- Published on
Learning to Build with AI: The Agents I Use
An overview of the AI coding tools I’ve been experimenting with—Claude Code and Codex—and how I’m structuring my workflow to stay tool-agnostic while maintaining strong human oversight.
Learning to Build with AI: The Agents I Use
As I’ve started documenting my work with AI-assisted development, I want to share not only what I’m learning conceptually, but also the practical tools and workflows that make this new way of building possible.
My use of AI tools for application development began shortly after the release of Claude Code in February 2025. Since then, a wave of new tools has emerged, each taking a different approach to enabling AI-assisted development.
For my own work, I’ve focused on Claude Code from Anthropic and Codex from OpenAI. These two tools best fit the development approach I prefer—one that maintains a high level of human oversight and code review. I want the AI to contribute meaningfully to the process, but I remain the ultimate approver: directing what work it does, reviewing every output, and ensuring all code meets my standards before it ships - this is "human in the loop" development.
As I’ve learned to develop alongside these new agentic assistants, I’ve been using the entry-level plans from both Anthropic and OpenAI, currently priced at $20 per month each. I often switch between agents for different tasks or when one has reached its usage limits. This ability to move seamlessly between platforms has become an important part of my workflow. It’s also shaping how I approach configuration, planning, development, and testing, ensuring everything I build is tool-agnostic and platform-independent. Managing tokens and usage is a big part of using AI for development right now and something I will be talking about more in future posts.
