Your tips are spot on in my opinion. Especially when working with larger code bases, first to understand the codebase with AI (or even ask human beings… somebody might know what is supposed to get what the code is about. Ideally have a recording of the conversation. This way you don’t interpret what you hear). I also let the AI extract coding styles, design patterns. Additionally, let it reflect on how many developers have written the code. What are the pieces which are looking good, which ones are from a junior developer. This alone lets you guide the AI which problems in the codebase might need either fixing or compartmentalization or rewriting. When faced with a new codebase (as in old, but new to me) I now always try to extract that information, put it into development guides and pass these onto Aider, so the LLM (sonnet) will know which coding styles to adhere to and how to interpret code provided to it through aider. Last but not least… THANK YOU FOR YOUR VIDEOS. They are great, informative and coming from experience paired with the AI. Not hype, but realistic. And if you happen to use aider, make sure to hop onto aider discord. You can gain quite some knowledge from just reading along.
Tim, you’re a professional—and, it shows.
Well structured and detailed implementation. It is really good, thanks for sharing.
Thank you for sharing your experiences. ❤
Thank you sir for your videos. I am new to software Development. AI has made it easy for me to start this journey and your videos are 100% honest and helpful i expect more contents from you my mentor. Thank you sir ❤
Great content as usually ! Thanks you!
Good Job, very helpful video, looking for more video on AI codding learning video 🙂
Insightful video! Your final tip about analyzing code bit by bit for improvements is intriguing. It got me thinking about the balance between incremental analysis and comprehensive overviews. In my work on SourceSailor, an open-source CLI tool, I take a different approach by using AI to analyze entire codebases at once. As this approach is useful to give the holistic approach in code-base onboarding. It's interesting to consider how these different methods might complement each other in a developer's toolkit. Though it's not without the issues, especially the large codebase where I can apply this tip verbatim... I'm curious - have you found AI tools particularly useful for either granular or broad code analysis? The rapidly evolving landscape of AI in development is fascinating to follow!
great video!
Well done Tim! What would you consider the average size chunks of code to work on interactively? One function at a time, one file at a time -+ 100 lines? For context I’m 99% working in python. Also what is your favorite stack for agentic frameworks, which is where all my use cases tends to lead based on the relatively unstructured user inputs.
In my previous role we were spending $25k/mo on LLMs, so a 2x or 10x optimization/inefficiency could make or break the business. 🙂
Thank you for the lessons. I have a question. When I want to use a fairly new framework how can I give the documentation of such framework to aider?
@AbhijitKrJha