I’m Zhengang Xu. I started in a small town in China, with limited resources and no established reputation.
Business taught me how to learn. Today, I lead a team operating 50+ e-commerce websites and design AI workflows for our work.
I didn’t start with a playbook.
I built businesses, then built a way to learn.
I organize AI into three roles: planning, execution, and review. Multiple Executors can work in parallel on separate tasks. I coordinate the handoffs and assess the proposed actions.
Operating many stores involves work across pages, product data, advertising, and SEO. I needed a clearer way to turn everyday business goals into tasks that AI could execute and check.
The Planner breaks a goal into scoped tasks. Multiple execution sessions can work at the same time, with outputs passed to the Reviewer. I relay instructions and feedback manually, and keep my judgment in the loop.
Three example workstreams · execution can happen concurrently
I define the three roles and how work moves between them.
I distribute scoped tasks across execution sessions and relay their outputs and feedback.
I evaluate suggested operations in the context of the business.
A retained WordPress + WooCommerce execution report shows a cycle of implementation, checking, and revision. Explore the recorded task below.
The task covered product structured data, one contextual internal link, and layout shifts in the mobile product gallery. This is a summary of the retained report; the original starting prompt is not available.
Source: retained execution report, not a reconstructed chat transcript. The report does not establish which AI role first found every issue. Lighthouse figures are laboratory measurements; original test screenshots are not included.
The task report checked public pages and compared product fields with backend data, rather than relying only on a completion message.
The report retained rollback instructions and left issues untouched when the checks had not demonstrated a fixable fault.
Reducing repeated context and token-related costs is a design goal. I have not yet established a controlled time or cost benchmark.
My team sells products internationally through independent e-commerce sites. We acquire customers through Facebook and Instagram advertising, Google Ads, and SEO.
I lead the team and organize how we apply AI to website creation, advertising-data monitoring, and SEO content generation and publishing.
The business is collaborative. Its results reflect our team’s combined work.Below are six anonymized store snapshots and a selected-host request chart. These are examples from our larger operation; additional store data is kept private.
April 1 – October 31, 2026
4,296 page views · 2,002 visitors. The blue series shows views; the dark series shows visitors.
Sites 01–05 only · Apr 1–Oct 31, 2026 · Jetpack.
Site 06 covers a longer period and is excluded from this total.
Page views count browsing activity, not unique people or purchases. Visitors are not deduplicated across sites. This selected sample is not the traffic total for all 50+ stores and does not measure AI’s effect on traffic. Dashboard screenshots retain their original interface language.
Requests may include bots, crawlers, and page resources. They are different from page views, visitors, or customers. The visible axis runs from early October to November 1; the complete date filter is not shown.
I struggled in a rigid school environment. Business gave me a reason to teach myself, test my assumptions, and take responsibility for what happened next.
I discovered an imported-beer delivery model on Douyin. Nearby convenience stores offered little of it, so I tested my own offer with short videos, local delivery, and customer groups for repeat purchases.
I continued while attending vocational high school, then explored bar partnerships and drink stalls in tourist areas.
An offer only becomes a business when I can reach a customer and deliver on it.
On a visit to a graduate-school campus, I noticed that new students needed transportation across a large site. I tested selling electric scooters alongside my beverage business and transported them from the shop to campus.
8-second archive edit.
Two selected clips, showing transport and scooters on campus.
Paying attention to an everyday inconvenience can reveal a business opportunity.
A friend introduced me to cross-border e-commerce. I researched the industry, moved on from beverages, and started with a cushion product on Temu. Later, we moved into independent stores.
A bigger opportunity has more connected parts. I had to understand how the whole business worked.
I first used AI for product titles and descriptions. As I explored further, I began using it for more of our operating work, eventually organizing the Planner, Executor, and Reviewer workflow.
Making money first pushed me to learn. Now, I also want to understand what I can build that creates value beyond my own immediate needs.
Explore the system I use today ↗Sanda. Muay Thai. Taekwondo.
Martial arts helped build my persistence long before I started working with AI.
Capable people. Complementary strengths.
The trust to build something together.
Entrepreneurship has taught me how much the people around a project matter. I want to meet partners whose abilities complement mine — people I can trust, learn from, challenge, and keep building with.
That is a central reason I want to join The Academy. I want to deepen my understanding of AI, work alongside ambitious builders, and discover what we could create together.
Customer acquisition, e-commerce operations, hands-on execution, team leadership, and persistence learned through real projects.
Complementary technical and product strengths, honest feedback, shared ambition, and the commitment to build and learn together.