Design
Building Tactile UX: Honoring Intentional Design With Lottie
This implementation case study explains how Lottie, DOM events, and distance-based calculations were used to preserve highly intentional motion in an interactive "squeeze toy" web experience.
https://www.smashingmagazine.com/2026/08/building-tactile-ux-honoring-intentional-design-lottie/
New EU Guidelines For AI Labelling
Smashing Magazine reviews the new EU guidance on AI-content labelling and the practical implications for product teams. It distinguishes cosmetic “AI” indicators from cases where transparency requirements apply.
https://www.smashingmagazine.com/2026/08/eu-guidelines-ai-labelling/
Interactive Websites Still Need Creative Judgment
Readymag examines how designers can use increasingly capable web tools without delegating aesthetic and conceptual judgment to them. The article positions experimentation, curation, and intentional constraints as central to distinctive interactive work.
https://blog.readymag.com/interactive-websites-still-need-creative-judgment-designing-with-readymag/
How Do You Actually Measure the Impact of AI on Design?
Figma describes the methodology behind its multi-year AI Index, which tracks the gap between expected and experienced effects of AI across design and product work.
https://www.figma.com/blog/measuring-the-impact-of-ai/
Workflow Lab: Moving Between Design and Code With Agents
Figma shows a workflow in which design-system context moves between Figma, FigJam, and coding agents. Its central point is that agents need structured, current context if design and implementation are to remain aligned.
https://www.figma.com/blog/workflow-lab-moving-between-design-and-code-with-agents/
Try These 10 Skills — And Show Off Your Own
Figma made it possible to author reusable markdown skills for its design agent directly in the product. The post provides examples of packaged instructions that let teams encode their own methods and visual point of view.
https://www.figma.com/blog/try-these-10-skills-and-show-off-your-own/
One AI Output Is an Example, Not an Evaluation
NN/g argues that a single successful model response does not establish a system’s reliability. It recommends representative inputs, repeated runs, and confidence intervals — treating AI evaluation more like quantitative UX research than deterministic software testing.
https://www.nngroup.com/articles/eval-ai-output/
Dogfooding vs. QA vs. User Research
This article separates internal product use, quality assurance, and user research, explaining the distinct questions each method can answer. It is a useful reminder that employees’ product knowledge makes them poor substitutes for target users.
https://www.nngroup.com/articles/dogfooding/
Rebuilding the Fashion Workflow
Krea profiles an AI-enabled fashion workflow that targets repetitive, expensive stages such as sampling and campaign production. The case study focuses on adapting generation processes to a brand’s real operating model rather than treating AI output as a one-size-fits-all replacement.
https://www.krea.ai/blog/rebuilding-the-fashion-workflow
Development
Gemini 3.7 Flash
Google announced Gemini 3.7 Flash as a faster, higher-capability general model for developer workloads. The release extends the company’s “Flash” line for tasks that need a balance of speed, reasoning, and cost.
https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-3-7-flash/
Visual Studio August Update — Work Smarter Across Models and Branches
Visual Studio added controls for Copilot reasoning effort, organization-level custom agents, and in-product usage visibility. Git support also gained worktrees and more integrated submodule management.
https://devblogs.microsoft.com/visualstudio/visual-studio-august-update-work-smarter-across-models-and-branches/
Preparing Your App for Broader Memory Limits
Android is extending per-app memory limits beyond Pixel devices to more manufacturers and RAM configurations. The guidance covers detecting throttling and termination, observing production memory use, and exercising limits with adb and profiling tools.
https://developer.android.com/blog/posts/preparing-your-app-for-broader-memory-limits
How To Evaluate LLMs Before Production
GitHub shares lessons from evaluating LLMs for secret scanning, emphasizing that production evaluations require realistic tasks, failure-mode analysis, and methods beyond benchmark scores. The article focuses on the properties that matter for a security-sensitive deployment.
https://github.blog/ai-and-ml/llms/how-to-evaluate-llms-before-production/
How Canvases Make Agentic Workflows Visible, Steerable, and Cost-Efficient
GitHub argues that a canvas makes an agent’s goals, intermediate work, and costs easier to inspect than a long chat stream. The article presents visual workspace state as a control surface for multi-step agent workflows.
https://github.blog/ai-and-ml/github-copilot/how-canvases-make-agentic-workflows-visible-steerable-and-cost-efficient/
Previewing the Model Hardware Standard
Anthropic opened a research preview of the Model Hardware Standard, a shared specification for AI agents operating laboratory and manufacturing instruments. MHS aims to replace bespoke device integrations with a common way to orchestrate physical equipment safely.
https://www.anthropic.com/news/model-hardware-standard-research-preview
How Claude’s Text Watermark Works
Anthropic explains its planned text watermarking, which encodes a detectable pattern through low-stakes token choices rather than visible or hidden characters. The company says the method does not add tokens, identify individual users, or materially change output quality.
https://www.anthropic.com/news/claude-text-watermark