Never Lose Context: Thread Documents
From the course: Agentic AI For Founders & Senior Operators
Keep a running thread document for each piece of work, a plain Markdown file that logs everything done and decided, such as our decision to stop using human video editors and edit with HyperFrames and FFmpeg instead. When a new chat session can't work something out, tell it to read the thread so those decisions are in its context.
Transcript
It is just a text file. It's a format of saving a text file. So if you actually open a Markdown file, this is what it looks like at the back end. Let me show you. So I just opened that. And yes, it has... it's basically because of AI that Markdown files are all the rage now. None of us really used to work in Markdown earlier. But at the back end, this is all it is. It's just a bunch of text. So in this case, for instance, I have... and so why this is useful is, I believe... I'm trying to recall. We basically decided to do a bunch of work with Claude for editing videos. And we had basically... I'm trying to look. So this has everything that we've done with the video editor, I think going all the way back to the 24th of July. So originally, we used to work with an editor. And then at some point, we pivoted and we actually logged the decision that we're never going to use an editor again. I'm going to try and look for that very quickly. So these tend to be a little bit unwieldy, but I don't worry about it. Let me see. Yep. So around this time was when we shifted to this thing called HyperFrames and this thing called FFmpeg that we use internally to edit videos and they work beautifully like, we just don't use human editors anymore, unfortunately. And so all of that is logged in this thread. So if I ever go to a chat session and I say, "Edit this video," and it's not able to figure it out, I'm just like, "Go read that thread, pick up everything that we decided, pull that into your context so you know what you're talking about," and then it will work. And then if that doesn't work, sometimes I'll just increase the model effort level and it's able to do it. And the kind of stuff we're able to create now in terms of like actual video is really quite... quite good.