Agentic Interfaces
An agent that edits files, runs commands, or books things on your behalf raises questions a chat box never had to answer. Should it show its plan first? Which steps need a person to approve them? What does someone see while it works for ten minutes, and how do they stop it halfway?
Products have converged on a few answers. 10 of 12 sources run long tasks in the background and notify when they finish, and 10 of 17 show in-flight work as a running activity log. Approval is less tidy. The largest group, 7 of 17, asks only before risky actions, 7 more leave it to a setting, and 3 ask every time. Showing a plan first and steering a run midway both lack a default.
- systems tracked
- 28
- sources cited
- 25
- evidence cards
- 142
- questions tallied
- 8
Popular questions
Each one is researched across every system we track. Pick one and it opens in chat, ready to send.
Systems tracked
The same core design systems are tallied on every topic, so answers compare like with like. See every source
- Claude
- GitHub Copilot
- ChatGPT
- Cursor
- Microsoft Copilot
- Gemini
- Notion AI
- Rovo
- Lovable
- v0
- IBM Carbon
- Salesforce Lightning
- Atlassian
- Microsoft Fluent
- Ant Design
- Perplexity
- Shopify Polaris
- GitHub Primer
- GitLab Pajamas
- Google Material 3
- Adobe Spectrum
- GOV.UK
- Mozilla Acorn
- Stripe Apps
- shadcn/ui
- Porsche Design System
- Base Web
- Radix UI
Got a agentic interfaces question?
Your first 3 questions are free. Every answer comes with the sources behind it.