Tips for Working with AI Jr Dev

Working effectively with an AI agent like AI Jr Dev involves clear communication and providing the right context. Here are some tips:

Provide Necessary Context

Mention Relevant Files: Mention files that may need to be written or modified. I’m a lot better at my job when I know where to look, even if I can sometimes find related files myself. If I need the full content, I’ll ask.
Specify Goals Clearly: State the desired outcome or the problem you’re trying to solve, not just the specific code you think needs changing.
Mention Constraints: If there are specific requirements, limitations (e.g., “don’t use external libraries,” “must be compatible with Python 3.8”), or style guides, mention them upfront.

Be Specific: Vague requests lead to vague results. Instead of “Improve this code,” try “Refactor this function to reduce complexity” or “Add error handling for file not found.”
Break Down Large Tasks: For complex features, break them down into smaller, manageable steps or requests.
Iterate: Don’t expect perfection on the first try. Review the AI’s suggestions, provide feedback, and ask for revisions. Use prompts like “That’s close, but can you also…” or “Please modify the previous suggestion to…”

Boilerplate Code: Generating standard code structures (e.g., classes, functions, HTML layouts).
Refactoring: Improving existing code based on specific instructions (e.g., renaming variables, extracting methods).
Adding Features (with context): Implementing well-defined features when provided with the necessary files and clear requirements.
Writing Tests: Generating unit tests for existing functions.
Translating Code: Converting code snippets between languages (accuracy may vary).

Highly Abstract Concepts: Designing complex system architecture from scratch without significant guidance.
Debugging Complex Issues: While it can help, debugging often requires deep understanding and context that might be hard to convey fully.
Subjective Style Preferences: Without a defined style guide, matching nuanced personal preferences can be difficult.
Tasks Requiring External Knowledge: Anything requiring real-time data, accessing external systems, or knowledge beyond its training data cutoff.
Very Large Codebases: Understanding the interplay of components across many files it hasn’t seen can be challenging. Provide relevant snippets.