Agentic UI
Interfaces for multi-step AI actions — action logs, reversibility, and confidence signals.
Agentic UI
Interfaces for AI that takes multi-step actions on the user's behalf. Transparency, consent, and reversibility are non-negotiable. The user stays in control.
Agentic task — plan & confirm
The agent shows its plan before executing. Users approve what will happen — not just the result. Never automate without consent.
Aria wants to search for properties
Here's what I'll do — approve to run, or edit any step.
- 1
Search 847 listings
2-bed · Dubai Marina · AED 1.5–2M
- 2
Apply quality filters
Verified listings, floor > 10
- 3
Analyse price-per-sqm trends
Comparing to 90-day market average
- 4
Prepare your shortlist report
Top 5 ranked by value score
Action log — AI transparency
Every AI action is logged with a timestamp and context. Users can audit and selectively undo steps — building trust through radical transparency.
AI Activity
4 actions · this session
Generated shortlist
5 properties · 3 pages
10:44
Ranked by value score
price/sqm × location factor
10:43
Applied quality filters
Removed 12 unverified listings
10:43
Searched 847 listings
2-bed Dubai Marina AED 1.5–2M
10:42
Reversible actions — undo AI
Destructive or hard-to-reverse AI actions require explicit confirmation. Design for mistakes — always provide an undo path.
Send offer to seller?
Filters applied — 12 listings removed
Some AI actions can't be undone
Confidence signals — probabilistic UI
AI is probabilistic. The UI should be too. Show confidence levels on AI-generated data so users know when to trust, verify, or override.
Estimated value
AED 1.94M
Confidence
92%
Based on 47 comparable sales
Rental yield
5.8% per annum
Confidence
74%
Rental data sparse for this floor
Days to sell
18–35 days
Confidence
51%
High variance in this community