Prashant Dumasia

Hi, #14 is the one for me. I've kind of fallen down in a rabbit hole on this exact thing, you finally get retrieval working nicely, the AI agents looking at all the right files, and then the code it gives you still feels off. Wrong patterns, ignores the stuff the rest of the repo does without thinking. I keep coming running into the same thing... what the model sees and what it actually gets about the database are two totally different problems. Retrieval nails the first one. But getting it to actually respect how your code is built? That's a whole other thing, and honestly it has eaten most of my time lately. Anyone else stuck on this, or is it just me?

Kevin Hwang Great work, thanks for sharing this. We came across the same issue that Sunny Choudhary raised, and we have built an approach to solve it. If you're okay with it, I'd love to share it here. Thanks!

Thanks for sharing this.

Though we are moving at a fast pace towards AI agents, I personally feel that AI agents could be used as assistants rather than replacing humans. As we know, sometimes AI agents start hallucinating. That is where engineers are needed the most—to check and rectify issues before shipping. OR, we could come up with something that would help AI agents perform better without hallucinating.