I spent the last six months on a project, converting dataweave applications to nodejs. Set up an empty nodejs repo w/skeleton code. Take the existing code, removing PII/password anything, and shoving it through Claude was the next step -- transformers were MADE for translation, and it did it's job wonderfully.
Then I had other realities to deal with - they changed the query engine to use JSON instead of XML, and there were already made libraries for interfacing with APIs for data fetching/storing with multiple services (think workday, box, etc). If I would have considered this earlier on in the project, I could have saved myself a lot of work with better prompting, but alas.
5% of my time was setting up the code for claude and pasting the results back
5% was setting up the repos
20% was spent on rewriting the little details that I didn't tell claude or that claude missed/glossed over
and 70% for testing/verification against existing systems.
So that's my real world example from the last six months - I'd love to hear from others
Then I had other realities to deal with - they changed the query engine to use JSON instead of XML, and there were already made libraries for interfacing with APIs for data fetching/storing with multiple services (think workday, box, etc). If I would have considered this earlier on in the project, I could have saved myself a lot of work with better prompting, but alas.
5% of my time was setting up the code for claude and pasting the results back 5% was setting up the repos 20% was spent on rewriting the little details that I didn't tell claude or that claude missed/glossed over
and 70% for testing/verification against existing systems.
So that's my real world example from the last six months - I'd love to hear from others
reply