I've been working on my PKM/Browser extension app for several years, adding features, hand rolling everything in plain JS/HTML/CSS and evolving my style and conventions over time. It's becoming harder to make progress and the output I get from AI is often not great - causing side effects, propagating older conventions or just plain missing required interactions. I want to modularize, standardize, modernize, invert some dependencies, add some facades etc. The goal is improve my velocity and the reliability of AI output, and to subsequently transition to a more AI-first development model.
I have a $10/mo Copilot license and access to the models it provides, with associated opaque limits on context and queries. I transitioned to working in VSCode to get the ai integration and to join the cool kids, but historically I've lived in emacs as much as possible. I'm working on a design.org text file giving an overview of the general structure, data flows etc. and have an initial work breakdown that I intend to feed into the context. I've previously found unit tests too cumbersome to maintain and not worth the time on a one person project.
I'm looking for all and any tips and tricks. Whats the best current model to use for a project like this? How should I think about preserving context over time? What do I do when I'm shut out for some time cos I've reached a limit. Automated tests are tricky given the complications and limitations of an extension interacting with web pages, how should I think about testing? As I'm at it are there non obvious things I should keep in mind to make the codebase more aligned with an AI-first development model? Anything else to keep in mind?
0 comments
—
Ask HN: ~25KLOC Indie project. Best practices for AI facilitated refactor? · HackerTrans
I have a $10/mo Copilot license and access to the models it provides, with associated opaque limits on context and queries. I transitioned to working in VSCode to get the ai integration and to join the cool kids, but historically I've lived in emacs as much as possible. I'm working on a design.org text file giving an overview of the general structure, data flows etc. and have an initial work breakdown that I intend to feed into the context. I've previously found unit tests too cumbersome to maintain and not worth the time on a one person project.
I'm looking for all and any tips and tricks. Whats the best current model to use for a project like this? How should I think about preserving context over time? What do I do when I'm shut out for some time cos I've reached a limit. Automated tests are tricky given the complications and limitations of an extension interacting with web pages, how should I think about testing? As I'm at it are there non obvious things I should keep in mind to make the codebase more aligned with an AI-first development model? Anything else to keep in mind?