Big week in AIUX. OpenAI dropped GPT-5.2 and overhauled image generation, Cursor shipped a visual editor that lets designers drag-and-drop directly in the IDE, and Google reimagined Deep Research as an autonomous agent.
The common thread? Every update is about reducing friction between intent and execution. Let's break down the UX patterns making it happen.
📱 This Week in AI Products
GPT-5.2 creates spreadsheets and presentations natively
OpenAI's new flagship model doesn't just describe how to build a spreadsheet. It builds it. Ask for a financial model and get a working .xlsx with formulas. Request a pitch deck and receive actual slides. The model also hallucinates 30% less than GPT-5.1 in testing. Source →
Pattern: Augmented Creation
New Images feature is 4x faster with precise editing
GPT Image 1.5 changes only what you ask while preserving lighting, composition, and faces. Users stay in control of exactly what gets modified. A dedicated Images section in the app lets you explore styles visually, no written prompt required. Source →
Pattern: Human-in-the-Loop
Visual Editor lets you drag-and-drop UI in your IDE
Cursor 2.2 bridges design and code: click any element, drag it around, adjust colors with sliders, then tell the AI agent to apply changes. It updates your actual React components. Designers can now "reach in and fix it" instead of describing changes to devs. Source →
Pattern: Multimodal Interaction
Debug Mode instruments your code to find root causes
Describe a bug, and the AI agent automatically adds logging statements to trace the issue. It verifies the likely cause, proposes a fix, and invites you to re-test. No more "hundreds of lines of speculative code," just targeted diagnostics you can follow. Source →
Pattern: Explainable AI
Claude Code launches in Slack for in-thread coding
Tag @Claude in any Slack thread to spin up a full coding session. It reads the conversation context (bug reports, feature requests) to determine the right repo, posts progress updates, and opens pull requests. Developers stay in their collaboration tool instead of context-switching to an IDE. Source →
Pattern: Contextual Assistance
Deep Research reimagined as autonomous research agent
Powered by Gemini 3 Pro, the new Deep Research autonomously plans multi-step investigations, navigates complex information, and synthesizes findings. Critically, it shows its research plan before executing so users can adjust the approach before the agent dives in. Source →
Pattern: Human-in-the-Loop
🎯 Steal This Week
Cursor's Visual Editor solves a problem every design-to-dev team faces: "Can you move that button 8 pixels left?" Now designers point, drag, and click. Then AI writes the code.
The UX insight: Multimodal beats single-mode for spatial tasks. If your AI product involves layout, positioning, or visual arrangement, consider combining direct manipulation with natural language. Let users show and tell.
📚 Pattern to Know
Human-in-the-Loop
This week's updates show two flavors of Human-in-the-Loop: GPT Image 1.5 lets users control what gets changed (precision editing), while Gemini Deep Research shows its plan before acting (approval gates). Both keep humans in control of AI actions.
When to use it: Any time AI actions are costly, irreversible, or consequential. Let users review before execution.
Want the full breakdown on any pattern mentioned above?
Explore All 28 Patterns →