Feedback / Suggestions / Concept Ideas
Features/New Tools (for the purpose of making Cody AI a tool that would be useful not only for developers but for everyone):
- Create a new AI search tool similar to Perplexity AI and Phind AI. The reason for this suggestion is that you could reach a lot more people and be helpful to a wider audience if it was available via a search, just like a Google search. Imagine if people started changing their default search engine to Cody AI. I really think it could be done with your team very easily.
Cody AI Suggestions:
-
Make Cody AI better at writing cody.json prompts and improving the prompts you already have in your cody.json. Yes, the codyception thing is great, but I constantly am trying to improve my prompts in cody.json, usually by hand, because Cody AI always messes up the formatting in my cody.json when trying to edit a particular prompt.
-
A lot of times, I’ve been trying to add a large chunk of Java code at once, basically the whole Java class. When I try this, usually it cuts off the end of my Java class. I realize there’s a content window and you can’t exceed the limit. However, I’ve observed a couple of other tools, like the Phind VS Code extension, that edit one line at a time, changing that line then moving on to the next line, and so on, changing line by line until it gets through the entire Java class. I will mention that Phind is not the greatest at actually writing good code, but the line-by-line editing approach seems to work very well and never cuts off the end of my Java classes. I feel like Cody would be way more efficient and powerful if it edited one single line or one single method from the selected code at a time (edit the first method correctly and then move on to the next method).
Additionally, another cool way you can make sure that the code Cody AI writes is exceptionally well-written is to have the AI go through your code before editing and, above each line of code that needs to be changed to implement the changes, have Cody AI write a comment line. For example:
//this method needs a try-catch block
(example method here)
Cody would be very efficient and productive if it went through and made comments within the code before making changes, then asking the user if they’re happy with the suggested changes in the comments. If the user says proceed, Cody AI will implement the changes and remove the comments.
AI Cody GitHub Client Concept Idea:
You guys should make an AI GitHub client. It would be really cool if you made a tool that can actually go out and touch your repositories and make changes at your request. I know that’s a dangerous topic, but if you made a tool that could do the following, I think a lot of people would pay for a premium subscription for this purpose alone.
AI GitHub Client Features:
This AI GitHub client should have the ability to “fork & create a new branch” of the specified repository, make the changes requested by the user, and when Cody AI believes all changes are implemented properly, double-check their work by having a second AI briefly fact-check, making sure that the changes are properly implemented and working as intended. Show the user a git diff with the changes (brief overview). Here, the user is asked if they are happy and want to proceed or decline. If the user is happy with the changes, a new pull request is created on their repository by Cody AI for review.