Shuichiro Ogawa
日本語

Notes · updated 2026-10-05

AI and Design Weekly Watch (2026-09-28 to 10-05)

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 T3 expert opinion. Anthropic released Claude Sonnet 5.5 and described it as strong at polishing interfaces and following slide templates.

Contents (5)
  1. T1v Vendor Primary Sources
  2. T2 Public Institutions and Research
  3. T3 Personal Views of Credible Individuals
  4. Recent Major Updates (Chronological)
  5. How to Read the Reliability Tiers

Developments in “AI and design” over the past seven days (2026-09-28 to 2026-10-05) were collected in three tiers by source reliability. The gap since the previous watch (AI and Design Weekly Watch (2026-09-21 to 09-28)) is exactly seven days, with no missed week. The scope covers four areas: integration of generative AI and design tools, AI adoption in design practice, design-related announcements from major AI vendors, and research and regulatory developments. Items already known from the previous watch are not repeated in this note. The full text of all eight references was retrieved, and figures and dates were checked against that text before writing.

Related: AI and Design Weekly Watch (2026-09-21 to 09-28) (the previous weekly watch) / Design Agent Tools in 2026: The Current State of Autonomous Production (the current state of autonomous production agents) / MCP and Design Systems: The Infrastructure Layer for Agent Integration (connecting design tools via MCP) / AX and Design: A Cross-Comparison of Academic and Industry Perspectives (contrasting academic and industry views of agentic experience design).

On the vendor side this week, the updates lowered model prices and widened what agents can do. Anthropic released Sonnet 5.5 at a lower price than its predecessor, and OpenAI added GPT-6.1 Sol and a computer use tool that runs a hosted browser. On the design-tool side, Figma made Motion settings savable and reusable, and said agents can apply them automatically. In T2 the only change within the window was one executive order, and in T3 Nielsen questioned how studies that remove AI and then measure performance should be read.

T1v Vendor Primary Sources

Anthropic released Claude Sonnet 5.5, the second model in the Claude 5.5 family (09-28). It is priced at $2 per million input tokens and $10 per million output tokens, with cache reads at $0.20 and cache writes at $2.50. The announcement says the model is strongest at well-scoped everyday tasks, bug fixes, documents, slides, and spreadsheets, and that it has “a sharp eye for design.” It also reports early-tester comments that the model adds polish to user interfaces and can follow slide templates to produce decks that need little editing. In an internal test, Anthropic gave the model a public company’s earnings materials, call transcripts, and a slide template, and asked for a 10-slide operating review. Two experts judged the first draft ready to send as is. Its Terminal-Bench 4.0 score is 70.6%, up from 10.3% for Sonnet 5. The comparisons “30%+ faster than Sonnet 5” and “up to 30% less for most work,” the slide judgment, and the benchmark scores are all Anthropic’s own evaluations, with no third-party verification. The smaller Haiku 5.5 is due to join the family in the coming weeks.

OpenAI posted three items in its API changelog on 09-29. The first is GPT-6.1 Sol (gpt-6.1-sol), for complex coding and professional work. The changelog says it costs less than GPT-6 Astra; for prompts of up to 272K input tokens, pricing per million tokens is $2 input, $0.10 cached input, $2.50 cache write, and $10 output. It also supports Multi-agent in beta, where the model delegates work to subagents within a single Responses API request. The second is computer use in the Agents API. Agents complete tasks in an OpenAI-hosted browser, while website access approvals and sign-in are handled by the calling application. The third is Ultrafast mode for GPT-6 Astra, enabled with service_tier: "ultrafast", which reduces the time between generated output tokens. It is subject to rate limits, uses global processing with US data residency, and does not support EU or other regional inference residency.

Figma added custom styles and other features to the open beta of Motion (09-30). An animation style stores easing, duration, and similar properties; styles can be published to libraries and combined into a single reusable pattern. The update page says Figma agent, and coding agents through MCP or Skills, can apply these styles automatically. Audio can now be placed on the timeline and exported with the animation in MP4 and WebM. Text can be animated by character, word, or line and stays editable afterward. Animations can also be exported as Lottie or dotLottie files.

On 10-01 Figma published a walkthrough of using Figma Weave to turn a design system into campaign assets. The workflow chains Figma, Image Describer, Color Palette, Gen Effect, and Any LLM nodes to carry a design system’s type, color, and layout into marketing material. The museum in the example is fictional, and the post is a usage guide rather than a feature announcement. Its claim that “design systems can power more than just designs” is not backed by any test.

Some vendors could not be checked within the window. The official pages for Adobe, Canva, and Google were unreachable, or no in-window date could be established. This means “undeterminable,” not “no announcements.”

T2 Public Institutions and Research

The only T2 item qualifying within the window was U.S. Executive Order 14434, “Inaugurating the Era of Super Intelligence” (09-29). The order sets a policy that executive agencies use “Super Intelligence” (SI) instead of “artificial intelligence” in official documents and communications. The term is defined as the technologies covered by “artificial intelligence” in 15 U.S.C. § 9401(3), and that definition governs implementation of the order. Within 60 days of the order, the Assistant to the President for Science and Technology is to submit proposed legislative language to the President. It is not a regulation that directly changes design practice.

No survey (analyst) was adopted this week either. No survey on design with a stated methodology was published within the window. The candidates examined fell outside it: the Pew international opinion poll (09-17) does not focus on design work, the Designer Fund and Foundation Capital survey uses a self-selected sample and carries two conflicting publication dates, and the Autodesk and Figma surveys come from sellers and do not state the survey period or question wording in the text.

The nine regulatory and standards items checked last time were not re-fetched this week. The EU AI Act transparency page that the subagent tried to retrieve returned a 404. No in-window publication was found from the U.S. Copyright Office, NIST, the Agency for Cultural Affairs of Japan, METI, the Cabinet Office, or the major academic conferences.

T3 Personal Views of Credible Individuals

In the 09-28 issue of UX Roundup, Jakob Nielsen argued that many studies of “AI deskilling” actually measure a situation in which a tool that people keep using has been taken away. The trigger was an experiment by Shang Wu and colleagues at UC Irvine. 124 participants solved logic puzzles in three phases (no AI, optional AI, no AI), and performance fell when the AI was removed after use. Nielsen reads this as true but irrelevant to work in which the tool stays, and says the skill to measure is the ability to instruct and correct AI. The same issue covered a preprint from Imperial College London and colleagues. Its three experiments involved 1,951 participants, with a 28-day field study conducted with OpenAI adding 981. After a single five-minute chat, 70% of those assigned to AI chose AI for future emotional sharing, against 46% of those assigned to a human stranger. All figures come from Nielsen’s summary, and the original papers were not checked.

In the 10-02 issue, Nielsen drew on three studies to discuss ambient AI that offers help softly. InsightToast, from the University of Waterloo, listens for knowledge gaps in a conversation, pulls evidence from about 40,000 Canadian parliamentary records, and shows a snippet of up to 280 characters or a small chart as a side-channel toast. In a 16-participant study, users’ own reactive searches fell by 74% and the sources cited in their written justifications rose by 85%. Thirty percent of the toasts were false alarms. In a separate proactive writing-support study, 16 people wrote 66 pieces over one week and received 1,100 interventions. They ignored 58%, and accepted 84% of the text the AI wrote when they let it write. MultiVerse, from CMU, is a tool for songs whose lyrics are regenerated for each listener, and 8 of 10 songwriters associated it with greater control than the comparison condition. The rating differences were not statistically significant. From these, Nielsen concludes that what a creator ships is no longer a finished artifact but an “adaptation envelope” of intent, constraints, and boundaries, and that authoring behavior is a different creative act from authoring artifacts. The figures come from Nielsen’s summary, and the original papers were not checked.

On 10-03, Simon Willison wrote that pay-by-usage services and APIs should have “default hard budget caps”: a monthly limit that cuts the service off and returns errors once reached. A warning email alone is not enough, he says, because a rogue agent can consume thousands of dollars while its owner sleeps. Willison reports, quoting the announcement, that AWS announced on 09-16 an experience in which a paid-plan user can set a monthly spend limit (the AWS page notes that it is being released to a limited number of customers). He adds that Google Cloud launched a similar feature, Spend Caps, in July. The post is not directly about design; it is an operational side issue for production work delegated to agents. No T3 post directly about design was found within the window.

Recent Major Updates (Chronological)

  • 2026-10-03: Willison, “default hard budget caps” (T3)
  • 2026-10-02: Nielsen, UX Roundup of 10-02 (T3)
  • 2026-10-01: Figma publishes the Weave workflow walkthrough (T1v)
  • 2026-09-30: Figma adds custom styles, audio, text animation, and Lottie export to Motion (T1v)
  • 2026-09-29: OpenAI adds GPT-6.1 Sol, computer use, and Ultrafast mode (T1v) / U.S. Executive Order 14434 (T2)
  • 2026-09-28: Anthropic releases Claude Sonnet 5.5 (T1v) / Nielsen, UX Roundup of 09-28 (T3)

How to Read the Reliability Tiers

  • T1v (vendor primary): The full text of the Anthropic, OpenAI, and Figma pages was retrieved and checked. Sonnet 5.5’s speed, cost, slide judgment, and benchmark scores are treated as the vendor’s own evaluations. Figma’s walkthrough is a usage guide, not a test of effect.
  • T2 (public institutions, standards, surveys): The full text of the executive order was retrieved. No surveys were adopted. The nine regulatory and standards items from last time were not re-fetched.
  • T3 (expert opinion): Two Nielsen posts and one Willison post. The figures in the studies Nielsen cites all come through his summaries, and the original papers were not checked.

Scope and Method of the Survey

The ledger details (position judgments, exclusion records, the list of unreachable sources, and per-claim verification evidence) are in the corpus (source/review/ai-design-watch-2026-10-05/ai-frontier.md). The retrieval record for the eight references is in fulltext-manifest.json in the same directory.

Unverified Items

  • Whether Adobe Firefly, Canva, and Google made in-window announcements could not be determined because their official pages were unreachable.
  • The EU AI Act transparency page returned a 404 in the subagent’s retrieval. The nine items from last time were not re-fetched.
  • The original papers behind the studies Nielsen summarized (Wu et al., the Imperial College preprint, InsightToast, the writing-support study, MultiVerse) were not retrieved.
  • The official announcements of the AWS spend limit and Google Cloud Spend Caps are confirmed only through Willison’s account.
  • The details of the range of documents covered by the executive order’s terminology change were not read closely.

References

All accessed on 2026-10-05.

T1v Vendor Primary

T2 Public Institutions and Standards

T3 Expert Opinions

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