Key Points

Introduction to AI-assisted Coding

AI coding tools range from autocomplete to highly autonomous development.
Higher AI autonomy requires greater human oversight and review.
Established software engineering practices remain essential when using AI.
AI can speed up development, learning, prototyping, and documentation.
AI-generated code may be incorrect, insecure, or poorly designed.
Developers remain responsible for validating all AI outputs.
Over-reliance on AI can increase technical debt and reduce understanding.
“Almost right” AI solutions often increase debugging and maintenance effort.
AI tools are evolving rapidly, and best practices are still emerging.
The Copilot free tier currently includes access to three AI models each with a different balance of speed and purpose.
The Copilot free tier currently allows 2000 completions and 50 premium requests per month.

The example Python code generates a basic set of statistical plots for all data files within the data/ directory.
The code has a number of deliberate issues that we want to resolve throughout this lesson using GitHub Copilot.

Configure Copilot privacy settings to disable public code suggestions and data-sharing before use.
Use Ask mode to explore and understand an unfamiliar codebase without making changes.
Copilot responses are probabilistic and non-deterministic, with the same question yielding different answers.
Write specific, concise prompts and break complex questions into separate steps for better results.
Use # to add explicit context (such as files or folders) to chat prompts.
A .github/copilot-instructions.md file provides persistent project-level guidance to Copilot for things like coding style and conventions.
Always critically evaluate AI-generated explanations and code, since they can be confidently incorrect.
Monitor your Copilot quota, especially when using higher-cost models (marked with multipliers greater than 1x).

Copilot offers three levels of refactoring support: inline suggestions, agent mode, and plan mode, in ascending order of autonomy and scope.
Inline suggestions appear as ghosted text while typing and are driven by immediate code context.
Agent mode can make targeted edits across multiple lines or functions when given a specific instruction.
Plan mode analyses the codebase and produces a step-by-step refactoring plan without making any changes, enabling review before implementation.
AI-generated suggestions should always be reviewed and understood before accepting; increasing autonomy requires increasing scrutiny.