AI Terminal Workflows

Overview

This project explores how modern AI tools can be used directly from the terminal to support real development work.

Rather than treating AI as a single web app, I wanted to understand how multiple tools behave, how they differ, and how they fit into an actual command-line workflow.

What I built

• Installed and tested multiple AI CLIs:
  • Claude Code
  • Codex
  • Gemini CLI
• Used Homebrew to manage versions and upgrades
• Created test repositories to experiment with:
  • Project-level instruction files
  • Prompt styles and constraints
  • Tool-specific strengths and weaknesses

How I worked

• Built a dedicated testing workspace
• Added project instruction files like CLAUDE.md to guide tool behavior
• Compared:
  • How each tool interprets instructions
  • Code quality and reasoning
  • How well each fits into real CLI-driven work

What I learned

• AI tools behave very differently depending on how they are guided
• Repo-local instruction files dramatically improve consistency
• Some tools are better for:
  • exploration and reasoning
  • others for fast code generation
• The best results come from treating AI as:
  • a collaborator
  • not a replacement

Why it matters

AI is becoming part of daily technical work.
Understanding how to integrate it thoughtfully — with structure, constraints, and intent — matters more than simply “using AI.”

This project shows how AI can become part of a real, human-centered development workflow.