Shuichiro Ogawa
日本語

Notes · updated 2026-11-16 · Markdown

AI and Design Weekly Watch (2026-11-03 to 11-16)

An integrated summary of 'AI and design' developments over the 14 days from 2026-11-03 to 11-16, collected in three tiers: T1v vendor primary sources, T2 public institutions and surveys, and T3 expert opinion. The 11-09 run did not take place, so the window was widened to two weeks.

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

Developments in “AI and design” over the 14 days from 2026-11-03 to 2026-11-16 were collected in three tiers by source reliability. The usual window is the past seven days. The 11-09 run did not take place after the previous watch (AI and Design Weekly Watch (2026-10-26 to 11-02)), so the window was widened to cover the whole gap. 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 in the previous watch are not repeated. Full texts were retrieved for all 27 references, and figures and dates were checked against them before writing. Retrieval took place on the afternoon of 2026-11-16 (Japan time), and anything published later is not included.

Related: AI and Design Weekly Watch (2026-10-26 to 11-02) (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).

Design-tool updates in these two weeks clustered around 3D, motion, and image generation. Figma released 3D transform and physics-based springs in Motion. Framer added image and shader generation to its Agent. At Adobe MAX on 11-10, Adobe announced new features in Photoshop and Firefly and previewed control of its apps from third-party agents. In T2, five items were adopted, against none in the previous watch. The UK government announced reforms to its designs system, and Adobe, Envato, and OpenAI published surveys and analyses of creative work. In T3, Nielsen wrote about the future of web design and an experiment on AI-designed games. Wroblewski wrote about design systems and agents.

T1v Vendor Primary Sources

On 11-12, Figma released an effect called 3D transform. Objects can be rotated along three axes (x, y, and z), and the Figma agent can also compose 3D designs. The result can be copied as CSS from Dev Mode. It can also be sent to a coding agent through the Figma MCP. On 11-10, Figma Motion gained three features. The first is physics-based springs (mass, stiffness, and damping, set for iOS and Android). The other two are previews of animation styles and a preview of the animation as a Lottie file before export. On 11-06, Select area in the AI image editing tools changed so that clicking an object in an image selects it. This feature is available only in the macOS desktop app. On 11-05, vector illustrations and icons generated with the Figma agent were improved. Layers are grouped by function. Line work uses strokes instead of fills, so stroke weight and end caps can be changed afterward. Existing illustrations can also be attached as a style reference. There were two updates on 11-03. When a code layer is created from design frames, the prototype connections and interactions between the frames are now passed to the agent as context (code layers is in open beta). Cloning is enabled by default for new GitHub repository connections (rolling out to all plans). Only the release-notes index page was read, not the individual entries.

On 11-12, Framer added image generation and shader generation through its Agent. The Agent can be asked for an image, texture, or illustration that fits the site. Given an image, it can also swap the subject, remove the background, or restyle it. Shaders can be requested in words (halftone, bokeh blur, CRT, pencil effect, and others). A generated image can also be set as the input of a shader. The text does not state which models are used or which plans are covered. On 11-03, Convert (the funnel and A/B testing feature) gained Agent support. The Agent can be asked to create, edit, and analyze funnels and tests. Harness 2, a rebuild of the Agent harness, was published on 11-02, which is outside the window, so it is not adopted.

On 11-10, timed with Adobe MAX, Adobe announced new features across Creative Cloud and Firefly on its official blog. Photoshop gains two features. Relight adds light sources whose position, color, and brightness can be adjusted non-destructively. Generative Restyle lets the user adjust the intensity of a prompted variation with a slider. Firefly gains Firefly Design (beta), which brings five kinds of work onto a single canvas (raster, vector, motion, 3D, and print). It also gains Firefly Graph (beta), which connects AI models with Adobe tools into repeatable workflows. Firefly Builder, which turns an idea into a working prototype (an app, website, or portfolio), is in private beta. In Premiere, Color Mode, a color-grading experience for video editors, is now generally available. In the Generative Media tool in Premiere, users can choose partner models as well as Adobe Firefly (Google, Kling, Runway, and Luma).

Adobe also previewed MCP integrations. Third-party agents that users already work with will be able to call on Adobe apps (Photoshop, Illustrator, Premiere, and After Effects). The integrations are scheduled to reach Creative Cloud customers in early 2027. According to a second article published the same day, the new Firefly features are in beta, and Quick Actions is generally available. Always-on agents will be available in beta in early 2027. In Premiere mobile, music from UMG and WMG can be used in personal, non-commercial videos. Availability starts on iPhone, with Android described as coming soon. Neither article states pricing. The article bodies carry no publication date, so the dates were confirmed from the article URLs.

On 11-10, Google announced an expanded partnership with Adobe. Adobe is expanding its connectors in Gemini and Gemini Enterprise with Acrobat and additional creative tools. Adobe’s creative tools will also come to AI Mode in Google Search. Adobe and Google DeepMind will explore new editing capabilities, starting with an image model for Photoshop. Google writes that this model could enable near-real-time edits. The work is at the exploration stage, however, and no timing is given.

On 11-12, Google released Gemini 4 Hydrogen. Pricing is flat across the 1 million-token context window. Input costs $1.50 per million tokens and output costs $7.50. Google lists live UI prototyping and design-to-code iteration among the uses and says the model generates more functional, feature-complete layouts in fewer prompts. No measurement method accompanies this statement. On 11-05, Google released the video model Gemini Omni 1.2 Flash. The maximum length for generating and editing video rose from 10 to 15 seconds. The price for 720p video is $0.07 per second, 30% lower than Omni 1.1. Outputs carry a SynthID watermark. Also on 11-05, Google described how to turn a photo of a hand-drawn sketch into app code with Canvas in the Gemini app. On 11-12, five tools were added to Google Flow (Character Poster, Character Arc, Pixel Machine, Lucid Glass, and Pocket Studio). All five were built by filmmakers and artists.

On 11-04, OpenAI released the image models GPT-Image-2.6 Sunburst and GPT-Image-2.6 Flare. According to the API changelog, they support image generation and editing, with adjustable complexity, SVG generation, and animated GIF generation. Markup-guided editing was also added for supported GPT-Image-2.5 and GPT-Image-2.6 models. The body of the announcement article returned 403 and was not retrieved.

On 11-09, Anthropic released Claude Fable 5.2 and Claude Mythos 5.2. The price is $8 per million input tokens and $40 per million output tokens. This is lower than the previous versions ($10 and $50). Mythos 5.2 is available only to verified organizations. On 11-03, text generated by three Claude models (Mythos 5, Opus 4.7, and Sonnet 4.6) began to carry a text watermark. Neither item is a design-specific announcement; they are listed as changes to the models that design tools depend on.

Some vendors could not be assessed for in-window announcements. The Canva newsroom has articles on its own models and new products (the design model Canva Brush, the image-generation model Canva Craft, Affinity for iPad, and others). None of these articles shows a publication date, so they could not be placed in the window and are not adopted. Webflow, Miro, and Google Stitch are also “undetermined,” not “no announcement” (Webflow returned a parse error, Miro returned 403, and Stitch shows no dated updates).

T2 Public Institutions, Standards, and Surveys

On 11-04, the UK government (Intellectual Property Office) announced reforms to the designs system. The announcement is the government’s formal response to the 2025 consultation (2025-09-04 to 11-27). A new deferment period will allow publication of a filed design to be delayed for up to 30 months. Protection for designs involving animation will be clarified, with new filing options for moving image files. The existing legal framework for computer-generated designs will remain in place. The government writes that it will monitor AI use in the design industry and consider wider developments in AI and intellectual property. The response document itself (56 pages) was not retrieved, and the statements on AI were read only in the press release.

On 11-10, the European Commission opened a consultation on the draft revised election guidelines under the Digital Services Act (closing on 11-23). The draft addresses very large online platforms and search engines. Deepfakes should be clearly labelled and should keep their label when shared. The context is elections, and the draft does not regulate creative or design practice directly. It is listed because it shares the issue of labelling generated content.

No other public institution or standards body produced an item that qualified. The two in-window items in the US Copyright Office NewsNet (11-05) are unrelated to AI. Five bodies had no relevant in-window publication (NIST, the European Commission’s AI Office, W3C, C2PA, and Japan’s Agency for Cultural Affairs). The agenda of the Agency for Cultural Affairs’ copyright subcommittee policy meeting on 11-09 is countermeasures against piracy, with no item on AI. Six bodies could not be assessed because their primary pages could not be reached (EUIPO, the Japan Patent Office, METI, the Cabinet Office, WIPO, and ISO). The USPTO AI page shows no dates on its items and could not be assessed. The Stanford HAI articles of 11-10 and 11-15 are commentaries on reports with no stated method, so they are not adopted.

Three surveys and analyses were adopted. On 11-09, Adobe published the results of a creator survey conducted with The Harris Poll. The survey covered more than 2,000 US creators in October 2026. Creators were defined as individuals who create and publish digital content several times per month. Of the creators who use creative AI, 78% said it has accelerated the growth of their business or following. At the same time, 68% said they at least sometimes settle for a result that does not quite match what they had in mind. In the past three months, 47% gave up on AI in the middle of a project because the result they wanted took too much effort. Of the creators surveyed, 77% said the final creative decision should always remain with the human creator, even when using agents. Adobe sells creative AI, and the answers are self-reported. Only the proportions and the sample were therefore taken.

On 11-10, Envato published its annual report on AI in creative work. It surveyed 1,033 working creatives in 103 countries and territories. Of these, 87% use AI at work, and 58% use it daily. Of the creatives who use AI, 49% said it has made them faster, but only 7% said they earn more. Taste and judgment were credited most for the quality of work, at 53%. Prompt skill came last of eight factors, at 19%. More than half of the creatives who use AI have used it on key client work without telling the client. The report page does not state the survey period or how the sample was recruited. Envato sells creative assets and has a stake in the subject.

OpenAI and Upwork published a report on the freelance market dated November 2026 (the cover gives only the month). The analysis compares Upwork job postings with a sample of more than 36,000 ChatGPT conversations. Posted fixed-price budgets are approximately 70% higher for AI-related postings than for other postings. They remain about 52% higher after controlling for work subcategory and posting month. The gap in posted hourly rates is about 9% and falls to roughly 1.5% after the same adjustment. Among hired hourly contracts, hours worked in the first 90 days are approximately 41% higher. According to the authors, the gap is more pronounced in the amount of labor purchased per contract than in the posted price per hour. Among the categories close to design, the report says that Video & Animation postings grew by about a third relative to the overall marketplace. The authors themselves note that the two datasets are not directly matched and that neither represents the labor market as a whole. OpenAI and Upwork both have a stake in the subject. The category-level figures inside the exhibits could not be checked against the retrieved text, so they are not reported here.

No other research firm produced an item that qualified. Gartner (403) and McKinsey (timeout) could not be retrieved. The a16z list shows no dates, so its items could not be placed. Six publishers had no relevant in-window publication (Forrester, IDC, Pew, BCG, Designer Fund, and Dribbble). The two in-window NN/g articles (11-06 and 11-13) are an explainer and an opinion piece with no sample.

T3 Expert Opinion

In the 11-13 UX Roundup, Jakob Nielsen discussed the future of web design. His view is that direct human visits to websites will decline and that chores will be delegated to agents. Nielsen advises investing in what stays true no matter who renders the interface (accurate product facts, fair policies, and service that keeps its promises). He predicts, however, that agents will not exceed 50% of ecommerce sales until after 2030. Driving direct human visits below 10% on major websites will take until 2036, he writes. In the same issue, Nielsen cites figures such as Shopify’s report that AI-driven traffic grew 8x year over year in the first quarter of 2026. The originals of these figures were not retrieved.

The same issue also covers design organizations and research methods. Citing a series by Ioana Teleanu, Nielsen writes that UX teams are “pancaking.” In the survey by Designer Fund and Foundation Capital, half of the more than 900 designers reported no significant structural change. He also quotes Cem Kurtaraner: “We’ve lowered the cost of proposing solutions, but not the cost of judging them.” On research methods, he summarizes a study by Hazra and colleagues. Critiques of a questionnaire mined from simulated participants scored 2.44 on a 0–3 quality scale. Critiques from paid human pilot participants scored 1.53. Nielsen notes that simulated users cannot replace real customers in the study itself. He then advises piloting with AI first and with a human second.

The 11-09 issue summarizes a market experiment by Conlon and Schwardmann. In an experiment with 1,025 people, small games designed by AI from a single prompt fetched $0.47. This was 27% above the $0.37 for amateurs’ games, and the AI’s games reportedly topped professionals’ games as well. Revealing authorship left the AI’s advantage intact (6.8 cents when hidden, 6.0 cents when revealed, p = 0.79). Nielsen writes that for small games “the taste barrier has fallen.” This is Nielsen’s interpretation, and it is limited to small games. In the same issue, Nielsen summarizes a study of minute-long waits on agent runs, called “micro-waits.” Pu and colleagues prototyped a read-only side chat. With it, the time users spent on other tabs fell from 35% to 13% of the wait. Six of 16 participants caught unwanted agent decisions during the run. Median mental demand, however, rose from 2.5 to 5.

The 11-06 issue addresses oversight of long-running agents. Mittal and colleagues analyzed 163,487 Reddit posts and extracted and classified 6,623 accounts of agent failures. In a study by Suh and colleagues, a map of the agent’s decisions raised median decision transparency from 3.8 to 6.0 on a 7-point scale. Workload (NASA-TLX) dropped from 12.0 to 8.8. Nielsen sums this up as users being unable to oversee what they cannot see. Another item in the issue summarizes a study of the division of work between humans and agents. Humans reportedly set three quarters of the goals, and two thirds of the methods were proposed by AI. Humans made 86% of the final choices on those methods. Nielsen himself notes that nearly all of the figures in this study are self-reported. None of the originals of the studies summarized in the three issues was retrieved.

On 11-03, Luke Wroblewski wrote about design systems and AI agents. He opens with the sentence that no design system survives contact with the product. In his team’s recent projects, the design system and agent instructions reportedly get the first pass 90% of the way there (earlier models reached 50 to 60%). The rest is closed through a lightweight feedback loop. The team shows the agent visually where things do not line up and has it correct each one. The next step would be to feed this feedback back into the design system. Wroblewski writes, however, that current models still lack the judgment to know what to integrate and how. Given free rein, he says, the models lean toward more documentation instead of simpler, more cohesive patterns. His conclusion is that a designer’s judgment still matters. The 90% figure rests on his own experience, and no measurement method is given.

On 11-11, Wroblewski wrote that Ask LukeW, the question-answering feature on his site, is now available as an MCP server and a custom agent. More than 17,000 people have asked Ask LukeW questions so far. Inside a coding tool, it can be used as an agent called Interface Designer. It returns an assessment of the interface and suggested improvements, together with prompts for a coding agent. The article includes promotion of his own products.

On 11-09, Ethan Mollick wrote about using Claude Fable 5.2 before its public launch. When he had it build a course website, the first version was “accurate and joyless.” When he asked for something beautiful, the next attempt was a children’s game in which the player rings bells. Mollick writes that jaggedness remains around taste and judgment. He leaves open whether this is a difference in taste or a failure on his part to specify what he wanted. In a project that travels through 3D worlds linked by portals, there were 204 worlds after twenty hours, and there are now over 450. Every world is reportedly drawn live by code, with no pre-made 3D models or image files. The figures are his own report.

On 11-15, Simon Willison released a Datasette plugin that shows geospatial data on a map. He says he built most of it with a single prompt to Claude Fable 5.2. Adding “be creative and think outside the box” to the prompt produced several ideas. His favorite concerns how GeoJSON shapes are displayed. The shapes appear first as SVG previews, and a WebGL preview loads only when a shape scrolls into view. On 11-03, he added a “panel” to Datasette Agent. The agent chat sits next to table and query pages. Combined with datasette-apps, a database-backed app can be prompted into existence and then modified from the chat beside it.

Among Willison’s in-window posts, those that only quote other people (11-05, 11-06, and 11-10) are not adopted. Brad Frost (latest 10-28) and Karpathy (latest 10-31) had no new in-window posts. Geoffrey Litt, John Maeda, and Julie Zhuo also had no new in-window posts. Dan Hollick and Maggie Appleton could not be assessed.

Major Updates (Chronological)

  • 2026-11-15: Willison released the Datasette map plugin (T3)
  • 2026-11-13: Nielsen published a UX Roundup (T3)
  • 2026-11-12: Figma’s 3D transform, Framer’s image and shader generation, and Google’s Gemini 4 Hydrogen and five Google Flow tools (T1v)
  • 2026-11-11: Wroblewski released the Ask LukeW MCP server (T3)
  • 2026-11-10: Adobe MAX announcements, the expanded Google and Adobe partnership, and the Figma Motion update (T1v) / the European Commission consultation and Envato’s annual report (T2)
  • 2026-11-09: Anthropic’s Claude Fable 5.2 (T1v) / Adobe’s creator survey (T2) / Nielsen’s UX Roundup and Mollick’s article (T3)
  • 2026-11-06: Figma’s Select area (T1v) / Nielsen published a UX Roundup (T3)
  • 2026-11-05: Figma’s improved vector generation and Google’s Gemini Omni 1.2 Flash (T1v)
  • 2026-11-04: OpenAI’s GPT-Image-2.6 (T1v) / the UK government’s designs reform (T2)
  • 2026-11-03: Figma’s code layers and GitHub connection updates, Agent support in Framer’s Convert, and Anthropic’s text watermark (T1v) / Wroblewski’s article on design systems and Willison’s Datasette Agent (T3)

How to Read the Reliability Tiers

  • T1v (vendor primary): Official materials from Figma, Framer, Adobe, Google, OpenAI, and Anthropic; full texts were retrieved and checked. Google’s statement about layouts generated by Hydrogen is the vendor’s own claim, with no measurement method.
  • T2 (public institutions, standards, surveys): Two public-institution items (the UK government and the European Commission) and three surveys and analyses (Adobe, Envato, and OpenAI with Upwork). All three come from parties with a stake in the subject, so only the proportions, the samples, and the authors’ own caveats were taken.
  • T3 (expert opinion): Three from Nielsen, two from Wroblewski, one from Mollick, and two from Willison. All figures from the studies and surveys Nielsen cites come through his summaries; the originals were not checked.

Scope and Procedure

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-2026-11-16/ai-frontier.md). Records of the full-text retrieval for the 27 references are in fulltext-manifest.json in the same directory.

Unverified Items

  • The Canva newsroom articles (Canva Brush, Canva Craft, and others) show no publication date, so whether they fall in the window could not be determined.
  • In-window announcements from Webflow, Miro, and Google Stitch could not be determined. The body of the OpenAI announcement article was also not retrieved.
  • In-window publications from several public bodies could not be determined (EUIPO, the Japan Patent Office, METI, the Cabinet Office, WIPO, ISO, and the USPTO AI page).
  • New in-window publications from Gartner, McKinsey, and a16z either could not be retrieved or carry no date.
  • The UK government’s response document and the text of the European Commission’s draft guidelines were not retrieved.
  • The Envato report page does not state the survey period or how the sample was recruited.
  • The figures inside the exhibits of the OpenAI and Upwork report and the classification procedure in its appendix were not checked.
  • The originals of the studies and surveys Nielsen summarizes were not retrieved (Hazra and colleagues, Conlon and Schwardmann, Pu and colleagues, Mittal and colleagues, Suh and colleagues, Designer Fund and Foundation Capital, Shopify, and others).
  • Only the Figma release-notes index page was read, not the individual entries.

References

All accessed on 2026-11-16.

T1v Vendor Primary Sources

T2 Public Institutions and Surveys

T3 Expert Opinion


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