Why Manage Project Context Via Epics

Note: This is an AI-assisted transcript and is not a verbatim or accessibility-certified transcript. There are some edits in the text to improve readibility. Text in square brackets, [], are in the video but are not needed for reading, so kept in brackets.

Hey, Sumit on this side. Good afternoon. I started off the second part yesterday, and I kind of realized that without having context as to why we should be avoiding some of these things or why we should be structuring some of these things in this way that I am showing, it might not be clear as to what are the benefits. Interestingly enough, something happened today which I feel like would set this context even more strongly as in why we should maintain context, why we should break it into different files, this structure that I showed with regards to epics and maybe tasks or some kind of stories in between.

Of course, around epics, tasks, stories, all of these concepts, I think you will figure out a pattern that works best for you. I'm not saying that you have to stick to this pattern. In fact, I don't even create stories. Like I said, the concept of epics, stories, and then tasks is very, I would say, quite popular, quite traditional when it comes to software management, software development, or in general project management. In probably many different domains, it is followed. But I'm not saying that you have to follow that strictly. I'm just saying that you have to break the structure, break the entire flow of what you want to build into something that you can go back. And this is why it is important. Let me show you an example.

This happened today itself. I wanted to build this. Let me show you also what I'm trying to build, so maybe that sets the picture very clearly. I wanted to build this page called Concepts. The reason for that is now that I have four posts, as you probably already know, my flow is very simple. I record these videos, and then I transcribe them with AI. So I usually use a mix of either Mimo, one of those large language models, or Eleven Labs. And then what I do is I get the text out of it. I then go ahead and correct a little bit, but I do not change the structure, break the entire flow of what you want to build into something that you can go back. And this is why it is important. Let me show you an example. This happened today itself.

What would really help is if I can extract some of these concepts as I have recorded. Because my recordings are, most of the times I don't plan — I don't have a very long-term plan as to how many such recordings, how many such parts, videos, etc. I'll be creating. So I know the concepts that I want to talk about, and I kind of pick a concept and just go with it for that one video. And then the next day I know something is important, and that is exactly what you will see happen today.

So yesterday this video that I posted, the one where I talked about how to break down context, what happened is when I now wanted to create this sidebar. So this is what I was going for. I wanted to basically extract concepts from within any post and put it on the sidebar. That's simple. That's it. That's simple. It's not complicated.

But what I wanted to do was I wanted to also have in the software. So when you go to the software that is running behind prompttogether, so that is hosted on GitHub. That is the raw software, not the website, not the website as you see it, but the actual software that runs it, which is a few other tools to help me, like for example, take the video and transcribe it to text, etc., etc. I wanted also a tool around this concept building. Of course, I'll get to that. I've not created the tool, but I wanted to create first, I wanted to create a first impression, like how does it look like to extract some of these concepts out of a post and just stick it in onto the sidebar, and also have a separate page.

So here is the prompt. This is just for context. Here is the prompt. So the prompt is, let's scroll up. So read agents and PRD for reference. This is, of course, if you have watched my videos, you already know that I have this as just copy paste friendly lines. I have a set of these depending on the project. Some are more Claude based. Some are now — I mean, even Claude now supports agents.md, so I don't really need a Claude.md reference anymore. Again, the agents and the Claude.md, they are supposed to be read automatically by Claude Code, Codex, Open Code, or any of these softwares. But I kind of had this habit from a long time before, a long time back, and so I kind of kept with it. But yeah, sure, you can also not refer to it. What is important is the PRD.md, the PRD.md, the markdown, which is — which has my project's requirements, which is my document for project requirement, and that is important. So it's just both of them in one line. I copy, I paste. So that's how I start.

And then what I'm saying is that I've been creating videos. I'm just giving it a bit of an explanation, and this is where I have referred to an existing epic. This epic has basically the documentation about that software, that small tool that I had created already, where I would take a video and transcribe it into text. That's how the content, the text content on this website gets created. And I just referred to it. I said I've been creating videos on, and then there are four posts, just to give it reference to what are the four posts. And then I said that I want to extract the concepts, the important concepts, and build a structure which makes sense, and that's what I explained.

What it did was I said take examples as Claude Code, Codex, agents, .md. These should become highlighted in the text, and they should also become a separate concepts page, which is what it has created. There is a separate concepts page, and if you go into a post, for example, this one, you'll see that on the sidebar now on this right hand side, you will see it nicely, which is great.

But where it falls apart, or where it will fall apart, is now let's say I come back to this two days later, and I want to build something on top. Let's say I want to extend this idea, and I want to make it either functionally different or visually different. Then it kind of falls apart because now I have lost track of what I told Claude Code. Because what Claude Code did, my expectation out of explaining this, by the way, you can pause the video and read this prompt right now if you want to, but what will happen is that if I come back to this a few days later, I will have forgotten my prompt, which is obvious, and Claude will have no way, or Codex, or any other such coding agent will have no way to figure out what was the user's direct thought to create this structural change in the code. Why did this change happen? And that's what we are going to miss when we come back to this.

Let's say, and I'll give you a very specific example. Let's say I want to add another concept from these blog posts. Let's say I'm reading them, I'm reviewing them, I'm like, hmm, you know what? OpenAI should also be highlighted, or maybe...

Harness should be highlighted. Now, what I have to do, and I'll show you the code that it has generated. Again, we are not supposed to look into code. That's the whole point. But just for the sake of this particular video, what it has done is it has actually created the table. This is in a database table. This database will live locally on my computer. This is SQLite. And what it has done is it has created concepts, which is all good. Again, don't want to deal with the technicalities right now.

But I just want to show you that it has actually put in the actual concepts one by one directly. It has generated this code, which has the concepts, which is just these few concepts, and that's the key part. So it's not extendable by me. How can I come back to it? Let's say I create another post and now I'm stuck. How do I increase? How do I rerun this software? Should I copy paste that entire prompt and create it again? No. But my structure is already there. It should not have had to go through this database file and create it again. No, that's not the point.

I should instead maybe have another tool in my software which would enable me to, hey, here is a new blog post. Let's say this particular video that I'm recording, which will become a new post. Hey, here are the concepts which I feel are important and a few other concepts maybe I missed in the past. Could be that I'm pointing out to different blog posts, which is totally fine. But as long as I can get to say, hey, take the blog post XYZ. Let's say this particular blog post becoming Agent Enable Software Builder, and I want to extract or I want to create now the concepts of OpenAI and Harness. So go ahead, check the internet, find the meaning of these, put this in the database wherever it needs to be, and make sure that the page then shows it on the sidebar.

So this now becomes a repeatable task. That's the key insight here. Anything that you think is repeatable is going to be part of your workflow, and this happens all the time in all sorts of businesses. Whenever an email is coming and I'm labeling it for some reason, maybe this customer is important, this is high value, this is low. Those are repeatable things. If I'm doing something on a weekly, daily, hourly, whatever basis, even if it is once a month, it's still repeatable. And repeatable things need their own tools because I want to go back to it and be able to do it again and again without restarting the whole prompt journey from scratch. I don't want to do that.

So in that case, what I would like to do is I would like to ask Claude Code, hey, build me a tool around this. And this, again, the showing the code is probably the most horrible thing I'm doing sometimes, but just go along with me for this. I already have a tool which converts the video to a markdown, and this is a CLI. CLI is a command line interface. Basically, it's a code. It's a piece of code which I can run on my computer. Let me show it to you.

So here is a simple example of how I have run this in the past. What I would do is I would go here and I would, just showing you. So video to markdown. I point it to my video and I have Eleven Labs as fallback. What it would do is it would go ahead and it would use this video file, extract it. By the way, sorry, I'm not sure you are seeing this in the video. Hopefully now you are seeing. So this is a command. This then what it does is this command is built into my project Prompt Together.

So the video to markdown is a command I have asked either Claude or Codex, I forgot which one, to build. And here is an epic for it. The epic is here. Video to markdown CLI. What's the benefit? Well, the benefit I get out of this is that the epic is something I can go back to because that was the thinking. That is how I had thought about what I wanted to do, why I wanted to have this process at all. And I can go back to a coding agent and say, hey, here is an epic, which is how I got this created in the first place. Now go ahead and modify. Please create another epic for it before you modify. Let's say I want some updates or whatever changes. Please create an epic for it and then go ahead and give me new updates or create the code that is needed.

But in the case of this particular thing that I'm talking about, which is this sidebar and the concepts being extracted, Claude Code did not create an epic, which means now I am left in a situation. I will, by the way, get the epic done. What I will generally do, and this is probably also important for this video, I will go ahead and ask. I will go ahead and ask what I can do. I will not press enter on this prompt because I have a few other things to ask of it, but I'll show you. Create a retrospective epic of this work so that I can continue making.

Further requests with coding agents in future. That line is good enough. It knows, or it'll figure out. I can say, by the way, Apex folder has Apex. Okay, that's good enough. Now that will create a retrospective epic. Retrospective in this case, I'm not sure if that is the best word to use, but basically the work has been done. So this is like thinking about it retrospectively, building an epic where, hey, this has already been built. I am just creating it for documentation purposes. So Claude will happily build this. But my way, generally speaking, and today I made the mistake, which is why I wanted to record this video, is that I usually get the epic created first and then go ahead and do the implementation. And I don't look at the implementation code, by the way. I usually run the command and I'm done with it. If the command doesn't work, I'll probably ask for more changes. But I don't look at the code. But the epic is still needed because the epic is how my thoughts kind of get locked. They get recorded so I can come back to them. Hopefully this helps a lot. Because again, this is part, like I said, it's not very well planned, but it's all part of as I see things, I record them, document them, build your own mental vocabulary out of all of these things, which is why the sidebar, the concepts, that's becoming key. Hopefully you will be able to even take these sidebars and the concepts and save it. We'll see if I can add those features. But the whole idea is for you to take those concepts and build software the way you want, find the workflow that works the best for you. Thanks.