Claude Code, GPT-5, and Practical AI Collaboration in Engineering Written by @Jeong JaeSoon

It’s been quite some time since AI coding assistants started blending into real-world software development.

Yet, the way each engineer actually uses them varies a lot.

Personally, I’ve built an AI-centric development workflow around Claude Code, combined with tools like MCP (Multi-Context Protocol), gh CLI, Slack MCP, Atlassian MCP, Context7, and Devin Wiki.

In this post, I’ll share how I use Claude Code and GPT-5 in my daily engineering work — and the insights I’ve gained while designing a practical AI-driven development environment.

⚙️ Setting Up: Prompt Engineering in Practice

Since I studied prompt engineering early on, I rarely rely on external prompts or OSS tools.

Instead, I design the prompt structure I need and let Claude Code handle everything end-to-end.

I always maintain a clear structure:

• Role – Define the agent’s responsibility
• Goal – Specify the target outcome
• Instructions – Provide concrete steps
• Precaution – Include constraints or warnings
• /output-style – Usually set to Explanatory

With Claude Sonnet 4.x, context precision became far more important, so I explicitly include which reference materials or repositories the model should use.