Coding with Claude: Pair-Program with AI and Review What It Writes

Set up a safe workflow where you stay the developer and AI is the assistant.

• What AI is good at in code: boilerplate, explanations, ideas, tests, refactoring
• Where it fails: made-up functions and libraries, outdated code, hidden bugs
• Plan first, small steps, run everything, read everything
• Giving context: language, version, files, goal, limits
• Why you must be able to understand and explain every line you keep

ClaudeVS CodePython

Practice: Build a small command-line tool step by step with AI, keeping a log of what you accepted or rejected

Use AI to understand unfamiliar code and then verify the explanation.

• Asking for line-by-line and big-picture explanations
• Reading open-source code with AI support
• Checking explanations by adding prints, running examples and reading docs
• Learning mode: ask for hints before answers

ClaudePythonVS Code

Practice: Choose a small open-source file, explain it using AI and then confirm each claim by running the code

Find and fix bugs faster by giving AI the right information, and learn the cause, not just the fix.

• Reading error messages and tracebacks
• A good bug report: expected result, actual result, code, what you tried
• Rubber-duck debugging with AI
• Verifying the fix: reproduce first, fix, test again
• When AI loops or guesses: step back and think

ClaudePython debuggerpytest

Practice: Fix five planted bugs and write a one-line cause for each

Create useful tests with AI and make sure the tests themselves are correct.

• Unit tests with pytest
• Asking for edge cases and invalid inputs
• Tests that pass for the wrong reason
• Test-first workflow: write tests, then ask AI for code
• Coverage as a hint, not a goal

pytestClaude

Practice: Add a test suite to an earlier project and break the code on purpose to confirm tests catch it

Improve code safely in small steps while keeping behaviour the same.

• Naming, functions, duplication and simple structure
• Refactor only when tests exist
• Ask for options and trade-offs, not one big rewrite
• Comments and docstrings that help future you
• Comparing AI's version with yours

ClaudePythonpytest

Practice: Refactor a 150-line script into small functions with tests and a short before-and-after note

Keep a clean history and stay in control when AI changes your code.

• Small commits and clear messages
• Branches for AI experiments
• Reading a diff before you accept it
• Undoing mistakes safely
• Pull requests with an honest description of what AI helped with

GitGitHub

Practice: A GitHub repo whose history shows each AI-assisted change, reviewed by you

Catch bugs and security risks in AI code before they reach real users.

• A review checklist: correctness, edge cases, readability, performance
• Checking that packages and functions really exist
• Security basics: never paste API keys, passwords or customer data into prompts
• Environment variables, .gitignore and secret scanning habits
• Input validation, SQL injection and unsafe code as examples to look for
• Licences and where code comes from

GitClaudePython

Practice: Review a piece of AI-written code using the checklist and write up three issues and fixes

Try agent-style and editor-based coding assistants carefully and know when each helps.

• Chat assistants versus tools that work in your editor or terminal, including Claude Code
• Other assistants such as GitHub Copilot or Cursor, compared honestly. Features change, so check current ones
• Permissions: what you allow an AI tool to read, run and change
• Working on a real project: small tasks, reviews, tests after each task
• Final project and next steps

Claude CodeVS CodeGit

Practice: Final project: build a small app with an AI assistant, with tests, clean Git history and a review note