Ami
Ambient Intelligence Interface
Overview.
Individual Project
Role
UI Design · UX Design · UX Research · User Flow Architecture · Prototyping
Jan 2026
Tools
Figma
Ami is a generative AI interface designed to transform fleeting conversations into permanent knowledge assets. While current AI tools suffer from cognitive uncertainty and information volatility, Ami proposes a dual strategy. It visualizes the AI's pacing to reduce wait-time anxiety, while integrating a seamless archiving system that allows users to capture, curate, and own valuable insights without interrupting the natural flow of dialogue.
Design Process.
Define
Waiting indefinitely for an answer creates anxiety
Current chatbots often display a simple spinning loader during generation. Users cannot predict if the response will be short or long, leading to undefined waiting periods and increased cognitive load.
Valuable insights are buried in the flow
As conversations lengthen, critical information scrolls away. While full chat history exists, extracting specific "key insights" for later use is cumbersome and disjointed.
Goal
Designing intuitive feedback mechanisms for AI chat interfaces through glow animations, progress indicators, and selective archiving
User Research
To ground Ami in real user needs, semi-structured interviews were conducted with 10 frequent AI chat users. The research focused on friction points during composing, waiting, and saving.
Participants (n=10): graduate students and early-career designers who use AI chat 3+ times/week.
Interview Questions
1. General AI Experience
-
Which AI platforms (ChatGPT, Claude, etc.) do you use most frequently, and for what tasks?
-
Walk me through your typical process of interacting with an AI from start to finish.
-
Have you ever felt "disconnected" or "uncertain" during a conversation with an AI? When and why?
2. Real-time Interaction
-
While typing a long prompt, do you feel the system is "listening" to you? How do you know?
-
How do you feel during the "loading" state? Does it provide enough information about the system's status?
-
Have you ever experienced "input anxiety," where you're unsure if your prompt was properly received?
3. Information Management
-
If an AI gives you a long, 5-paragraph answer but only one sentence is useful, how do you save it?
-
Can you describe your current process for organizing valuable AI-generated insights?
-
What is the biggest frustration when you try to revisit a specific piece of information from a past conversation?
Key Insights
-
70% (7/10) of users felt anxious when the system provided no feedback during the composing stage.
-
All participants (10/10) found it tedious to manually copy-paste specific snippets into external note-taking apps.
-
60% (6/10) cited context preservation as a key need for next-gen AI chat.
These findings directly informed Ami’s three core features: input glow (composing), time estimate + progress bar (waiting), and drag-to-archive (saving).
Competitive Research
AI chat has become a daily tool, but the interaction still carries moments of ambiguity. Users don’t only evaluate the final answer. They constantly read cues from the interface: Does the system understand me? Is it working? How long will this take? Will I be able to find this again?
I reviewed major conversational AI services (ChatGPT, Gemini, Perplexity, Copilot, Claude) to analyze how they communicate system state during typing, generation, and saving
*Services were selected based on worldwide AI chatbot market share (StatCounter, Jan 2026)
Service
Response while typing (input detection & feedback)
Visualization of estimated answer length or progress status
Archiving / Highlighting
ChatGPT
-When you start typing, the send button becomes active
- A Stop generating button appears in the chat window
- A blinking dot appears where the answer is being typed
- Manages history at the full-conversation level
- Archive is hard to find -low discoverability
- Pin
Gemini
-When you start typing, the send button becomes active
- A Stop generating button appears in the chat window
- Circular animation around the diamond icon + Seeking + rotating diamond
-Collects images in one place (“My Stuff”)
Perplexity
-When you start typing, the send button becomes active
- A Stop generating button appears in the chat window
- Thinking text appears where the answer is being typed
- Content is organized into Threads, Media, Apps, and Documents, but users can’t manually archive items.
Copilot
- Real-time autocomplete suggestions
- When you start typing, the send button becomes active
- A Stop generating button appears in the chat window
- Pin
Claude
-When you start typing, the send button becomes active
- A stop indicator is shown in the chat window.
- A blinking icon appears where the answer is being typed.
- An icon below the answer indicates it is still generating.​
- Users can create their own categories to save content (Projects) and can also leave comments
- Star
Key Insights
"Ambiguous System Feedback"
-
Lack of Attention Cues: Most services offer passive feedback, merely activating the send button when the user starts typing. They fail to provide active visual cues that assure the user the system is tracking their intent.
-
Unpredictable Waiting: Response generation is represented by basic animations like "blinking dots" or "thinking text" alongside a simple "Stop generating" button. The absence of estimated time or progress indicators increases user anxiety and cognitive load during the wait.
"Rigid Archiving Systems"
-
Full-Conversation Storage: Platforms like ChatGPT primarily manage history at the "full-conversation level". Users are forced to save entire chat logs, which mixes valuable insights with irrelevant noise and prompt iterations.
-
Absence of Curation Workflows: Partial saving is heavily restricted to simple "Pin" or "Star" actions, and in some cases, manual archiving is entirely unavailable. Current tools do not support extracting and curating specific interactions into reusable knowledge assets.
Key Gaps
Composing
Confidence-building cues while writing are rare.
Waiting
Most UIs only show “generating,” with little visibility into time or progress.
Saving
Chat history is stored as full sessions; selective saving and reuse are limited.
Solution
This project proposes three interface signals aligned with a user’s mental model of a responsive partner. A subtle input glow supports confident writing, time and progress indicators make waiting predictable, and drag-to-archive turns chat history into a curated library of reusable snippets.
Design Principles
1. Attention should be felt while composing.
2. Waiting should be predictable, not mysterious.
3. Memory should be selective and intentional.
Wireframe
This wireframe maps Ami’s core loop: Compose → Wait → Curate, showing how three interface signals (input glow, time/progress, drag-to-archive) guide users from drafting to saving reusable snippets.
Usability Testing & Iteration
To ensure the interface successfully reduced cognitive load without introducing new friction, I conducted a guerrilla usability test with 5 participants using low-fidelity wireframes. The test focused on input animation and the archiving logic. The insights led to two critical iterations before finalizing the visual design.
Reducing Visual Distraction: From Rotating to Pulsing Glow
Initial Design: The input field used a rotating glow along its border to signal that the system was active. Listening or processing the user’s intent while they typed.
Rotation Glow Animation
User test (n=5): Participants were asked to save a short snippet while reading a long AI response.
Feedback:
"“It looks nice, but the moving glow keeps pulling my attention. When I’m typing, I end up tracking the animation instead of focusing on my words.”"
Archive Drag
Pulsing Glow Animation
Save Confirm
Iteration: The rotating motion was replaced with a soft, subtle pulse. This preserves the same “system is active” reassurance while reducing visual distraction and cognitive load, helping users stay focused on writing and reading in the chat.
Minimizing Reading Interruption (Context Preservation)
Initial Design: A center-screen modal asked for folder/tag inputs after text selection.
Archive Drag
Archive Tray
Save Confirm
User test (n=5): Participants were asked to save a short snippet while reading a long AI response.
Feedback:
"The large popup completely blocks the chat. It interrupts my reading flow and forces me out of the context just to save a short sentence."
Archive Drag
Archive Tray
Save Confirm
Iteration: Replaced the modal with an inline, selection-anchored menu, so users can save without losing their place in the conversation.
Input Glow
A subtle glow animation appears inside the input field while the user writes,
signaling that the system is tracking intent, rather than passively waiting for send.
Interaction Rules
- Glow softens during pauses to indicate 'still listening', not 'stopped'.
- Glow briefly stabilizes when the user edits earlier text, reinforcing that changes are recognized.
User action: begin typing → System response: glow fades in after 300ms → Reason: confirms attention without distracting single taps.
Key Interactions
Time Estimate & Progress Bar
Interaction Rules
- The time estimate is range-based for honesty and it narrows as generation progresses.​
- The progress bar reflects completion of the response.​
When the model starts generating, Ami displays an estimated remaining time in seconds plus a progress bar.
User action: waits → System response: range estimate narrows + progress advances → Reason: waiting becomes predictable and less interruptive
Archive
Instead of saving an entire chat session, users can drag-select any span across messages to archive as a reusable snippet.
Interaction Rules
- Drag selection supports cross-message ranges.
- After selection, an Archive Tray slides in.
- Saving creates a snippet card with source link.
User action: drag-select across messages → System response: tray appears and saves a snippet card → Reason: captures distilled insights without saving entire noisy transcripts.
User Flow
Chat experience is structured into four phases: Input, Anticipation, Consumption, and Curation, to keep system feedback consistent across the conversation lifecycle.
Input: Glow confirms typing/pausing states.​​
Anticipation: ETA range shapes waiting expectations.
Consumption: Progress stays synced with streaming output.
Curation: Selection turns chat into reusable snippets.
Archive System
Archive as a Snippet Library
Ami turns chat history into a reusable snippet library. Instead of saving entire sessions, users can save only what matters, organize it, and return to it later.
Archive Home
All
A central library where every saved snippet lives in one place.
Group (By Folder)
Snippets can be grouped by topic or intent to support lightweight curation.
Check
Snippets can be grouped by topic or intent to support lightweight curation.
In-Chat
Selective saving, directly in the conversation
Users highlight only the useful part of a response. A minimal Save entry point appears without interrupting reading.
Save to the Right Place
A quick folder dropdown lets users store the snippet where it belongs, turning saving into a small curation step.
Saved Confirmation
A confirmation message shows exactly where the snippet was saved, so users don’t second-guess whether the action worked.
Sign-in
Just-in-time sign-in
If the user isn’t signed in, Ami prompts login only at the moment saving becomes meaningful. This keeps exploration frictionless while protecting long-term memory.
Final Product
A high-fidelity interactive prototype demonstrating Ami’s full flow, from composing to predictable waiting, and from selective archiving to retrieval. The demo video shows the end-to-end interaction in real time, while the embedded prototype lets you explore key states and transitions.