have" feature drift earlier than teams doing only reactive research. 4. AI-assisted "craft" evaluation (e.g., automated micro-interaction or empty-state quality scoring) is currently less reliable than AI-assisted competency evaluation (e.g., automated accessibility contrast checking) due to the judged-vs-measured distinction Hunt draws. 5. Users can distinguish between AI-heavy "competent but generic" digital products and human-crafted "differentiated" products in blind comparison, and prefer the latter when the product category doesn't imply commodity status. ### 📊 Methodology - H1: Longitudinal case study of 5-10 design systems, coding each for AI-tooling reliance vs. human-judgment practice, tracking brand-distinctiveness and user-delight survey scores (e.g., Kano-style delighter mapping) over 12-24 months. - H2: Retrospective survey of design-system practitioners correlating presence/absence of a named steward role with system-relaunch/rebuild frequency. - H3: A/B comparison of teams randomly assigned continuous vs. discovery-only research cadences, measuring time-to-detection of shifting feature expectations via periodic Kano surveys. - H4: Inter-rater reliability study comparing AI-automated vs. human expert scoring on a set of "craft" artifacts (empty states, micro-interactions) vs. a set of competency artifacts (contrast ratios, load times). - H5: Blind A/B user study comparing AI-generated-only vs. AI-assisted-plus-human-crafted product variants across non-commodity categories, measuring preference and perceived quality. ### ❓ Questions - Where exactly is the line between "measurable" and "judged" — is accessibility's writing-level/style criterion actually competency or partly differentiator, and does that ambiguity matter operationally? - Can AI evaluation tooling, given enough training data on "craft" judgments (design critiques, dark-pattern audits), eventually approximate taste — or is there a principled reason (not just current capability) it can't? - What does "stewardship" look like operationally at organizations too small to have a dedicated design-system team — does the argument scale down? - How do you measure "convergence" (everyone's AI output looking the same) at an industry level, and is there early empirical evidence for it yet, or is this still anecdotal? - Hunt's scorecard/dashboard measurement system is self-described as "alpha" — what would a mature version need to actually unify quantitative and qualitative differentiator signals? ### 🔮 Future Research - Empirical measurement of AI-driven design convergence across a sample of live sites/products (e.g., visual/interaction similarity scoring pre- and post-widespread AI design tool adoption). - Development of judged-quality benchmarks (craft, empathy, respect) with inter-rater reliability testing, to see whether "judged not measured" is a durable claim or a temporary tooling gap. - Case studies on design-system stewardship models (dedicated role vs. distributed ownership vs. none) and their correlation with system health/longevity. - Study of how Lean UX-style continuous feedback loops perform specifically in agency/client relationships (vs. in-house product teams), given Hunt's own context is agency-adjacent (Lone Rock Point serving clients). ### 💡 Implications - **Practice**: Design teams should explicitly split design-system QA into a competency checklist (automatable, AI-assistable) and a differentiator checklist (requires design critique, user research, human sign-off) rather than treating "the system passed automated checks" as sufficient. - **Policy**: Organizations adopting AI design tooling should pair adoption with an explicit named-owner/steward policy, per Hunt's "machine can't be responsible" argument — this is a governance gap, not just a tooling gap. - **Security**: Not directly addressed, though reliability's "error handling" and "bug reporting" competency items intersect with security practice tangentially (not elaborated in source). - **Education**: Design education should teach Kano-model and competency/differentiator frameworks explicitly, since Hunt treats them as foundational vocabulary the audience is assumed not to fully know. - **Funding models**: Agencies/clients should budget explicitly for "ongoing research" and continuous feedback loops as an operating cost, not a one-time discovery-phase line item — Hunt frames this as a funding-model change, not just a process change. ### 🧠 Learning Styles Assessment - Reading level estimate: approximately Flesch-Kincaid grade 8-9 on the underlying spoken transcript (conversational register, short clauses, frequent filler/repair typical of live speech; ideas are conceptually dense despite simple sentence structure). #### 👀 Visual learners - Present: the talk references several visual aids (satisfaction charts, Kano model diagram, user journey map) not captured in the transcript/audio-only source. - Improve: add the actual slide deck or a rebuilt diagram set (ceiling/floor chart, Kano quadrant, competency-vs-differentiator matrix) alongside the transcript. #### 🔊 Aural learners - Present: this source is itself the full audio/talk — inherently well-suited to aural learners. - Improve: add chapter markers/timestamps to the original video for the eight major sections identified in Main Content above. #### 🗣️ Verbal learners - Present: Hunt uses extended analogies (car/map metaphors, cruise control, online traffic maps) and clear term definitions (Kano model, stewardship, ownership from "the dictionary"). - Improve: a glossary sidebar defining competency marker, differentiator, stewardship, ownership, and Lean UX loop terms in one place. #### 🧩 Kinesthetic learners - Present: none — this is a lecture-format conference talk with no hands-on exercise. - Improve: pair with a worksheet applying the assume/build/measure/learn/adjust loop to the learner's own project. #### 🧠 Logical learners - Present: the competency-marker and differentiator sections are structured as clean parallel four-item lists; the Lean UX loop is a clear five-step cyclical process. - Improve: a decision-tree or table cross-referencing the four competency markers against the four differentiators with example tooling for each cell. #### 🧑‍🤝‍🧑 Social learners - Present: live Q&A segment included at the end. - Improve: a discussion-guide version of the Q&A question ("how do you counter AI homogenization?") for team workshops. #### 🧘‍♀️ Solitary learners - Present: the "assume" hypothesis-writing template is directly usable for individual reflection/journaling. - Improve: a self-audit checklist ("is my project's design system doing floor work, ceiling work, or both?"). #### 🌿 Naturalistic learners - Present: none directly — no sustainability/ecosystem framing in this talk. - Improve: could tie the "entropy/reactivity as downward forces" metaphor to ecological succession/maintenance analogies if adapted for this audience. ### 📊 Accessibility Assessment #### Screen reader semantics - Present: N/A for this source (raw transcript, no original HTML/document structure to assess). - Improve: when publishing this transcript, use semantic heading structure matching the section breaks identified in Main Content/Notes. #### Keyboard navigation - Present: Not applicable to this source — audio/video content, no interactive UI to navigate. #### Contrast and readability - Present: Not applicable to this source — no visual text/UI in the transcript itself. #### Text resizing and reflow - Present: Not applicable to this source — plain text content reflows natively. #### Media and captions - Present: this document *is* the transcript (captions), satisfying the core accessibility requirement for the source video. - Improve: add speaker-labeled timestamps and an audio-description note for the visual aids (charts/diagrams) referenced but not described verbally in enough detail for a screen-reader/audio-only user to reconstruct them (e.g., the Kano model quadrant is named but not fully described). #### Forms and interactions - Present: Not applicable to this source — no forms or interactive elements. #### Standards reference - Primary applicable WCAG 2.2 AA criterion: 1.2.2 Captions (Prerecorded) — satisfied by this transcript's existence. Gap against 1.1.1 Non-text Content: the visual charts/models discussed (satisfaction curve, Kano quadrant, user journey map) have no text-equivalent description precise enough to fully convey their content to a non-sighted user from audio alone. ### 🌍 DEIB Assessment #### Cultural breadth - Present: examples are US-centric (Amazon Basics, US phone plans, big-box stores); no international/multi-region examples. - Improve: supplement with non-US examples of "delighter becomes must-have" feature drift and commodity-vs-differentiated product categories. #### Inclusive language - Present: gender-neutral throughout ("users," "people," "your audience"); jargon (Kano model, Lean UX, stewardship) is defined in-talk. - Improve: a written glossary would help non-native-English listeners parse the conversational/filler-heavy live-speech register. #### Neurodiversity support - Present: consistent four-item parallel structure for both competency markers and differentiators aids predictability; car/map metaphors are consistently reused as organizing devices. - Improve: the live talk's structure (long unbroken verbal sections, minimal explicit signposting between the ~8 major topic shifts) would benefit from added headers/chunking in any written adaptation — which this note provides. #### Representation - Present: the speaker identifies her own role/credentials directly (director of UX engineering); no other named individuals/case subjects to assess for representation. - Improve: N/A — a solo conference talk has limited scope for broader representation critique. #### Psychological safety - Present: Hunt explicitly validates the audience's likely defensiveness ("I'm not going to be a technical talk... you can leave") and validates "fine" as a legitimate outcome rather than shaming commodity work, and self-positions as "neither an AI enthusiast nor an AI hater," reducing polarization framing. - Improve: N/A — this is already a strength of the talk's rhetorical approach. ### ✅ Logical Validity #### Strong reasoning - Strong analogical reasoning: the Kano model's delighter-to-must-have migration (cruise control, unlimited talk/text) is used effectively to support the "expectations aren't static" claim — concrete, testable-in-principle examples. - Strong structural/definitional reasoning: grounding the entire talk in one authoritative definition (NN Group) and returning to parse its exact wording ("a set of standards" → who sets them) is a clean rhetorical device that keeps the argument anchored. - Strong causal framing: "competency work only lifts the floor" is supported by a clear mechanism (competency work targets worst-experience remediation, not best-experience enhancement) rather than asserted without explanation. #### Fallacies - **False dichotomy risk** (located: the "measured vs. judged" framing throughout the differentiators section) — Hunt presents competency-as-measurable and differentiators-as-judged as a clean binary, but her own examples (accessibility's writing-level/style, reliability's "does it feel trustworthy") show measurable and judged qualities overlapping. Fix: acknowledge the axes as a spectrum/mixed rather than a strict partition. - **Unsupported assertion presented as settled** (located: "a machine can produce things and a machine can evaluate things, but it can't be responsible") — this is the talk's most load-bearing claim and it's stated as a given rather than argued for, despite being exactly the kind of claim AI-skeptics and AI-optimists actively contest. Fix: either cite a source/argument for machine non-responsibility (e.g., legal/philosophical accountability literature) or flag it explicitly as a normative stance rather than a settled fact. - **Hasty generalization risk** (located: the "slop" section's claim that even best-case AI output will always be "generic and hollow and sterile") — this predicts a permanent ceiling on AI output quality based on current-state extrapolation rather than demonstrated evidence, presented with high confidence ("I believe...it's still going to be soulless"). Fix: frame as a hypothesis/bet rather than a certainty, especially given her own caveat elsewhere that today's AI tools are "exciting beta software" that will look "hilariously obsolete in six months." #### Tone of voice - Conversational, self-deprecating, and audience-aware (jokes about the room's seating layout, "spicy take" framing, informal filler-heavy delivery). Balanced/non-polarizing stance explicitly stated ("neither an AI enthusiast nor an AI hater"). Confident and declarative when stating her core framework, more hedged/exploratory when describing her own team's in-progress measurement tooling ("we're in the very early stage... alpha version"). ### 🛡️ DARVO Analysis - Not applicable to this source — this is a conference talk presenting a conceptual framework, not a response to an accusation or a defensive narrative. No deny/attack/reverse-victim-offender pattern is present; Hunt does not frame herself, AI vendors, or any party as an accused party responding to allegations. ## 🎯 Conclusion - **Bottom line**: AI is genuinely useful for raising design-system competency (consistency, accessibility, performance, reliability) — the measurable floor — but cannot substitute for the human-judgment work (personality, craft, empathy, respect) that raises the ceiling and differentiates one product from another. Without active human stewardship and ownership of a design system's evolving standards, an organization risks producing technically competent but generic, undifferentiated ("fine but not good") output, and risks building "just another artifact" rather than a living design system. - **Practical next steps**: (1) Audit your own design system against the competency/differentiator split — where are you AI-assisted vs. AI-assistable vs. requiring human judgment? (2) Name an explicit steward/owner for standards evolution if one doesn't exist. (3) Replace one-off discovery research with an ongoing research practice. (4) Adopt or adapt the assume/build/measure/learn/adjust loop for at least one active initiative and run it continuously rather than once. (5) When evaluating AI design tooling, explicitly ask "is this lifting the floor or claiming to raise the ceiling?" before adopting it for judgment-requiring work. ## 🔄 Refactor Appendix — Design Systems in the AI Era: Competency vs. Craft > Density: 2–5 Literature atoms (H3), 0–3 Permanent candidates (H4). > Defaults: 3 Literature, 1 Permanent. Resonance filter: if you can't > articulate in one line why a claim matters, drop it. > > Konik claim-header rules: present tense, declarative, ≤10 words when > possible, searchable, unique, mirrors annotation focus if present. > > Atomicity: one declarative sentence per atom. "and"/"also" → split. > > Tag pool: #bmf · #addendum · #xref · #articleseed · #storystem · #share · #aphorisms ### 📚 Literature Atoms (H3) #### [H1] AI lifts design floors, not ceilings > [!quote] ID:rw_transcript_floor_ceiling > "we're kind of only lifting the floor... The ceiling is still being impacted by the same short-sighted decision-making as before." **Context & use**: - Mid-talk, after the four competency markers (consistency, accessibility, performance, reliability) section. - Core mechanism explaining why AI-assisted design-system work alone doesn't produce differentiated output. **Tags**: #xref #aphorisms **Highlight**: [view highlight](readwise://01m0czz48mney7rrvg0646rptw) <!-- atomicity: 1 idea ✓ --> #### [H2] Differentiation is judged, not measured > [!quote] ID:rw_transcript_judged_not_measured > "differentiation is judged not measured... judgment takes taste and experience and opinions." **Context & use**: - Immediately after the personality/craft/empathy/respect breakdown. - The talk's central axis-defining claim; everything else (stewardship, feedback loops) follows from it. **Tags**: #xref #aphorisms **Highlight**: [view highlight](readwise://01m0czz48mney7rrvg0646rptw) <!-- atomicity: 1 idea ✓ --> #### [H3] Machines can't be responsible for design systems > [!quote] ID:rw_transcript_machine_responsible > "a machine can produce things and a machine can evaluate things, but it can't be responsible." **Context & use**: - Stewardship/ownership section, tied to the Nielsen Norman "repository + people" framing. - The talk's most contestable claim (see Logical Validity → Fallacies); worth flagging for cross-reference against AI-accountability literature. **Tags**: #xref **Highlight**: [view highlight](readwise://01m0czz48mney7rrvg0646rptw) <!-- atomicity: 1 idea ✓ --> ### 🧠 Permanent Candidates (H4) #### [P1] Competency-vs-differentiator framework organizes design-system maturity **Claim** (one sentence, Courtney's voice): A design system matures along two independent axes — measurable competency (consistency, accessibility, performance, reliability) that AI tooling can raise, and judged differentiation (personality, craft, empathy, respect) that requires human taste and stewardship. **Evidence pointer**: [[#H1]] [[#H2]] **Scope & conditions**: Holds for products that need to stand out from commodity alternatives; explicitly does not apply to genuinely commodity/utility products, where "fine" is an acceptable ceiling. **Links** (≥1 with annotation): - [[Lean UX]] — because applies-in the continuous-feedback-loop mechanism this talk adapts for ongoing differentiator work. - [[Design system stewardship]] — because extends the "repository + people" framing into an accountability argument. **Optional**: `proof_state: working` (empirical only) | tests would include the H4/H5 hypotheses above (blind product comparisons, longitudinal AI-tooling-reliance studies). <!-- atomicity: 1 idea ✓ --> ### ✏️ Fleeting Queue - [F1] Is "AI output ceiling is permanently generic" a testable hypothesis or an unfalsifiable aesthetic claim? — cross-reference against AI capability-progress literature #xref - [F2] Compare Hunt's "stewardship/ownership" framing to FAIR's human-accountability governance principles — possible cross-project synthesis #addendum ### → Things 3 (max 2 actions) - [ ] Draft a competency-vs-differentiator self-audit checklist for courtneyr.dev's own design system work — WordPress/design-systems context #things <!-- deep-read:sections end --> 📄 **Full text:** [[Full Document Contents/Articles/Design Systems in the AI Era Competency vs. Craft.md|Design Systems in the AI Era: Competency vs. Craft by Michelle Hunt (Lone Rock Point)]] ## Highlights - So all of those things are really important and great. But the problem is that all of this competency work is only acting on your worst experiences which by bringing them up to expectations which is extremely important, right? Um but it does mean that we're kind of only lifting the floor. We're only lifting the bottom with all of these with all of these competency markers. The ceiling is still being impacted by the same short-sighted decision- making as before. ([View Highlight](https://read.readwise.io/read/01m0d0a3hrmth9spdhjg3d7x6a)) - Note: Core mechanism: AI-assisted competency work (consistency/accessibility/performance/reliability) only lifts the floor. Ceiling stays capped by the same short-sighted decisions. - Tags: atom/literature/h1, ai-automation, design-systems - And we do also have the problem of slop and I don't want to I'm not uh neither an AI enthusiast nor a AI hater. I I kind of fall somewhere in the middle. I feel that the current state of AI and the tools we're using around AI are are more like really exciting beta software. Like we are con they're constantly changing. Something we're doing now is probably going to be hilariously obsolete in six months. That's just how it goes, right? We're kind of on the cutting edge. Um so when I say slop, I don't necessarily just mean the current state of slop where things are broken or nonsensical or hallucinating or inaccurate, right? That's like very that's careless slop. Um I'm being optimistic even in looking into the future state. I believe uh we can solve for those things and have highly competent work but it's still going to be even the best most competent most accurate work is going to be very generic and hollow and sterile. It's going to be soulless. So we're kind of converging into this this baseline of everything's fine but nothing's good. ([View Highlight](https://read.readwise.io/read/01m0d0adegy4wnwgex3qpv4fn9)) - Note: The "fine but not good" convergence risk: even best-case AI output trends generic/hollow. Flagged in Logical Validity as a hasty-generalization risk — presented as near-certain rather than a hypothesis. - Tags: design-systems, ai-automation, atom/literature/h2 - So the big difference here with these differentiation things is that differentiation is judged not measured. All of those competence metrics that we were talking about before, those are measurable, right? You know, you can say green check mark, I have the correct contrast ratios, my load times meet this metric, my test coverage is x%, I've got this much uptime, you know, I've passed all my visual regression tests, all of these things are happening. I can say yes or no. But judgment takes taste and experience and opinions. In other words, judgment connects with our shared humanity. ([View Highlight](https://read.readwise.io/read/01m0d0ah7nq1pvpf3jwk8mfbfr)) - Note: The talk's central axis-defining claim: competency is measurable, differentiation is judged. Everything downstream (stewardship, feedback loops) follows from this split. - Tags: design-systems, ux-research, atom/literature/h1 - Neielson Norman group uh even says there are two important parts to a design system. Uh one is the design repository you know all the stuff that we created and the second one is the people who are managing it. And if you're only worrying about the first one and not thinking about all the responsibilities that come with the second one my spicy take is I would argue you are not building a design system at all. You are just building another artifact that's going to become obsolete along with everything else. A design system that remains successful over time is made up of both these comprehensive outputs, right? All of the artifacts, all of the tokens, all of the rules, all of the guides, and also thoughtful inputs, which are the evidence-based updates that you're making to the system as we go on. ([View Highlight](https://read.readwise.io/read/01m0d0an83zp70mdwkjmkyatw5)) - Note: Stewardship/ownership argument: a repository without active human management isn't a design system, just an artifact heading toward obsolescence. - Tags: design-systems, atom/literature/h1 - Yeah, I mean that's exactly when I was making this talk I was thinking about like you know that convergence of everything becoming the same right because AI is kind of standardizing everyone right you know lifting up that floor and making everyone follow the same standards but like no one's doing anything cool so that's that thing um I mean right now today what I would say is don't have your AI I be a decision maker because it's not one. It's a builder and it can be a good builder, but you need to be a decision maker. So if everything's looking the same because AI is building it, it's because we're letting AI build the same things, right? [snorts] So that's the account. I mean that part is hard because it's like well yeah now I have to do like work and be creative and stuff and that takes time and energy and it's expensive and the people who are just letting AI produce the fine things will do it really fast but we're not competing to be the cheapest. I don't think I think we're trying to be good. ([View Highlight](https://read.readwise.io/read/01m0d0ar90bp5e8ffhd5p2qx9p)) - Note: Q&A answer to "how do you counter AI homogenization": AI is a builder, not a decision-maker; homogenization is a human decision-ceding problem, not an inherent AI limitation. - Tags: atom/literature/h2, ai-automation, ux-research