My role
I led the end-to-end design of Pond AI’s AI agent chat experience, a core feature of its AI model ecosystem. I work closely with founders, visual designer, and engineer team to ship this feature from 0-1 within 4 weeks.
Project Type
To C web app
0-1
MVP
Launched
Sector
AI
Web 3
Team
CEO, CPO, CTO
Engineering team
Design team
Highlights
Result
Goal
More than
108K
new users started a conversation
Context
Pond AI is building a AI model community where developers can deploy, share, and monetize on-chain machine learning models. While the tech is powerful, its value isn’t immediately clear to newcomers, many are overwhelmed by complex, technical features.
To bridge this gap, the founders envisioned a conversational AI agent platform, more than just a chatbot, this interface would serve as the gateway to the Pond ecosystem. It would allow users to engage with community-built models through natural language, helping them discover real-world applications, understand the platform’s potential, and ultimately become active participants.
The project aimed to address two key business priorities:
Design goal based on their business priorities:
Although the high-level scope and goals were clear, many details remained uncertain, including:
How do we reduce friction for first-time users?
How do we showcase the value of community-built models without overwhelming users?
...
I realized that instead of waiting for all the answers, I needed to dive in headfirst, conducting quick research, sketching key wireframes, and proactively proposing right ideas to the founders to drive more effective discussions.
Design strategy 1
Apart from having user simply click agent card to enter the chat, what are possible ways for different types of user to enter the chat.
It felt right to ask the most fundamental question first:
I designed the agent universe entry points around four mental models, enabling different users to engage with AI agents based on their intent.
Design strategy 2
Through competitive analysis, I discovered:
I brought this insight to the founders, sparking a debate: “should we prioritize sign-up rates or focus on demonstrating value first?” I successfully convinced them that showcasing the platform’s capabilities upfront would drive stronger engagement, leading to a compromise: users could send 3 messages before being prompted to sign up.
1
Prompt users to sign up, but they can try first
2
Showcase the models powering the agent, but limit visibility to 3 to avoid overwhelming users.
Design strategy 3
Deliver clear, actionable responses
Through research, I found that our target users, the younger generation interested in emerging technologies like Web3 and AI, aren’t fans of reading long, dense text.
More than
70%
of young people showing strong interest in AI's potential in the crypto space (Source: KuCoin)
Around
300 words
young people has short attention spans, prefer concise content
(Source: MarTech)
But how do current AI chat platforms handle this? Competitive analysis revealed a common pitfall:
To address this, I proposed breaking responses into structured, digestible formats:
1
1
Fold the response to show only the result first with a tag indicating the outcome, as many crypto-related tasks can be rated.
3
2
2
Show step-by-step processing and allow users to like models, encouraging model builders.
3
Subtly highlight that AI can be wrong, especially for finance-related issues where users need to be more cautious.
Reflection
It’s easy to get lost in the hype of AI, but at the end of the day, great design still starts with understanding your audience. Who are they? What’s stopping them from using your product? AI should remove friction, not add to it.
Designing from 0–1 often means there’s no clear problem statement, just a vision. Stakeholders might not have all the answers (and sometimes, they need our guidance). Instead of waiting, put ideas on paper. Sketch, iterate, and let unknowns reveal themselves along the way. Conversations become more productive when there's something tangible to react to.
Next steps
Launching the AI agent platform is just the beginning—there’s plenty more to refine and expand.
First, since Pond previously lacked a structured UX approach and design system, the entire platform is due for a revamp alongside new feature rollouts.
For the AI agent experience itself, real-world testing will give us better insights into our target audience, allowing us to collaborate with the new PM team to fine-tune the product strategy.
Some key areas for improvement:
Richer AI responses – Adding graphs, charts, and structured insights for better readability.
Smarter discovery – Introducing search functionality to help users quickly find relevant agents.
Personalization – Enhancing recommendations so returning users can instantly connect with the right agent.
Scaling for B2B – As Pond expands, we’ll explore tailored enterprise chat solutions.