100% agree on the creative side, but as a coding assistant, ChapGPT and Claude 2 are very handy. An example is “help me code an in-memory graph data store” and you’ll get solid code with edges and nodes. They are great at writing unit tests. Boilerplate code gen is like tiny hyperspace jumps.

I’ve heard from multiple coders that it works “sorta okay” in that context. Fair enough! But you really can’t equate solving a constrained problem (code that does X) and the unbounded problem of creating good writing / art