AI UX DAILY
Wednesday, May 6, 2026
4 stories · curated for designers
The stories
Today in AI Products
 Figma Make |
May 4 |
More Ways to Add References When Using AI to Generate or Edit Images
Figma Make now lets designers provide multiple reference images and sources when generating or editing images through AI. The update also includes version history with instant revert, question cards for structured decision-making, and a one-click context reset. Integration with Zapier connects to 9,000+ apps including Google Drive and Microsoft Office.
| “ |
Use multiple reference sources when prompting for image generation to reduce iteration cycles. Consider building prompt templates that ask structured questions (via the new question cards) rather than relying on free-form text to get more predictable AI outputs. — Designer's Takeaway |
 General UX Practice |
May 5 |
Designing Agentic AI Requires Thinking Like a Manager, Not a Prompt Engineer
New research from UX practitioners argues that agentic AI design is fundamentally about setting boundaries and establishing trust, much like managing a team. Rather than treating agents as tools that respond to commands, designers should approach them as delegated decision-makers that need clear scope limits, escalation paths, and accountability structures.
| “ |
When designing agentic workflows, map out decision boundaries and trust thresholds before building UI. Ask yourself: what decisions can the agent make autonomously, what needs human sign-off, and when should it ask for clarification instead of assuming intent? — Designer's Takeaway |
 Guidepoint on Claude |
May 5 |
Guidepoint Embeds Expert Insights into Claude via Model Context Protocol
Guidepoint launched an MCP (Model Context Protocol) integration on Claude that lets researchers access verified expert insights directly within their AI-powered workflows. This allows Claude to pull from Guidepoint's network of trusted consultants and real-time expert data without leaving the chat interface.
| “ |
Consider how your product can surface source credibility alongside AI-generated content. When users see that an agent pulls from named, vetted sources (not just general web knowledge), it changes their confidence in the output and your product's trustworthiness. — Designer's Takeaway |
| General UX Practice |
May 5 |
The Psychological Fine Print of AI: What Users Aren't Told
A short-form essay examines the implicit expectations and emotional contracts users form with AI systems, often without realizing how their behavior is being shaped. The piece highlights invisible design choices (like response tone, correction patterns, and error messaging) that influence trust and reliance.
| “ |
Audit your AI interface for hidden affordances that might inadvertently encourage over-reliance or false certainty. Check if your error messages, confidence indicators, and tone subtly suggest the AI is more capable or neutral than it actually is. — Designer's Takeaway |
Today's Idea
Boundaries Beat Blank Canvases in Agentic Design
This week's strongest pattern: designers building AI agents should stop thinking about expanding what the AI can do and start thinking about what it should NOT do. The shift from prompt engineering to management thinking, combined with concrete features like multi-source references and expert-verified credibility signals, shows that user confidence in AI comes from clear limits and transparency, not capability alone.