Using Claude Code or Codex as a Teacher, Not an Autopilot

Using Claude Code or Codex to enhance your coding skills requires a different approach than simply relying on the tool to do the work for you. It's about treating the AI as a thought partner that can deepen your understanding, identify gaps in your knowledge, and guide you toward independent problem-solving. Here are four strategies to transform your AI coding assistant into a valuable mentor rather than a shortcut:

1. Let the AI ask you questions instead of requesting code right away. If you're working on a dashboard that fetches metrics, prompt the AI to ask you about the API call placement, handling loading and error states, and why TypeScript types alone aren't sufficient for real-world data validation. This approach forces you to articulate your current understanding—and reveals what you're still unsure about—before diving into any code.

2. Instead of immediately asking for a complete solution when you encounter an error, request a series of progressive hints. Start with a guiding question, move on to a conceptual hint, provide a minimal example, and only finally offer the correction if needed. This scaffolding ensures that help arrives only after you've exhausted your own problem-solving attempts, thereby reinforcing your ability to tackle challenges independently.

3. Share your imperfect code with the AI and ask it to rephrase what your code is attempting to do, point out what's correct, highlight the two most significant risks, and suggest a question that will lead you to the next step in fixing the issue. This process turns your mistakes into learning opportunities. You'll likely discover fundamental issues such as the potential for fetch to fail, the need for proper handling of HTTP errors, and the fact that response.json() returns unvalidated external data—lessons that are more impactful when you uncover them through reflection rather than immediate correction.

4. Encourage the AI to compare trade-offs rather than simply presenting one solution as definitive. Ask it to generate two approaches for fetching metrics, such as doing so directly within a server page versus through an API route, and then compare these options based on factors like simplicity, security, caching, testability, and error handling.

The aim is not to select the most complex or "fancy" solution, but to gain insight into why one approach might be more suitable for your particular context than the other. The goal isn't to avoid using AI, but to leverage it in a way that strengthens your judgment and decision-making skills. Over time, your prompts will become more precise, your understanding more profound, and your reliance on the AI more intentional.

You'll learn to recognize when the AI is guiding you correctly—and when it may be overlooking crucial aspects.