AI Agent in Action: How a Gen-Z Creator Uses 3 Steps to Sell 3D-Printed Sandals Globally

The barrier to cross-border e-commerce is being rapidly leveled by AI Agents. A Bilibili creator used a platform integrating multiple AI agents to demonstrate a complete workflow for selling 3D-printed sandals overseas from scratch — covering market research, sample costing, and storefront assets, all handled by AI. The entire process takes just three steps, yet covers the groundwork that traditionally takes cross-border entrepreneurs weeks to complete.

Step 1: Use the Industry Opportunity Agent to Gauge the Market

When going global, the first hurdle is always product selection and market assessment. The traditional approach means trawling through e-commerce platforms, industry reports, and customs data. In this demo, the creator simply invoked an agent called the "Industry Opportunity Assistant" with a plain-language prompt: which countries sell 3D-printed shoes best, and whether consumers care more about weight or value.

This highlights a core capability of AI Agents — converting vague, natural-language requests into structured data analysis. The agent outputs a Markdown-formatted report by default, but also supports HTML web page output, making the data easier for non-technical users to interpret. For self-described beginners, chart-based and web-page-style outputs lower the comprehension barrier — a practical design choice for consumer-facing Agent products.

As the demo shows, the real value of this market research step isn't just the conclusions — it's the direct link between product selling points like "lightweight" and target market preferences, providing a foundation for downstream design and pricing decisions.

Step 2: Multi-Agent Group Chat for Sampling and Cost Breakdown

What makes this workflow truly stand out is the "group chat" collaboration model. The creator added several agents with different roles — a product selection expert, a store-building expert, a design expert, and an e-commerce manager — into a group chat named "Sandal Shopping Spree," letting them collaborate on the same task.

This multi-Agent collaboration model essentially breaks a complex task into sub-tasks distributed to agents with different specializations, while maintaining information consistency through shared context (e.g., the group chat can directly access the previously generated market report). Compared to a single large model tackling the entire workflow, this division of labor offers advantages in both professional depth and output controllability.

The part that most impressed the creator was the cost calculation. After handing over the sampling requirements (same size range, as light as possible, non-slip, comfortable, multiple colors) and the costing task, what came back wasn't a vague estimated range — it was a fully itemized breakdown covering net weight, unit material cost, material waste rate, labor, OEM manufacturing, sea freight, customs duties, quoted price, and sampling inspection fees.

The agent concluded that the production cost per pair was approximately ¥32.98, with an attached gross margin calculation and route comparisons across different batch sizes — 50 pairs, 1,000 pairs, and 5,000 pairs. This kind of analysis — from fixed costs to landed costs to sensitivity analysis across different scenarios — is already approaching the work output of a junior supply chain consultant. For first-time entrepreneurs lacking industry experience, these quantified conclusions can directly shape pricing and ordering decisions.

---

Multi-Agent Collaboration is a major architectural evolution in AI applications today. A single large language model is constrained by context window size, domain depth, and task-switching overhead, often struggling with cross-domain complex tasks. The core idea of multi-Agent systems is "divide and conquer": breaking a large task into sub-tasks distributed to individual agents fine-tuned with specialized instructions (Prompt Engineering), with agents exchanging intermediate results via shared memory or a message bus, and ultimately aggregating into coherent output. This architecture already has mature implementations in software development (coding Agent + testing Agent + documentation Agent) and business analysis. Its advantage lies in each agent only needing to maintain high accuracy within its own domain, while the overall system's controllability and debuggability surpasses single-model end-to-end solutions. The group chat interface further reduces the cognitive load for ordinary users trying to understand multi-Agent collaboration.

Step 3: One-Click Generation of Store Design and Product Landing Page

With market research and sampling complete, the final step was producing all the assets needed to launch the store. The creator asked the agent to design based on the physical shoe and generate a full product line HTML page — including brand name, title, selling points, hero image, and key visuals.

The quality of the output exceeded expectations. The product landing page featured not just hero images and selling points, but was specifically optimized for the North American market. The detail the creator found most impressive was the cross-section diagram of the shoe — denser structure at the arch, with options for different layer counts, thickness, weight, and shoe sizes listed below. Even the color palette directly reflected the preferences surfaced in the Step 1 market research report.

This means the entire pipeline is connected: the conclusions from market research feed directly into the final product presentation, rather than each step operating in isolation. In cross-border e-commerce, consistency from product selection to listing assets often determines conversion rates.

---

Cross-border e-commerce platforms (such as Amazon, Shopify independent stores, AliExpress, etc.) have strict requirements for product landing pages — covering hero image specifications, A+ content, keyword density, and localized copy. Traditional execution requires collaboration across operations, design, and translation roles, with timelines typically ranging from several days to two weeks. The value of AI-generated HTML product pages lies in compressing the "brand concept → visual presentation → localized selling points" chain down to minutes, while using context continuity (directly leveraging prior market research data) to eliminate the information loss that occurs during manual handoffs. Notably, the generated HTML page can be deployed directly as an independent landing page, or used as a reference prototype for communicating with factories or designers. It serves as a "deliverable" during the startup phase — not merely a rough draft.

Key Signals Worth Watching

This demo is product-experience in nature, but it reveals several trends. First, AI Agents are evolving from "answering questions" to "completing tasks" — work like cost calculation and asset generation, which previously required dedicated specialists, is being end-to-end automated. Second, multi-Agent collaboration (group chat mode) is better suited than a single model for handling complex, multi-stage, highly specialized scenarios like cross-border entrepreneurship. Third, output-format accessibility — free switching between Markdown and HTML, chart visualization — is becoming a key factor in whether ordinary users can actually adopt Agent products.

Of course, the accuracy of the cost figures and market conclusions presented in the demo still needs validation in real-world business operations. Whether the ¥32.98 per-pair production cost holds up against an actual supply chain is a question these tools must answer before moving from "looks great" to "genuinely useful." But the tools have already compressed the information barrier and operational barrier to cross-border entrepreneurship to a remarkably low level.