Hello fellow AI coders,
Since I picked up AI coding (i don’t want to call vibe coding) very early, I collected some useful tips/tricks and strategies over the last few years. They are all explained in the video (6 video playlist) below:
Please feel free to comment on the video with your feedback!
Okay, great work! I will definitely watch. Intro was nice already.
But my problem with education like this is, that it is always old. These things evolve very fast. I think the general goal for that development is to make AI as smart and as tolerant for bad human input as possible. So in the end, in my mind, tricks should not be needed.
@Serp I absolutely agree, and some small parts are already out of date probably. However, i have mostly focused on high level strartegy, and how to think about coding with AI on the highest level (knowing full well that the low level stuff will be outdated). 60-80% of the content should be solid at least for a few years more 
Please let me know if you watch it to the end!
Well, in lesson2 where Im now, you speak of context size. I think it got old. I mean it is still problem, but for example Cursor is quite efficient to choose which context to use limiting the need. It can handle 50k line project just fine. In the core, I suppose the problem still exist, but situation between 3 months ago and today is like night and day in my eyes.
I agree 100%. I worry about context size much less these days. You will make me record a version 2 
Good first videos!
Small error: TabTab can indeed kinda read your mind, since it knows ALL of your previous actions - git history, chat, edits .. “if you look a programmer over the shoulder, you can predict the next 30mins”.
Alright, finished all and it was good talk.
Quite in-depth thoughts of work environment and everything, but one thing came to my mind.
Social media. At the moment my workflow is pretty much ask and wait style, where I then go to some in window next to Cursor and see posts. I cannot do anything more serious because the terminal might get stuck and I need to be ready to un-freeze it before the agent timeouts completely.
2 agents in parallel works pretty well - complex stuff gpt-5-fast-high using Cursor/Codex, and fill the pauses with quick Claude code bursts in other modules. E.g. debug/refactor + new UI stuff.