AI UX DAILY
Thursday, September 24, 2026
4 stories · curated for designers
The stories
Today in AI Products
 Figma + Applitools |
via TLDR Design |
Figma frames now serve as the visual baseline to catch where AI-built code drifts from approved designs
Applitools' September 15 release lets a Figma frame URL act as the visual test baseline. Coding agents can inspect, review, and resolve visual drift from chat via the Eyes MCP Server.
| “ |
Set a Figma frame as the Applitools baseline on your next AI-assisted build sprint to catch implementation drift before it ships. — Designer's Takeaway |
 Rabbit OS3 |
Sep 22 |
Rabbit abandons dedicated AI hardware and ships OS3, a cross-platform agent app for phones and desktops
Two years after the r1 device, Rabbit is launching OS3 as a software-only agent that runs on existing screens. The pivot acknowledges that users want agents inside current devices, not new ones.
| “ |
Consider how your product's agent features fit into the apps users already have open, rather than requiring a dedicated surface or mode. — Designer's Takeaway |
| Meta Muse |
Sep 23 |
Muse gets email addresses, video chat, and Mac computer-use in a rapid post-launch update
Meta shipped Muse updates adding agent-owned email addresses, video chat access, and the ability to control a Mac. Each adds a new surface where the agent can act on a user's behalf.
| “ |
Map every new surface a persistent agent can act from and design explicit permission moments for each one, so users know what the agent can reach. — Designer's Takeaway |
| Jakob Nielsen (UX Tigers) | Sep 21 |
Users say they distrust AI but share secrets with it anyway, and cutting form fields lifts conversions
Nielsen's roundup cites data showing users disclose sensitive information to AI despite stated distrust. A separate finding: every form field removed increases conversion. Users also direct AI in two modes: specify upfront or correct as they go.
| “ |
Audit your AI product's data-collection moments against what users actually share, then design explicit trust signals at those points rather than relying on stated preferences. — Designer's Takeaway |
Today's Idea
Users trust agents with actions before they trust agents with data.
People share secrets with AI they say they distrust, and agents are rapidly gaining new surfaces to act from. Design explicit, per-surface permission moments and visible trust signals before your agent earns the access, not after.
Stop shipping AI slop
Turn your design into Claude skills
Drop a screenshot. See which of the 38 patterns you are missing and take them away as Claude Code skills. Free, no signup for the first audit.
Was this issue worth your time?