Shuichiro Ogawa
日本語

Notes · updated 2026-10-26 · Markdown

AI and Design Weekly Watch (2026-10-19 to 10-26)

An integrated summary of 'AI and design' developments over the past seven days, collected in three tiers: T1v vendor primary sources, T2 public institutions and surveys, and T3 expert opinion.

Contents (6)
  1. T1v Vendor Primary Sources
  2. T2 Public Institutions, Standards, and Surveys
  3. T3 Trusted Individual Opinion
  4. Key Updates (Chronological)
  5. How to Read the Reliability Tiers
  6. Scope and Method

Developments in “AI and design” over the past seven days (2026-10-19 to 2026-10-26) were collected in three tiers by source reliability. The interval from the previous watch (AI and Design Weekly Watch (2026-10-12 to 10-19)) is exactly seven days, so there is no gap week. The survey covers four areas: generative AI integrated into design tools, AI adoption in design practice, design-related announcements from major AI vendors, and trends in surveys and regulation. Items already reported last week are not repeated. Full texts were retrieved for all 10 references, and figures and dates were checked against them before writing.

Related: AI and Design Weekly Watch (2026-10-12 to 10-19) (previous weekly watch) / Design Agent Tools in 2026: The Current State of Autonomous Production (state of autonomous production agents) / MCP and Design Systems: The Infrastructure Layer for Agent Integration (connecting design tools through MCP) / AX and Design: A Cross-Comparison of Academic and Industry Perspectives (academic and industry views of agentic experience design).

On the design-tool side, this week centered on how to make an agent follow a team’s own design system. Figma published an article explaining how to combine library connections, guidelines, and Skills, and listed six case studies of code layers. Framer’s MCP lets outside AI tools operate its canvas and CMS. In T2, no public-institution item qualified, and the only survey was NN/g’s survey of UX practitioners. In T3, Nielsen discussed AI use by design phase, and Willison, Mollick, and Wroblewski each reported something they had built with AI.

T1v Vendor Primary Sources

On 10-23, Figma’s official blog explained how to make the Figma agent use a design system (written by PM Wayne Lin). When a published library is connected through “Libraries” in the agent chat, the agent uses the components and variables the team has approved. A specific component or token can also be named with @. Guidelines are instruction files attached to a library, set under Library → Manage guidelines. The article itself says guidelines work best when the agent creates designs from scratch, and that consistency when editing existing designs is still being improved. Skills are instructions taught once and invoked with / for repeated procedures. The agent is available to Full seats on the Professional, Organization, Enterprise, and some Education plans, with limited availability on Starter. Collab, Dev, and View seats can use it in their drafts.

On the same day, Figma published six case studies of teams using code layers. Code layers moved into open beta on 10-15 and use AI credits. On paid plans, creating and editing is free through 2026-11-04. Among the cases, Merkle says iteration and feature development are 30% faster and that time spent on documentation fell by up to 50%. A product manager at Cox Automotive says she went through 19 versions with about an hour of hands-on-keyboard time, which she describes as “at least 10 times more efficient, or about 90% time saved.” These figures are customers’ self-reports with no stated method of calculation. They are not Figma’s own measurements either.

On 10-22, Figma added controls for Enterprise admins. Default access to paid AI credits can be set at the organization and billing-group levels. Credit access and limits can be managed through SCIM, and creating Starter or Professional plans with company email addresses can be restricted. New admin REST APIs manage teams, folders, and files. The controls roll out to Enterprise plans over the coming days.

On 10-22, Framer released Framer MCP. Through the Model Context Protocol, an AI tool can operate Framer’s canvas, manage CMS content, and publish the site. Only the entry in the updates list was read; the individual page was not retrieved.

Some vendors’ in-window announcements could not be checked. The OpenAI news and ChatGPT release-notes pages, Adobe, and Canva returned 403. These are “undetermined,” not “no announcement.” OpenAI’s RSS feed (19 items), Anthropic’s release notes, and Google’s changelog contain no direct mention of design. Google’s latest item is the general availability of Gemini 4 Argon (10-13), which falls outside the window. Figma’s iPhone 18 frame presets (10-23) are not an AI feature and were not adopted.

T2 Public Institutions, Standards, and Surveys

No public-institution or standards document published within the window and directly related to design was found. The latest item on the U.S. Copyright Office’s AI page is dated 2025-05-09. NIST’s October item and the in-window UK IPO and DCMS item (10-24) have nothing to do with creative work or copyright. The same holds for the in-window items from the Agency for Cultural Affairs (10-22, 10-23) and the latest Cabinet Office item (2026-09-18). The European Commission’s AI Office, METI, and ISO returned 403, USPTO returned a 404 and a list with no items, and WIPO shows no dates, so whether anything fell within the window could not be determined.

From the surveys (analyst), one is adopted: NN/g’s survey of UX practitioners (Maria Rosala and Rachel Banawa, 10-23). The survey ran from 2026-08-24 to 09-11, drawing a convenience sample from a practitioner panel, a newsletter, invite-only Slack communities, and LinkedIn ads. Of 2,814 responses, incomplete responses, bot responses from LinkedIn, and responses from people not currently working were removed, leaving 521 (246 designers, 145 researchers, 21 PMs, and 109 others).

Access to generative AI was above 99%, but 59% could use only an approved set of tools and 12% were free to use any tool. 66% had used at least one agentic tool, ranging from 85% at startups to 36% at nonprofits. The top barriers to adoption were output quality (54%) and data security and privacy (50%); only 9% reported no major barrier. The effect on task time averaged 4.2 on a five-point scale where 3 means no difference and 5 means much shorter. For interviewing and testing with users, many respondents never used AI: 61% for interviews and 51% for testing. Sentiment averaged 3.57 out of 5, significantly above the neutral point of 3, but not by much. 43% of respondents answered the open-ended question; 69% of those comments contained some negative sentiment, and 41% were purely negative. 18% expected their role to shrink or be eliminated, and 65% expected it to remain as important or grow. Mean job-security ratings (five-point scale) were 3.81 for PMs, 2.99 for designers, and 2.94 for researchers.

The sample has limits. The authors state that it is a convenience sample and that people with strong views are more likely to leave a comment. The PMs were recruited through NN/g’s own channels and may be more familiar with UX than PMs in general. There were only 21 PM respondents, and the 3.81 figure rests on that number. Two AI models ran the sentiment analysis with the same written rules, and a researcher checked a random 50 comments and every comment on which the two models disagreed. NN/g sells training and consulting, and the article points readers to its own AI training courses. It recommends no product and discloses its method, so only the numbers and facts were taken.

Other surveys were not adopted. Figma’s reports page shows no dates, so whether any report falls within the window could not be determined. Surveys by Designer Fund and Foundation Capital, Designlab, Autodesk, and Ornamo fall outside the window. The official sites of Gartner, Forrester, McKinsey, and similar firms were not fetched directly, so the absence of in-window items is not asserted.

T3 Trusted Individual Opinion

In his 10-19 UX Roundup, Jakob Nielsen discussed designers’ AI use by design phase. He cites an arXiv preprint (Tajarmakan and Hojjati Emami) that surveyed 443 designers in 43 countries and reports AI use of 72% in Discover, falling to 41% in Deliver. Nielsen reads this as support for his own view that ideation is cheap while finishing is not, and predicts that Deliver will catch up with Discover by 2028. He also points out the weakness of the sample. Respondents were recruited through social media and personal networks, 47% were from Iran and 60% were in their twenties, and he writes that the 80% adoption rate is probably inflated because AI enthusiasts are more likely to answer. The same issue covers a randomized trial of Figma Make (Stewart and colleagues, Figma and MeasuringU, arXiv). With 50 designers and 50 PMs, task time fell by 35% for PMs and 17% for designers, narrowing the gap between the roles by about 70%. The 70% is Nielsen’s own calculation. He notes that participants were shown a picture of the finished design, so AI closed the execution gap while the judgment gap went untested. Elsewhere he argues that AI agents need stop controls with Pause, Stop, and Undo that actually halt the work, and that icons need text labels.

The 10-23 issue discussed the design of AI products. He argues for showing the evidence beside an AI answer, for offering an editable starting point instead of a blank “Ask me anything” prompt, and, citing a Gartner survey (5,728 customers, late 2023) in which 64% would rather companies not use AI for customer service, for keeping a path to a human open. He argues that a thumbs-up button can reward flattery and that the feedback mechanism itself should be usability tested. As support he cites a study (Frik and colleagues) in which 8 of 40 products with generative AI features had no feedback mechanism, and OpenAI’s statement that thumbs data contributed to its sycophantic 2025 GPT-4o update. He also tells readers to recheck every “AI can’t do that” assumption each quarter. All figures from studies and surveys he cites come through Nielsen’s account; the originals were not retrieved.

On 10-23, Simon Willison published Quick Study. It repaints a photo stroke by stroke; users choose a medium (oil, watercolor, or pointillist), a detail level, and a duration of 2 to 10 seconds, and can hold a button to compare against the original photo. He writes that it was built from a single prompt to Claude Opus 5.5 High, uses Sobel edge detection, and has no external libraries beyond Google Fonts. He offers no evaluation, so it is listed as an example of an AI-generated interface.

On 10-20, Ethan Mollick reported work he delegated to agents in a post timed to the release of his new book, Co-Existence. From a couple of sentences of instruction, an orchestrator agent handled the research, graphing, layout, images, and citations, and he only chose among options and replied “make it better.” He calls this way of working the conjurer, and says it is powerful but makes the AI’s behavior harder to see. He also writes that Opus 5.5 redesigned the book’s website. He says humans keep an advantage in deep knowledge, wide knowledge, taste, and agency, though the link to design is thin. Figures in the post, such as the METR measure (four minutes in 2024 to 17 hours in April 2026), are his own statements.

On 10-20, Luke Wroblewski wrote that he had extended Ask LukeW on his own site. It now ingests new content every hour, transcribes audio and video, has AI describe PDFs and images, and pulls in posts from X. Automating ingestion let him start the experience with an answer instead of waiting for a question. He presents this as a response to the problem that users of AI products do not know what the product can do.

Willison’s report on a game he built (10-20) has little design content and was not adopted. Karpathy and Brad Frost showed no new posts in the window. Dan Hollick, Julie Zhuo, and John Maeda could not be reached or could not be confirmed as the authors, so they remain undetermined.

Key Updates (Chronological)

  • 2026-10-23: Figma published its guidance on the Figma agent and design systems and six code-layers case studies (T1v) / NN/g published its survey of UX practitioners (T2) / Nielsen published a UX Roundup and Willison published Quick Study (T3)
  • 2026-10-22: Figma added controls for Enterprise admins and Framer released its MCP (T1v)
  • 2026-10-20: Mollick published his Co-Existence post and Wroblewski published his Ask LukeW update (T3)
  • 2026-10-19: Nielsen published a UX Roundup (T3)

How to Read the Reliability Tiers

  • T1v (vendor primary): Full texts of Figma and Framer were retrieved and checked. The Merkle and Cox Automotive figures in the case studies are treated as customers’ self-reports.
  • T2 (public institutions, standards, surveys): No public-institution items. The only survey is NN/g’s; because the publisher sells training and consulting, only the factual parts were taken. The sample is a convenience sample.
  • T3 (expert opinion): Two items by Nielsen and one each by Willison, Mollick, and Wroblewski. Every figure Nielsen cites comes through his summary, and the originals were not checked.

Scope and Method

Details of the ledger (position judgments, exclusion records, the list of unreachable sources, and per-claim verification evidence) are in the corpus (source/review/ai-design-watch-2026-10-26/ai-frontier.md). Records of the full-text retrieval for the 10 references are in fulltext-manifest.json in the same directory.

Unverified Items

  • In-window announcements from OpenAI, Adobe, and Canva could not be determined because the official pages returned 403.
  • In-window publications from the European Commission, METI, ISO, USPTO, and WIPO could not be determined.
  • The originals of the studies and surveys Nielsen cites (two arXiv papers, Frik and colleagues, OpenAI, Gartner) and of the METR and GPQA figures Mollick cites were not retrieved.
  • New in-window publications from research firms such as Gartner, Forrester, and McKinsey were not fetched directly.

References

All accessed on 2026-10-26.

T1v Vendor Primary Sources

T2 Surveys

T3 Expert Opinion

Author: Shuichiro Ogawa (Design Researcher / Design Consultant) About me →