AI UX DAILY
Thursday, July 23, 2026
4 stories · curated for designers
The stories
Today in AI Products
 Claude |
Jul 22 |
Claude can now record your screen to learn your workflow and replay it as a skill
Anthropic's new 'Cowork' feature lets Claude watch your screen as you complete a task, then extract that sequence into a reusable skill it can execute on your behalf later. You demonstrate once, and Claude turns the recording into a repeatable automated workflow. This is a direct shift from prompt-based instruction to observation-based instruction.
| “ |
Consider how your product's most repetitive user flows might expose a 'record this' affordance — and then design the review screen where users can inspect, name, and edit what the AI captured before it runs autonomously. — Designer's Takeaway |
 Loom |
Jul 22 |
Loom video prompts let you show an AI agent what you want instead of writing it out
Atlassian has added video prompting to Loom, letting users record a short screen video and feed it directly to an AI agent as context. Instead of describing a task in text, you show it. The agent uses the visual and audio from the recording to understand intent and take action.
| “ |
Apply this pattern to your own product's onboarding or support flows — when users struggle to articulate a problem in text, a short screen recording can carry far more signal, so consider whether a 'show us' input deserves a spot alongside your existing text or form fields. — Designer's Takeaway |
| Dovetail |
Jul 21 |
Dovetail adds 18 integrations and first-party MCP connectors so AI tools can query your research
Dovetail expanded its ecosystem with 10 new inbound integrations to pull customer feedback into the platform, plus eight first-party MCP connectors that let external tools like Claude and ChatGPT query Dovetail's research insights directly. Teams can now ask an AI assistant a question and have it pull relevant user research from Dovetail without leaving their workflow.
| “ |
Notice how research repositories are becoming queryable by AI agents — if your team uses Dovetail, map out which AI tools your teammates already have open, then run a quick test to see whether the MCP connection actually surfaces the right insights or flattens important nuance. — Designer's Takeaway |
 Figma |
Jul 20 |
Figma publishes a plain-language explainer on MCP for designers, not developers
Figma's resource library added a guide explaining what the Model Context Protocol actually is and how designers can use it in their workflows, without requiring any engineering background. The piece covers practical scenarios like connecting design files to AI tools and using MCP to pull live content or data into prototypes.
| “ |
Use this as a team resource to run a 30-minute workshop where designers identify one repetitive context-switching task in their current workflow, then explore whether an MCP connection could collapse it into a single tool. — Designer's Takeaway |
Today's Idea
Showing beats telling, but showing needs a review step
This week's launches share a common thread: AI tools are moving from text-only instruction toward watching, listening, and observing what users actually do. That is genuinely useful, but it puts real design pressure on the moment after capture — the screen where a user reviews what the AI saw, corrects any misreading, and decides whether to trust it. That review UI is the most important surface in any observation-based AI feature, and right now most products are barely designing it at all.
Stop shipping AI slop
Audit your AI design against 36 patterns
Drop a screenshot, get specific gaps and a Claude Code prompt to fix them. Free, no signup for the first audit.
Audit your design →