A developer transitioned from using GitHub Copilot for inline edits to giving full control to an AI agent for a complete portfolio website refactor. Using Claude Haiku 4.5, the agent rebuilt the site, preserving the original framework syntax and component structure while making only minor syntax fixes.

The experience proved more effective than inline suggestions, as the agent operated with full code context. As a safety measure, the developer limited terminal command execution and prompted the agent to scan for security vulnerabilities, which it detected and fixed. The agent also provided structured documentation for each iteration.

This experiment highlights the growing capability of AI coding agents to handle larger, context-aware tasks. For broader adoption, developers are advised to use sandboxed environments and avoid granting auto-merge rights to production systems. The author plans to continue exploring agent security as these tools become more integrated into development workflows.