Building a Personal AI Factory with Claude Code, Sonnet, and O3 (July 2025 Snapshot)

In the rapidly evolving world of AI-driven development, building a personal AI factory can significantly streamline your workflow. This July 2025 snapshot provides insights into how I've set up my own AI factory using tools like Claude Code, Sonnet, and O3 to handle planning, execution, verification, and continuous improvement.

Guiding Principle: Fix Inputs, Not Outputs

The core principle of this setup is to focus on improving the inputs rather than patching the outputs. When something goes wrong, I don't manually fix the generated code or argue with Claude. Instead, I adjust the plan, prompts, or agent mix to ensure that subsequent runs are correct by construction. This approach is akin to building a self-improving factory in Factorio-a game where you set up automated systems to produce and improve themselves.

Basic Day-to-Day Workflow

Step 1: Planning

My primary interface for this setup is Claude Code, which has become my go-to tool for all development tasks. I also use a local Management Control Program (MCP) that runs Goose and O3. Goose is integrated with our Azure OpenAI subscription, making it easy to access the necessary models.

• High-Level Task Definition: I provide a high-level task to Claude Code.
• Plan Generation: Claude Code forwards this task to O3, which generates a detailed plan.
• Plan Documentation: O3 writes out a <task>-plan.md file that includes both my original request and the implementation plan. This document serves as a reference and ensures clarity.

Step 2: Execution

The execution phase involves several steps to ensure that the generated code meets the specified requirements.

• Task List Creation: Sonnet 4 reads the plan, verifies it, and breaks it down into a task list.
• Code Generation: Claude Code executes the plan using either Sonnet 3.7 or Sonnet 4, depending on the complexity of the task. Since most of my work is in Clojure, I often use Sonnet 4 to handle the syntax correctly.
• Commit Management: An important instruction is for Claude to write commits as it completes each task step. This allows for easy reversion if issues arise.

Step 3: Verification and Feedback

After the code is generated, it undergoes a rigorous verification process to ensure quality and alignment with the original plan and ask.

• Initial Verification: Sonnet 4 verifies the generated code against the implementation plan.
• Comprehensive Verification: O3 performs an additional verification, checking the code against both the plan and the original request. O3 is particularly strict:
  • It calls out unnecessary backward compatibility code that Claude might leave in place to please.
  • It flags any "lint ignore" flags added by Claude.

Any issues identified during these verification steps are fed back into the plan template, not fixed inline. This ensures that future runs are more accurate and efficient.

Concurrent Development with Git Worktrees

Git worktrees allow me to open multiple instances of Claude Code, each working on a different feature or task. This setup enables concurrent development without conflicts. While I still manually merge changes, the process is much smoother and more organized.

Conclusion

Building a personal AI factory using tools like Claude Code, Sonnet, and O3 can significantly enhance your productivity and code quality. By focusing on improving inputs rather than patching outputs, you create a self-improving system that grows and adapts over time. This setup is not just about automating tasks but about creating a sustainable development environment.