What did you measure to call it on hard refactors versus autocomplete? The tasks you've named don't sit on the same axis. When a diff comes back from each one, how much of it survives unedited is a different question from how long the tests take to go green. The interesting variable is how much context each one gets to see before it starts.
Adam Lewis
Product Engineer/Architect navigating the AI revolution
What did you measure to call it on hard refactors versus autocomplete? The tasks you've named don't sit on the same axis. When a diff comes back from each one, how much of it survives unedited is a different question from how long the tests take to go green. The interesting variable is how much context each one gets to see before it starts.