This is a great anecdote. SOTA models will not provide “engineering” per se, but they will easily double productivity of a product manager that is exploring new product ideas or technologies. They are much more than intelligent auto-complete. I have done more with side projects in the last year than I did in the preceding decade.
Not hours; but days instead of months: porting around 30k lines of legacy livescript project to typescript. Most of the work is in tweaking a prompt for Claude (using Aider) so the porting process is done correctly.
Thankfully it seems like AI is best at automating the most tedious and arguably most useless endeavor in software engineering- rewriting perfectly good code in whatever the language du jour is.
Use Undermind to gather a literature review of a field adjacent to the one you’re working in but with a wealth of information that you don’t yet know.
Use OpenAI to convert a few thousand lines of code from a language you're familiar with to one you’re not, as all the state-of-the-art tools in the field above use that language. Debug all the issues that arise from the impedance mismatch between the languages. Recreate the results from the seminal paper in the field to verify that the code works, and run it on your own problem. Write a stream-of-consciousness post without spell-checking, then throw it into GPT and ask it to fix it.
