MarkeTrade Agentic Project
A full React app design and implementation.An end-to-end experiment in autonomous product creation, where an AI agent transformed a high-level goal into a working application with minimal human guidance.
The experiment
This project tested how far an AI agent can go on its own. The whole process ran in Visual Studio Code with the Claude Code extension, where the agent could research, plan, write code, and change the project directly.
The agent got only a high-level objective and a set of workflow instructions. There were no detailed product requirements, no design guidelines, no design system, and no predefined visual style.
What the agent did on its own
The agent researched the market, analyzed competitors, and identified common design patterns. It wrote a Product Requirements Document, made the product and UX decisions, designed the interface, and built the application end to end.
It also added features nobody asked for. It introduced account verification and authentication flows on its own, which showed awareness of common product and compliance requirements.
Where it fell short
Compared to my guided MarkeTrade project, where I directed and validated each stage, the autonomous approach reached a working first version faster, but with clear compromises:
- It simplified key parts of the product and left out advanced filtering and sorting.
- It used a more basic interaction model.
- It ignored responsive layout and flex best practices on desktop and mobile.
- It made the trading cards fully clickable, with no dedicated Yes/No buttons. The cards looked cleaner, but the primary action had no strong visual cue, so it was harder to discover.
My role
I set the goal and the workflow rules, reviewed each output, and directed the fixes. After the first version, I used impeccable skill to improve the design.
On top of that, I led 2 reviews and refinement iterations to complete the missing functionality, improve interaction patterns, and fix responsive behavior.
What I learned
An agentic workflow can cut the effort from idea to working prototype by handling research, planning, design, and development on its own. But for now, turning that first version into a polished, production-ready product still depends on human expertise, critical review, and iteration.
I set the goal and the workflow rules, reviewed each output, and directed the fixes.
View Live Project Online >
