Most AI coding demos work because the project is empty. The interesting question is what happens in a repository with ten years of history and no tests.
These are the tools that hold up there, sorted by the job they do best.
Claude Code
Reads the repository, runs commands and makes multi-file changes; the strongest option for real refactors.
claude.com/claude-codeGitHub Copilot
Still the best inline completion, and it is where most teams already have a licence.
github.com/features/copilotCursor
VS Code rebuilt around chat and multi-file edits, with the extensions you already use.
cursor.comContinue
Brings an assistant into VS Code or JetBrains using whichever model you choose, including local ones.
continue.devAider
Pairs with you from the command line and commits each change separately, which makes review easy.
aider.chatThe tools are not the bottleneck any more: reviewing what they produce is. Keep the diffs small enough to actually read.
Need it built, not just chosen?
Web apps, APIs, infrastructure and interfaces. Freelance from Barcelona, working with clients anywhere.
Leave a Reply