NEW!
UX DESIGN
AI AGENT
Agentic Marketplace
An agentic marketplace that helps buyers and sellers move surplus inventory—equipment, materials, ingredients—faster and with more confidence. Designed around intents like search, compare, list, and negotiate, with trust and structure built in so the agent can actually do work.
Timeline
1 month
Team
Solo
Role
Strategy, Design, Architecture
Problem & Challenges
At the start, the challenge wasn’t building an agent—it was designing a marketplace that stays coherent when inventory and user behavior are unpredictable. People switch between buying and selling, listings are inconsistent, and “value” depends on context. That shaped the direction toward intent-first design and a trust-led experience.
Key challenges that drove strategy:
No default path: users bounce between browse → list → negotiate
Decision-heavy commerce: condition, compatibility, and risk matter as much as search
Trust as a prerequisite: credibility signals must be consistent across categories
Messy inputs: listings vary from spec sheets to vague descriptions
Ambiguous pricing: fair value is contextual, often negotiated
Research
I learned that an agent-led marketplace lives or dies on clarity of intent. When users can switch from “I’m browsing” to “I’m listing” to “I’m negotiating” in minutes, the product can’t rely on linear flows or role-based navigation. That pushed the design toward intent-first entry points, consistent states (draft listing, active negotiation, ready-to-buy), and UI patterns that make “what happens next” obvious.
I also learned that “helpful” isn’t enough—marketplaces need confidence. So the design centered on decision support: comparisons that highlight meaningful differences, trust signals that travel with the user across contexts, and structured prompts that reduce ambiguous listings without slowing people down.
AI Workflow
I used vibe coding as a strategy tool to move fast without skipping the hard thinking. I started in Figma Make to explore clear layout directions for an intent-first marketplace, then used quick prototypes (Claude Code and Gemini AI Studio) to pressure-test the workflow end-to-end—where users hesitate, what information they need to commit, and how seamlessly they can switch between buying and selling.
The key learning was that AI makes exploration cheap, but good product design still comes from choosing the right questions: grounding prototypes in user intent, market realities, and decision risk. Once a direction “clicked,” the work shifted from generating variations to systemizing the experience—tightening IA, strengthening trust at decision points, and shaping patterns that could scale inside a real design system.
Key Solution
A raw-material marketplace where manufacturers can sell excess inventory and buyers can source smaller quantities for pilots, R&D, or short runs that don’t meet minimum-order requirements. Listings follow a standard format, and the platform guides each trade step by step with clear confirmations so both sides can transact with confidence.
Final Outcome
The final outcome was a cohesive, agent-led marketplace concept that makes trading feel guided rather than transactional—users can move from discovery to listing to negotiation without losing context or switching “modes.”
Over time, the work evolved from an early focus on agent-led search into a broader product narrative: intent-first navigation, decision support, and trust scaffolding became the backbone, so the experience could handle diverse categories and role-switching without collapsing into a generic marketplace.
Success Metrics
Intent-to-completion rate
% of sessions that successfully complete the user’s primary intent (Search → qualified shortlist, List → published listing, Negotiate → agreement reached, Purchase → order placed).
Time-to-first value
Median time from landing to the first meaningful outcome (e.g., first relevant match saved, first listing published, first offer sent).
Liquidity / match efficiency
% of active listings that receive a qualified inquiry or offer within X days, plus median time-to-first-inquiry (a direct read on marketplace health).
Trust & transaction quality
Successful transaction rate with low dispute/return incidence, paired with post-transaction confidence score (CSAT) or “felt confident buying/selling” rating.
Conversational Search
I designed the entry experience around intent-first conversational services (e.g., “find,” “compare,” “list,” “price-check”) so users can describe what they need in natural language and quickly transition into structured results and next steps. This approach reduces navigation overhead and keeps the experience coherent across categories by anchoring the UI to goals, not pages.
Order Management
I created a unified order management model that supports both sides of the marketplace—buyers track requests, offers, shipments, and receipts, while sellers manage inventory status, fulfillment responsibilities, and documentation. The key decision was to make orders state-driven (clear statuses and handoffs) so users can switch roles without losing context or duplicating work.
Negotiate & Coordinate
Negotiation was designed as a guided, trackable workflow where an agent mediates offers and terms, while allowing direct chat and structured forms when details must be explicit (price, quantity, condition, shipping, timelines). This hybrid approach keeps negotiation efficient and auditable, while preserving the human nuance required to reach agreement.
Next Steps
If this were an established product, I’d treat vibe coding as the front end of a disciplined delivery cycle. Prototypes would still be used to validate intent flows and user confidence early, but the goal would be to translate what works into shippable patterns—aligned to the design system, measurable in production, and supported by operational guardrails for trust and accountability.
Discovery + alignment would come first: I’d define target intents, category constraints, and trust requirements in partnership with Product, Legal, Ops, and Support. From there, collaborative prototyping becomes the acceleration layer—short, time-boxed sprints with PMs and key stakeholders to converge quickly on the right workflow while the context is fresh and shared.
Once a direction proves itself, the work shifts into systemization: updating the design system and component library so the UI is consistent, accessible, and buildable. Next is evaluation, where the experience is instrumented and tested—tracking intent completion, time-to-first-value, dispute rate, and user confidence through structured studies and lightweight experimentation.
Finally, launch + iteration happens as a phased rollout by category and intent, supported by continuous feedback loops and ongoing refinement of trust signals and decision support as real usage reveals edge cases and new opportunities.
