Ami
Designed an AI interface that makes system behavior and response progress more transparent, helping users interact with less uncertainty and preserve useful insights from conversation.
Type
Solo Project
Tools
Figma
Timeframe
Jan 2026
Role
Product Designer
Problem
AI conversations often feel useful in the moment, but uncertain while they happen and hard to retain afterward
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.
Solution
Presenting AMI
A generative AI interface that makes writing feel acknowledged, waiting feel predictable, and valuable insights easier to keep. By combining subtle attention cues, honest progress feedback, and selective archiving, AMI reduces uncertainty across the full lifecycle of AI chat.
Primary Research
Interviews revealed that uncertainty in AI chat happens during composing, waiting, and saving
To ground the project in real user needs, I conducted semi structured interviews with 10 frequent AI chat users, including graduate students and early career designers who use tools like ChatGPT and Gemini more than three times a week. The interview focused on three stages of interaction: composing a prompt, waiting for a response, and saving useful information afterward. The interview pointed to a clear opportunity: AI chat interfaces need to support confidence during writing, predictability during waiting, and selective memory after the conversation.
“While I’m typing, I’m not sure it’s really following me, and when the answer is good, it’s hard to save just the part I need.”
70%
felt anxious when the system gave no feedback while they were still composing.
90%
said it was tedious to manually copy and paste useful snippets into external note taking tools.
60%
emphasized the importance of preserving context when returning to past conversations.
Competitive Analysis
Most AI chat interfaces provide answers, but few make system state or memory truly legible
I reviewed tools such as NuraPhone and Samsung Galaxy Buds Pro, which offer sound personalization, active noise cancelling, and ambient mode. While these products help users adjust their listening environment, they still rely largely on hearing alone and provide limited visual support for understanding what is happening. This revealed an opportunity to design not just better listening controls, but a hearing care interface that helps users interpret their auditory condition more clearly and connect awareness to action.
O
O
O
Design Goals
Signal Attention
Let users feel that the system is tracking their intent while they are writing, without distracting them.
Make Waiting Predictable
Replace vague loading states with feedback that helps users anticipate how long a response will take.
Support Selective Memory
Enable users to save only what matters from a conversation instead of forcing them to archive an entire chat.
Persona
Designing for frequent AI chat users who rely on conversations, but struggle to trust or retain them
Jenny Lee is a 27 year old graduate student and early career designer who uses AI chat tools several times a week for brainstorming, rewriting, planning, and reflection. She often has good conversations with AI, but still feels uncertain while waiting for long responses and frustrated when useful ideas get buried inside chat history. She wants the interface to feel responsive while she writes, honest while she waits, and helpful when she wants to save only one valuable sentence instead of an entire session.
“Sometimes I’m not sure if it’s really following what I’m writing, and by the time the conversation gets useful, I already know I’ll have trouble finding the important part later.”
Frustrations / Pain Points
• No clear signal that the system is attending while she is still writing
• Waiting feels indefinite and cognitively irritating
• Useful snippets get buried in long chat histories
• Copy and paste is tedious and breaks the flow
• Saving an entire conversation feels too broad when only one part matters
User Flow
Placing the AI user in a real chat scenario reveals what the interface needs to support
A user enters a conversation with a rough idea, begins writing a long prompt, waits for a response, reads through the output, and then decides that only part of it is worth keeping. At each stage, the interface has a different job. While composing, it should reassure the user that the system is attentive. While waiting, it should reduce anxiety by making progress legible. After reading, it should help transform a fleeting exchange into something intentionally saved and reusable. This journey reframes AI chat not as a single interaction, but as a cycle of drafting, anticipating, consuming, and curating.
Wireframe & User Testing
The first wireframes mapped AMI’s core loop from composing to curation
The initial wireframe structured the experience around three signals across one continuous flow: input glow for composing, time estimate and progress bar for waiting, and drag to archive for saving. This made the core product logic visible before moving into the final visual system.
Usability testing showed that reassurance should be subtle and saving should preserve context
To evaluate whether AMI’s signals actually reduced uncertainty, I conducted a guerrilla usability test with 5 participants. The test focused on two key moments in the chat experience: how the interface responds while users are typing, and how users save short but valuable snippets while reading long responses. The goal was not to add more feedback, but to make the system feel attentive and supportive without disrupting focus.
Reducing Visual Distraction
The initial input field used a rotating glow along its border to signal that the system was active and listening while the user typed. In response to the user testing, the rotating motion was replaced with a soft pulsing glow to preserve the sense that the system was active while reducing visual distraction and cognitive load.
“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.”
Preserving Reading Context
The initial archive interaction used a center screen modal that appeared after text selection and asked users to assign a folder or tag. In response to the user testing, the modal was replaced with an inline, selection-anchored archive tray to allowe users to save useful snippets without losing their place in the conversation.
“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.”
Final Outcome
Ami turns AI chat into a more legible and reusable interaction system
Based on the PUI testing, the final physical output developed along two complementary directions. One direction preserved familiarity while improving fluidity through touch based controls. The other made AI physically explicit by introducing a dedicated AI key. Together, these concepts explored how the device body itself could help users recognize, access, and trust AI more naturally.
Solution highlights
Input Glow
A subtle pulse appears while the user writes, signaling that the system is attending without distracting from composition.
Time Estimate and Progress
A range based estimate and progress bar reduce wait time anxiety by making generation feel more predictable and honest.
Selective Archive
Users can drag select only the most valuable parts of a conversation and save them as reusable snippets instead of archiving entire noisy transcripts.
Takeaways
AMI shows that the value of AI chat depends not only on what the model says, but on how clearly the interface communicates attention, progress, and memory
Through user research, competitive analysis, and iterative testing, AMI reframed three moments of friction in AI chat as interface problems: uncertainty while composing, anxiety while waiting, and loss of value after reading. The final design translated those moments into three connected signals that made the system feel more responsive and the output more reusable.
As AI becomes part of everyday workflows, users need more than good responses. They need interfaces that reduce cognitive uncertainty and support intentional use over time. AMI suggests that trust in AI can be shaped not only by model quality, but by how well the interface makes system state and valuable content legible.
Future iterations could explore how these signals adapt to different AI use cases, from casual conversation to deep research or collaboration. The next step is not simply adding more feedback, but making responsiveness and memory feel even more integrated across the full lifecycle of AI supported work.