AIUX News

Shopify lets AI agents complete purchases, a design audit finds 855 font sizes in one app, and Jim Nielsen makes the case for interfaces that protect rather than exploit.

Three separate agents destroyed or exposed user data this week without a single confirmation step, making intent preview the most urgent pattern on your backlog.

Three concrete signals this week: a veteran designer argues the tool-first instinct is wrong, OpenAI agents publicly posted 53 user images without anyone knowing, and a Singapore GovTech designer hit the wall where AI prototyping stops working.

Three stories this week all point to the same design problem: AI agents acting without visible intent, and the UX debt that creates.

Meta turns Muse into a wearable, Figma shares what separates fast prototypes from shipped products, and a UX consultant publishes concrete spatial UI rules for AR design.

A visual regression tool makes Figma the source of truth for AI-built UIs, Rabbit abandons dedicated hardware for a cross-platform agent app, and new data shows users trust AI with their secrets even when they say they don't.

This issue covers Claude's new document editor, a real-world playbook for making design judgment legible to AI, and what happens when websites start gatekeeping AI agents.

Claude shipped Docs and Slides as artifact types, and the feature list is still the only thing written about them. Two new courses cover both: the anchored comment that is the spine of each product, the empty design system picker that looks broken and is not, and the PowerPoint export unpacked to see what actually survives.

This week's releases and research all push toward the same question: how much should an agent do before a human sees and approves what it has done?

Three concrete product developments this week that push designers to think harder about trust, safety, and who is actually in the room with your AI.

This issue covers how power users shape AI agent context, a designer's live experiment replacing walls of text with visual answers, Pinterest's room-redesign AI feature, and OpenAI's sponsored agent model that changes how ads enter conversations.

Three concrete product moves this week give designers new tools for brand-controlled content generation, beginner-friendly AI editing, and a sharper question to ask about every agent they ship.

Three concrete product moves this week that change how designers think about tools, web accessibility, and smart home control surfaces.

Designers this week have concrete reasons to rethink planning flows, local AI controls in creative tools, and what happens when half of all AI conversations break down.

Apple ships Siri AI with ChatGPT and Claude integration, a federal court rules agents act for users not developers, and Adobe Premiere adds multi-model AI generation directly in the timeline.

This week's releases and research all point at the same problem: AI agents are taking real actions in the world, and the UI for authorizing, monitoring, and recovering from those actions barely exists yet.

A pattern this week: AI systems surfacing the wrong thing at the wrong moment, and the design decisions that let it happen.

This issue covers how to catch the interface states AI generators miss, why interactive prototypes let you test complex flows earlier, and how AI is reshaping user research practice.

Three concrete moves for designers this week: how Microsoft's CoreAI design system routes guidance to both Figma and coding agents, what OpenAI's Ed Bayes learned building Codex into ChatGPT, and how a pluggable design system lets agents generate consistent UI.

Four concrete design moves this week: a workspace tool hands teams direct control over agent AI models, Apple ships its first foldable for designers to study, a Smashing Magazine piece argues buttons are the wrong starting point, and Instinct gives its agent a real email address.

LukeW makes the case for starting AI experiences with answers instead of prompts, Jim Nielsen warns about the hidden cost of interface rebuilds, and Meta's Muse launch surfaces a trust design problem worth studying.

A new AI that generates interfaces frame-by-frame, Jakob Nielsen's research on when graph UI beats chat, and a parallel-branching workflow from Lovable.

Agentic products are shipping faster than the safety and consent patterns that hold them together, and this week gave designers concrete tools to close that gap.

Two new tools put designers in control of what AI agents see and build, while a week of agent misbehavior makes the case for designing containment and transparency into every agentic flow.

A new frontier model reshapes what agents can do in your product, a family AI assistant wins users by refusing to train on their data, and a multi-platform outage is a reminder that your AI feature needs a fallback.

New AI agents are stepping into sales flows, Figma adds a long-requested variable feature, and a privacy comparison of personal AI agents surfaces real design questions about trust signals.

Real numbers from Coinbase's Code Connect experiment, a playbook for team prototyping sandboxes, and a practical color guide for dynamic data states.

A new benchmark scores how well AI coding agents actually follow design systems, Figma ships expanded generative plugin controls, Google launches a prompt-first image tool for Workspace, and Android adds motion-sickness and accessibility features.

This issue covers how Vercel teaches agents to stay on-brand with a markdown skill file, why thumbs-up and thumbs-down are too blunt for real user feedback, and how to check whether your site is ready for agents to navigate it.

This week's stories all point to the same pressure: AI is shipping faster than UX can evaluate, and designers need concrete methods to stay upstream of the output.