GitHub Copilot

Copilot greys in the code it thinks follows your cursor. It doesn’t roam a project the way an agent does, but by sheer frequency of small time savings it’s the form of assistance that helps most often.

Install link

Install the GitHub Copilot extension from the VS Code marketplace and sign in with your GitHub account. Plugins exist for JetBrains, Neovim, and Visual Studio too.

  # Neovim
git clone https://github.com/github/copilot.vim \
  ~/.config/nvim/pack/github/start/copilot.vim
# then run :Copilot setup
  

Basic controls link

Getting better suggestions link

Copilot predicts from surrounding context, so the clearer that is, the more accurate it gets.

1. Write the intent as a comment first.

     # Read the CSV, parse the date column, drop rows with missing values
   def load_sales(path: str) -> pd.DataFrame:
     

2. Write the signature and types first. A declared return type sharply improves the body it produces.
3. Keep related files open. Open tabs are used as context.
4. Name things precisely. parseInvoiceRow draws better suggestions than handleData.

Where it shines link

• Writing test cases (write one and it continues the pattern)
• Repetitive mapping and transformation code
• Regexes and date format strings — the things nobody memorises
• Doc comments (JSDoc, docstrings)

Cautions link

• Plausible wrong answers happen, especially invented API signatures. For an unfamiliar library, check the docs.
• Don’t ship security code unreviewed. Authentication, encryption, and permission checks always get a human read.
• Check your organisation’s policy. Suggestion filtering or usage limits may apply to private code.

Alongside Claude Code link

The two barely overlap.

Next link

For an editor built around AI from the start → Cursor