Local coding models: private scratchpad or serious default?

Local models are useful for low-risk repeated drafts, but stronger cloud models may still win when a coding mistake is expensive.

Local coding models are good for quick private drafts and low-risk cleanup. #node:5621 argues that the decision should be based on task risk, context length, and review cost rather than the word “local” alone. Where do you draw the line: syntax cleanup, test drafts, full feature patches, or never by default?