A Software Workflow with Claude Opus, Codex CLI and Gemini CLI

In a software project, architectural decisions, code generation and infrastructure operations each come with different requirements. Rather than assigning everything to a single AI tool, dividing tasks according to these requirements can offer a more balanced way of working. At X Mind Solutions, we explore this approach using Claude Opus, Codex CLI and Gemini CLI. Our focus is on establishing a clear connection between the project's planning, development and operational stages while keeping a limited token budget in mind.

In this workflow, we assign Claude Opus the role of strategic planning. We first examine the project's overall structure, define its main functions and clarify which components are needed. This stage also covers how the components will fit together and how the workflow will be structured. Instead of starting code generation with a vague idea, we establish an architectural framework that can be translated into concrete development tasks. We also use Opus to reassess the generated code.

During code generation, we pass clearly scoped tasks to Codex CLI. Specific definitions, such as what a function should do or how an API endpoint should behave, form the basis of this stage. We test the resulting code and refine it through revisions when needed. One reason we choose Codex CLI in this setup is to keep costs balanced. However, generating the code does not mean the work is complete; testing and evaluation remain part of the workflow.

On the operational side, Gemini CLI and MCP connections come into play. At this layer, we handle tasks such as creating branches and managing pull requests on GitHub, deploying applications to a VPS, and configuring DNS and CDN settings through Cloudflare. This division of roles does not position Gemini CLI as a replacement for architectural decisions or code generation, but connects it to the infrastructure steps that follow development. We organize operational tasks as a separate area of responsibility.

An example API development workflow illustrates this division of tasks. First, we design the API architecture with Opus. We then assign Codex CLI the task of writing and testing a specific endpoint. We evaluate the generated code through tests and use Opus for another quality review. Next, we push the code to GitHub and deploy it to a VPS through Gemini CLI. The output of each stage becomes a concrete input for the next tool to work with.

The core of this approach is not to increase the number of tools, but to give each tool a specific responsibility. By using Opus for architectural thinking, Codex CLI for code generation and Gemini CLI for operational connections, we distribute token usage across tasks. We see this not as providing unlimited resources, but as a way to make more balanced use of the available budget. This division of work, which we also discussed at Claude Code Meetup Istanbul, shows how everyday development practices can provide a concrete basis for discussion and mutual learning.