This essay reflects the author’s insights on leveraging AI in coding, emphasizing the importance of developing intuition about when AI is effective and when it isn’t. The author categorizes tasks where AI excels and struggles:
-
Green-field tasks: AI performs well with isolated, complex methods like utilities or base UI components. Providing a function signature helps ensure the AI aligns with the developer’s vision.
-
Pattern expansion: AI can replicate established patterns well, especially when provided with organized examples, making it suitable for tasks like generating endpoints or expanding schemas.
-
Documentation: It effectively writes concise docstrings due to its strength in summarizing complex information.
-
Code explanation: While it offers directionally correct summaries of code, these shouldn’t be taken verbatim.
-
Technical spec reviews: AI can identify potential issues but struggles with document creation due to context specificity.
-
Debugging: AI is less reliable here; it risks offering incorrect solutions in unclear scenarios.
Ultimately, the author advocates for focused task selection over intricate prompting to optimize AI utility in coding.