The Question: Design System Collaborative Learning

Episode 074 Recap: AI and Design System Visibility with Ben Callahan and Kaelig Deloumeau-Pregent

In this recap of Episode 074, Ben Callahan is joined by Kaelig Deloumeau-Pregent to share what we learned on the subject of AI and design system visibility. The conversation traces a question Kaelig first answered nearly a decade ago at a Salesforce symposium: How do we actually know how our design system is being used? We wondered how that question lands today as AI accelerates content, design, and code production. The survey was sent to over 1,000 design system practitioners and received 78 responses across four questions: (1) current level of visibility into how design system assets are used across disciplines, including by AI; (2) biggest design system concerns as agents and automation produce more content at scale; (3) how the right balance between enforcement and enablement has shifted as AI enters the picture; and (4) the one thing they'd implement today to improve visibility without becoming the design police. Ben and Kaelig dig into a striking correlation between visibility maturity and enforcement-versus-enablement preferences, the "fog of war" metaphor for systems work, why accessibility may not belong inside design systems, and what shifting roles mean for designers in an agent-driven future. Show Notes 00:00 — Welcome and reintroducing Kaelig 00:28 — The 2016 Salesforce symposium and a decade-old magic wand question 03:03 — A business opportunity: cross-discipline visibility tooling 03:39 — Walking through the four survey questions and methodology (1,000+ sent, 78 responses) 05:53 — Kaelig's in-progress article and crowdsourcing community thinking 08:26 — Question 1: Self-reported visibility levels and what surprised us 09:14 — Why design systems are for people, and the limits of robotized outreach 10:38 — Visibility as stacked layers, not a single maturity rung 11:00 — Question 2: When every concern is a top concern 12:24 — Why feedback loops may be the most critical concern 13:04 — Question 3: The balanced split on enforcement vs. enablement 14:35 — Pace layers, time, and when to enforce vs. let people roam 16:35 — Does accessibility actually belong inside the design system? 18:53 — Design system teams becoming the org's AI product-builder definers 19:30 — Educating the designers of tomorrow (including agents) 20:30 — The legal-approval bottleneck slowing AI enablement 20:55 — A standout open response: lightweight embedded signal collection at the point of consumption 21:46 — Just-in-time guidance and bringing developer experience to designers 22:31 — Pegah Amadi's Magnolia and weaving signals into workflow 23:00 — The "Fog of War" metaphor: attention as a system team's scarcest resource 24:48 — Sending scouts: proactive visibility across Slack, Drive, and roadmaps 27:03 — De-risking and saying no when the landscape shifts 28:30 — Cheap scouting with sentiment analysis and lightweight tooling 29:26 — Mapping Question 1 against Question 4: a clear visibility-to-enablement gradient 31:22 — Taming chaos vs. the plateau of sameness (Polaris, CalPete, Yesenia) 33:20 — Curiosity over policing: the posture of successful system teams 34:23 — What this means for designers facing a big role pivot 35:10 — Redwoods Community Hike at Muir Woods in June 37:11 — Closing thanks Where to Find the Hosts Ben Callahan is Founder of Sparkbox (https://sparkbox.com) and Redwoods Design System Community (https://bencallahan.com/redwoods). Read his writings, have him present at your event, or engage with him as a coach or consultant at https://bencallahan.com Kaelig Deloumeau-Pregent is a design systems leader with experience at Salesforce, Shopify, and beyond. Connect with him on LinkedIn: https://www.linkedin.com/in/kaelig/ Get the Raw Data Access the complete survey data from Episode 074 to conduct your own analysis: https://bit.ly/4t4rYv6 Review the FigJam Notes Dig into the collaborative notes we took as a community during the deep dive: https://bit.ly/3OWYGk3 Join the Conversation The Question explores design systems topics through community research and deep-dive discussions. Participate in future episodes and contribute to the next survey: https://bit.ly/answerTheQuestion

Episode 074 Deep Dive: AI and Design System Visibility with Ben Callahan and Kaelig Deloumeau-Pregent

Episode 074 Deep Dive: AI and Design System Visibility with Ben Callahan and Kaelig Deloumeau-Pregent In this deep dive, Ben Callahan is joined by Kaelig Deloumeau-Pregent—a veteran design system practitioner whose career has spanned the BBC, The Guardian, Financial Times, Salesforce, Shopify, Netlify, and most recently Intuit—to explore the intersection of AI and design system visibility. Kaelig shares how a question raised back in a 2016 design systems symposium ("If you had a magic wand, what would you change?") still resonates today: practitioners want more visibility into how their systems are actually being used. The survey was sent to 1,081 design system practitioners and received 78 responses across four questions: current level of visibility into design system asset usage, biggest concerns as AI agents produce content at scale, how the enforcement vs. enablement balance has shifted with AI, and what one thing they'd implement to improve visibility without becoming the "design police." The conversation explores the "fog of war" metaphor for incomplete knowledge in systems work, the tension between surveillance and creative freedom, librarians vs. police as governance models, and how AI changes who (or what) is deviating from the system. Show Notes 00:39 — Kaelig's background: from a French web agency to BBC, Guardian, FT, Salesforce, Shopify, Netlify, and Intuit 06:56 — Becoming a systems thinker before "design systems" was a career 07:38 — The 2016 magic-wand question and why visibility is still the wish 08:34 — Walking through the four survey questions 09:22 — Survey methodology: 1,081 practitioners, 78 responses 10:04 — Reviewing Q1: most teams have manual or partial visibility, very few have robust automated tracking 12:01 — Visibility isn't just internal; the end customer dimension and zombie code 12:27 — Q2 results: AI concerns are "all of the above," and Brandon's optimistic reframe 13:26 — Q3 results: enforcement vs. enablement is balanced, with 14% choosing "other" 14:35 — The "fog of war" metaphor and the risk of a design system surveillance state 17:02 — Peter on cultural contracting and counterbalancing forces in an org 18:58 — The "helpful Clippy" view: visibility as a signal for better docs and training 21:24 — Doug's question: is resistance to tracking a designer-specific concern? 22:13 — Greg on discipline, rigidity, and adapting design practices for AI workflows 24:22 — Lightweight, embedded signal collection at the point of consumption 25:31 — Magnolia and ESLint-style "disable with a reason" patterns for design 27:10 — Jeff on measuring adoption and building relationships to capture wins for leadership 29:48 — Alexander on percentile-matching to surface emerging patterns and snowflakes 32:02 — Pedro on treating deviations as a "confession room," not policing 33:53 — The correlation between visibility (Q1) and enablement (Q4) responses 35:20 — The "plateau of sameness" and how the design system kicks back at scale 36:16 — ToniAnn: less visibility breeds more assumptions; talk to people 37:44 — Stephen on AI flipping enforcement toward enablement, and tracking why agents deviate 39:41 — Robin on enforcement and enablement as intertwined, not opposing 42:00 — Greg on building decision points into AI skills and rules 44:57 — Danita: what level of accountability belongs to the human using AI? 45:27 — Trust cultures, talent pools, and where the cursor sits on enforcement 47:38 — Non-negotiables: accessibility and regulated environments 49:01 — Closing announcements: Redwoods Compass alpha, Config hike, Sparkbox, Southleft Where to Find the Hosts Ben Callahan is Founder of Sparkbox (https://sparkbox.com [https://sparkbox.com]) and Redwoods Design System Community (https://bit.ly/44lzHL5 [https://bit.ly/44lzHL5]). Read his writings, have him present at your event, or engage with him as a coach or consultant at https://bencallahan.com [https://bencallahan.com] Kaelig Deloumeau-Pregent writes about design systems and AI at https://www.kaelig.fr [https://www.kaelig.fr] Get the Raw Data Access the complete survey data from Episode 074 to conduct your own analysis: https://bit.ly/4t4rYv6 [https://bit.ly/4t4rYv6] Review the FigJam Notes Dig into the collaborative notes we took as a community during the deep dive: https://bit.ly/3OWYGk3 [https://bit.ly/3OWYGk3] Join the Conversation The Question explores design systems topics through community research and deep-dive discussions. Participate in future episodes and contribute to the next survey: https://bit.ly/answerTheQuestion [https://bit.ly/answerTheQuestion]

Episode 073 Recap: Design System AI Automation with Ben Callahan and Davy Fung

Episode 073 Recap: Design System AI Automation with Ben Callahan and Davy Fung Host Ben Callahan and co-host Davy Fung, a product designer on the Atlassian Design System and host of the Design System Office Hours podcast, sit down immediately following the Episode 073 deep dive to reflect on what they heard from the community. The survey was sent to 1,077 design system practitioners and received 101 responses across four questions: what percentage of your workflow could be automated with AI today; what percentage should be automated; in what areas should we avoid AI automation and why; and what does craft mean to you in a 2026 design systems context. The conversation covers the gap between "could" and "should," the fear of loss embedded in resistance to automation, how process maturity should gate automation decisions, Bill's insight that automating broken processes masks their flaws, and the community's rich catalog of ways AI is already being put to practical, targeted use in design system workflows. Show Notes 00:00 - Introduction and episode overview 00:14 - Topic recap: AI as automation in design systems, four questions asked 01:51 - Davy's starting point: Zero Height report showing 63% not using design system automation 02:48 - Top-down AI mandates vs. practical decisions about what to automate 03:18 - What's missing from the conversation: automation's impact on human connection rituals 03:40 - The "could vs. should" gap: respondents who decreased their answer between Q1 and Q2 04:00 - What the decreasers said: loss of organizational context, institutional memory, and learning 05:01 - Davy's pushback: documented knowledge scales better than single points of contact 05:58 - The language of "loss" as sensitivity to losing control, not losing value 06:16 - Ben's process maturity model: automate after you've learned the lessons manually 07:11 - The risk of skipping straight to AI before understanding the work 07:45 - Davy: scalability vs. the trap of being the sole expert in your org 08:10 - Bill's insight from the deep dive: automating a process exposes its flaws — AI won't 17:24 - Ben recaps Bill's argument: AI is powerful enough to automate things you shouldn't 18:40 - Davy on CI pipeline linting: signals over blockers, data over gatekeeping 19:55 - Ben: injecting human review earlier in the process keeps the PR doing its job 20:35 - FigJam roundup: how community members are already using AI for automation 21:00 - Use cases shared: single-use plugins, token automation, GitHub workflows, dashboards, prototyping 21:37 - Davy: Atlassian's push toward higher-fidelity prototyping with AI tools 22:35 - Davy's underrated use case: Slack MCP to capture keywords and surface support patterns 23:11 - Ben: thin slices of AI help throughout the process vs. wide-scope automation 24:15 - Closing reflections on craft: Samantha's quote — "AI is the average; craft is rising above it" 25:21 - Thanks and outro Where to Find the Hosts Ben Callahan is Founder of Sparkbox (https://sparkbox.com) and Redwoods Design System Community (https://bencallahan.com/redwoods). Read his writings, have him present at your event, or engage with him as a coach or consultant at https://bencallahan.com Davy Fung is a Product Designer on the Atlassian Design System and host of the Design System Office Hours podcast (https://bit.ly/3AQYjjI). Connect with him on LinkedIn (https://bit.ly/3XrcF2W). Get the Raw Data Access the complete survey data from Episode 073 to conduct your own analysis: https://bit.ly/4cIjAv8 Review the FigJam Notes Dig into the collaborative notes we took as a community during the deep dive: https://bit.ly/4tW5ZHA Join the Conversation The Question explores design systems topics through community research and deep-dive discussions. Participate in future episodes and contribute to the next survey: https://bit.ly/answerTheQuestion

Episode 073 Deep Dive: Design System AI Automation with Ben Callahan and Davy Fung

Episode 073 Deep Dive: Design System AI Automation with Ben Callahan and Davy Fung Host Ben Callahan is joined by co-host Davy Fung, a product designer on the Atlassian Design System (previously Meta) and host of the Design System Office Hours podcast, to explore AI as automation in design systems—what could be automated, what should be automated, where practitioners draw the line, and what "craft" still means in 2026. The survey was sent to 1,077 design system practitioners and received 101 responses across four questions: what percentage of your workflow could be automated with AI today; what percentage should be automated; in what areas should we avoid AI automation and why; and what does craft mean to you in a 2026 design systems context. The conversation covers the surprising gap between "could" and "should," the risk of using AI to automate broken processes without questioning them first, the tension between deterministic tasks and those requiring human judgment, and how community remains the best antidote to feeling overwhelmed by an ever-accelerating tooling landscape. Show Notes 00:00 - Introduction and welcome 00:29 - Guest background: Davy Fung on design systems at Atlassian and Meta 01:27 - Design System Office Hours podcast approaching episode 100 01:56 - Topic framing: AI as automation in design systems 02:22 - Survey overview: the four questions asked 03:14 - Survey stats: 1,077 sent, 101 responses 03:44 - Framing quote from Greg: craft-driven practitioners as guardrail-keepers 04:37 - Q1 & Q2 findings: could vs. should be automated 04:59 - Davy's reaction: Zero Height report showed 60% not using token automation 05:28 - Ben's take: design systems are ripe for automation by definition 09:46 - Low-level manual work as craft: some practitioners prefer curation over automation 10:17 - Community opens up: automation as habit vs. automation as know-how 13:00 - The "could vs. should" gap: more caution than capability suggests 17:00 - Davy's workflow: starting ~60–70% of work with AI or automation support 23:34 - Bill's 0%/0% answer: automation exposes flawed processes AI won't question 25:27 - Key insight: automating a hard process can mask that the process itself is wrong 26:33 - Stephen's framework: black-and-white tasks vs. tasks needing intelligent reasoning 28:01 - Practical example: using AI to write consumer-friendly token changelog messages 29:57 - Connection to Episode 072: extreme support and openness to direct conversation 30:12 - Lauren: AI used to train teams on new tools, preserving human knowledge transfer 33:00 - Q3: areas to avoid AI automation — relationships, decision-making, creative direction 36:15 - The "CEO said something" problem: top-down AI mandates without practical grounding 36:43 - Skills vs. MCP: a lively side thread from the community 38:00 - Craft in 2026: intentionality, systems thinking, and human judgment 43:00 - The V0/AI coding tool support burden falling unexpectedly on design system teams 45:02 - Community as the antidote to feeling overwhelmed by tooling change 45:31 - Doug's question: how to expose design documentation to AI via MCP 46:29 - Davy's answer: Atlassian's JSON-structured content powering their ADS MCP 47:28 - Closing reflections; encouragement to dig into Q4 raw answers on craft 47:55 - Community updates: Redwoods writing accountability group, Guy's "Cost of Yes" article 48:51 - Upcoming events: Zeroheight Converge in Newcastle (October), UX London (June, code: JOIN_BC for 20% off) 49:25 - Outro Where to Find the Hosts Ben Callahan is Founder of Sparkbox (https://sparkbox.com) and Redwoods Design System Community (https://bencallahan.com/redwoods). Read his writings, have him present at your event, or engage with him as a coach or consultant at https://bencallahan.com Davy Fung is a Product Designer on the Atlassian Design System and host of the Design System Office Hours podcast (https://bit.ly/3AQYjjI). Connect with him on LinkedIn (https://bit.ly/3XrcF2W). Get the Raw Data Access the complete survey data from Episode 073 to conduct your own analysis: https://bit.ly/4cIjAv8 Review the FigJam Notes Dig into the collaborative notes we took as a community during the deep dive: https://bit.ly/4tW5ZHA Join the Conversation The Question explores design systems topics through community research and deep-dive discussions. Participate in future episodes and contribute to the next survey: https://bit.ly/answerTheQuestion

Episode 072 Recap: Extreme Design System Support with Ben Callahan and Doug Neiner

Episode 072 Recap: Extreme Design System Support with Ben Callahan and Doug Neiner Host Ben Callahan and co-host Doug Neiner, a design system practitioner at Planview, sit down immediately following the Episode 072 deep dive to reflect on what they heard from the community. The survey was sent to 1,081 design system practitioners and received 49 responses across four questions: what support do you currently offer; how would you change it without constraints; what prevents better support; and share a story of going above and beyond. The conversation covers the standout data points: the written vs. video documentation gap, the surprisingly high rate of dev environment access, embedding, private vs. public support channels, the balance between high-touch support and burnout, and the importance of being perceived as a helper rather than a blocker. Show Notes 00:00 - Introduction and episode overview 01:46 - Q1 data highlights: written vs. video documentation gap 02:13 - Dev environment access: higher than expected at nearly 50% 02:48 - Lowering the bar for video production with modern tooling 03:15 - The perfectionist/design system practitioner Venn diagram 04:00 - Q3 data: unclear ownership is low; headcount and competing priorities dominate 04:30 - What "competing priorities" really means for system teams 05:46 - Doug's support approach at Planview: docs, Slack channels, onboarding, and local debugging 07:53 - Going beyond "access": running consumer products locally for deeper support 08:28 - The most extreme example: getting an org-issued PC to support a heavy product 09:42 - DMs vs. open channels: why private requests matter for trust 10:34 - Not everyone is comfortable asking publicly—meeting people where they are 11:20 - The problem with ticketing systems and over-streamlining support 11:49 - How private support builds trust that eventually leads to public participation 13:25 - Prioritizing relationship over efficiency: creating tickets on behalf of consumers 14:10 - Scale vs. effort framework for thinking about support types 15:42 - Embedding: initially looks high-effort/low-scale, but the impact compounds 16:21 - Doug on embedding: modeling behavior, referencing docs together, building self-sufficiency 17:50 - The other side: high-touch support and the risk of design system team burnout 18:47 - How to gauge when a support request warrants deep mentorship vs. a quick fix 21:56 - Recap of embedding discussion: Sean's reverse embedding process from Spotify 23:28 - Doug's one experience with reverse embedding and its lasting impact 24:06 - Alexander's story: misaligned incentives can undermine embedding programs 25:08 - Rebecca's insight: being a helper vs. a blocker, and how hard trust is to rebuild 26:06 - What embedding teaches you about your own system's pain points 26:31 - Staying connected to product work keeps system teams grounded in consumer reality 27:31 - Mapping stakeholders: identifying high-influence non-advocates and converting them 28:35 - Doug: influence can come from the product, not just the person 29:57 - AI in design system support: useful for self-service, but reduce touch points with caution 31:01 - Closing reflections and thanks 31:39 - Outro Where to Find the Hosts Ben Callahan is Founder of Sparkbox [https://bit.ly/4a4Gg6G] and Redwoods Design System Community [https://joinredwoods.com/]. Read his writings, have him present at your event, or engage with him as a coach or consultant at https://bencallahan.com [https://bencallahan.com] Doug Neiner is a Principal Software Engineer at Planview. Connect with him on LinkedIn [https://bit.ly/3PvuhGe]. Get the Raw Data Access the complete survey data from Episode 072 to conduct your own analysis: ** [https://bit.ly/41H6Tf7]https://bit.ly/41H6Tf7** [https://bit.ly/41H6Tf7**] Review the FigJam Notes Dig into the collaborative notes we took as a community during the deep dive: ** [https://bit.ly/4mm3uLZ]https://bit.ly/4mm3uLZ** [https://bit.ly/4mm3uLZ**] Join the Conversation The Question explores design systems topics through community research and deep-dive discussions. Participate in future episodes and contribute to the next survey: ** [https://bit.ly/answerTheQuestion]https://bit.ly/answerTheQuestion** [https://bit.ly/answerTheQuestion**]