Refactor a Legacy ETL Codebase With Cursor

Cursor is a fork of VS Code, so if you already use VS Code, the transition is mostly familiar.

1. Download and install Cursor from the vendor's site.
2. Sign up for the free Hobby tier first, no credit card required, to confirm the editor works with your existing extensions and workflow.
3. Open your ETL codebase as a project folder.

What you should see: An editor that looks and largely behaves like VS Code, with your existing extensions available if they're compatible.

Troubleshooting: If your company's device management software blocks installing new applications, check with IT before assuming Cursor isn't allowed. Some orgs already have it on an approved list for engineering roles.

1. If you've confirmed the editor works for you, upgrade from Hobby to a paid tier.
2. This step matters specifically for this use case: the free tier restricts AI context to the current file, which defeats the purpose of a codebase-wide refactor tool.

1. Once on a paid plan with the repo open, Cursor indexes the codebase so its AI features can reference code across files, not just what's open.
2. Wait for indexing to complete before starting a serious refactor task. A partial index means partial context, which produces worse suggestions.

What you should see: An indexing status indicator, and once complete, AI features that can reference functions and files you haven't explicitly opened.

1. Before asking the AI to analyze anything, check whether your repo has a way to tell the editor which files to ignore when indexing (most AI-enabled editors support some form of ignore file for this). Exclude credential files, .env files, and any config containing real connection strings or secrets.
2. Confirm this exclusion is actually working by checking whether the AI can "see" a piece of content from an excluded file before you trust the boundary.

Why this matters: A legacy ETL codebase is exactly the kind of place where an old credential or hardcoded connection string might be sitting in a file nobody's touched in years. Indexing the whole codebase means that content becomes visible to the AI unless you've explicitly excluded it.

1. Open the AI chat or multi-file editing feature (often called Composer or Agent mode, depending on your version) and describe the refactor goal in plain language: what's inconsistent, what pattern you want the codebase to converge on.
2. Ask for a plan, not code changes yet. Review the proposed approach, the order it suggests tackling files in, and whether it's identified the same pain points you have.

What you should see: A structured plan referencing specific files and functions in your actual codebase, not generic refactoring advice.

1. Work through the plan one file or one small group of related files at a time.
2. Review every AI-suggested change before accepting it. Don't accept a batch of changes across many files without reading each one.
3. After each file or small batch, run your test suite before moving to the next.

Troubleshooting: If the AI suggests a change that breaks a pattern used elsewhere in the codebase (a naming convention, an error handling approach), point that out explicitly in your next prompt. It's working from what it can see, and a codebase with genuinely inconsistent patterns gives it mixed signals to learn from.