AI Coding and Review Tools I Use and How I Use Them (2025-10-06)

AI-powered development assistance has become commonplace, significantly changing our workflows. With various tools emerging, how we choose which tool to use in which situation has become a critical factor affecting productivity.

This article introduces how I utilize and differentiate between AI tools in my daily development work as of October 6, 2025.

Tools for Design and Implementation

During the design and implementation phases, I mainly use two CLI tools in combination. Each has its strengths, so my basic approach is to use them depending on the situation.

Main: OpenAI Codex CLI (gpt-5-codex)

My main implementation tool is OpenAI Codex CLI. I use the latest gpt-5-codex model.

│  Model:            gpt-5-codex (reasoning none, summaries auto) │
│  Approval:         on-request                                   │
│  Sandbox:          workspace-write                              │
│  Agents.md:        AGENTS.md                                    │
│  Account:          (Pro)                                        │

Sub: Claude Code CLI (Sonnet 4.5)

I use Claude Code CLI as a sub-tool. I utilize it when Codex CLI (gpt-5-codex) is unavailable due to rate limits, or when I need to generate human-oriented text.

 Login method: Claude Max Account
 Model: Default Sonnet 4.5 · Smartest model for daily use
 Memory: project (CLAUDE.md)

Review Process After Implementation

After writing code, rather than immediately doing a git commit and push, I perform AI-powered reviews in two stages: locally and on GitHub.

1. Self-Review in Local Environment

Before creating a Pull Request, I perform a self-review that can be easily executed locally. This allows me to discover and fix obvious bugs and improvements early.

codex /review

/review

claude /review & /security-review

/review

/security-review

coderabbit

Hit ↵ to hop into the review

2. Review on GitHub Pull Requests

To ensure quality in team development, I leverage multiple AI review tools on Pull Requests. Each tool reviews from a different perspective, increasing comprehensiveness.

By integrating these tools into the CI/CD pipeline through GitHub Actions, human reviews can now focus on more essential discussions about design and specifications.

Summary

This article introduced my AI utilization approach as of October 6, 2025.

Design & Implementation: Main driver Codex, Sub & text creation support Claude
Review: Combination of rapid local self-checks and multifaceted automated reviews on PRs

AI tools are evolving daily, and this configuration may change again tomorrow. I will continue to explore optimal workflows and improve productivity.

That’s all from the Gemba.