So I'm looking at available AI tools, and it seems like most that they can do is: 1. Provide code snippets using common knowledge, like 'how to build a React component that calls something on click' 2. Generate boilerplate code based on your code input, like 'build a unit test for this function' That's useful, but not a huge improvement over googling for examples, or using IDE templates. And that's only a small part of the actual work. What I would like to know is if there are any tools, that can read a whole codebase, and add features or fix bugs based on prompts? This is where most of the work is done - reading the code, untangling the complexity, changing without regressions, ensuring completeness, and refactoring for simplicity.
> refactoring for simplicity. I've had some success using Chat GPT for this, though it's most useful on Javascript, I'm guessing because that makes up the majority of its training corpus Overall I've had the same experience. I've had some luck using it to automate "busy work" code tasks. For example, I needed to create a handful of basic CRUD rest endpoints. I wrote one endpoint and asked Chat GPT to use that as a guide to generate endpoints for the other entities. The output was say 90% usable and I was able to clean up the bugs.