Juggle 10+ projects in parallel without getting overwhelmed

I'm going to be talking, about bash, little bit of bash, Python, and Go. Actively using AI to build things to to to use AI to develop things. How many? Raise hands. Okay. Claude Code or others? What others are you guys using? Cursor and Anti clarity. NewClarity. Hi. Hi. And so are you using Claude with these things or using other agents? I don't. Have you are you any of you using hooks with Claude or with any of the a few of you. Have you building hooks? How about MCPs? I assume most most of you are using MCPs. Are you building MCPs? Using but not working. Okay. Alright. So, I didn't really wanna have slides because that wasn't supposed to be the thing, but but if I don't have something to go from, I'm going to I'm gonna forget. So, what's the problem that I wanna talk about here? Problem is about, I've been, working with AI for about 2 plus years now, which is, you know, anybody who started from the beginning was about 2 plus years ago. When I say that, I don't mean building LOMs. I mean, using it AI to help me write code. And then I've been doing it to help me write Go code. And I helped Deep tight control of the code because it kept doing stuff I didn't like. And then about 2026, beginning of 2026, it just started Cloud Code just got really good, and I just gave up of holding on to control. So I started doing lots of planning for what I wanted to to have it do and just let Cloud build. But that turned into lots and lots of waiting. You know, we spend 10, 15 minutes doing planning and then wait 2, 3, 4, 5, 6, 7, 8 minutes for it to build something. I've never been somebody who could do parallel projects, but I started thinking, you know, there's LLC these round 2 Project I have. All these things I wanna work on. So I started doing them on the side as I was working on the main project. And it's kind of the point to where I have well over 10 projects concurrently. And And a lot of those projects are because I'll be working on a Project, and I realize I need something else, in order to do that Project, but I'm like, wait a minute, that's really a standalone thing. So I'll create a new project to do that standalone thing and I'll do them in parallel. And then that just keeps multiplying. The task switching was killing me. I got to the point to where I go between each session, each cloud code session, and I couldn't remember which what it was do which 1 was doing what. It was just it was a nightmare. So, and then all the markdown files because I have this tendency. I'm not very good planner. I'm not 1 of these people who can plan out something that that I can then give an AI and have it go for, you know, build the entire app. I have to iterate iterate with it and constantly see what's what it builds and go, okay. Now let's do this. Let's do this. And every time I did did that, it would like, okay. Let's create a new plan. So I have plan.md, and I'd have plan._new.md and, you know, 17 others of those. And I had an epiphany. Why don't I just build a tool for this? So I built something called Endless, which is because I have Endless projects. And I started a lesson 2 weeks ago. I was going to bed 1 night. I was just totally dejected and I was like, man. And I thought of this and the next day I got it started working on it. Now I don't lose track of what I'm working on and I don't have to constantly ask Claude about the markdown files anymore. So, and I'm looking at the time. I'm going pretty slow. So let me try to pick this up. What, what is, what is it that Dives built? I'm not gonna read from these. I'm just gonna, sorry. Trying to make it bigger so you can read it. It's a CLI that tracks Staff. It's a local web dashboard that's in a critical component. Apps to monitor, call it hooks, and monitor using the bash ZSH Project, SQLite database to keep track of activity and tasks, an MCP server, a very specific type of MCP server that implements a brand new feature of cloud called cloud channels. And channels are the ability to have 2, or have multiple sessions open and have something from the outside. Actually, it could be just 1 session. Have something from the outside talk to that session that you're working on so it doesn't have to interrupt it. I'm using it to actually let 2 sessions talk to each other, and it leverages a Windows manager called TMUX. How many of you have used TMUX or are using it? So a couple of you. So web dashboard. Let me switch over to the web dashboard. So I'm not going to because of time, I'm not going to show the entirety of it, in large part because the only really significant page I've worked on is this page. So I'm sure you've heard the term dogfooding. I'm dogfooding. I'm using Endless to build Endless. And as you can see, I have all of these tasks in here, that this is Dives. This is this is my this is what I'm working on. And then over here you can see the other projects that I have working on, some of which I have not started using Endless with yet, but that are projects that are on my system and so forth. But Endless is the 1 that I'm currently using it a lot. So that's the web dashboard. Commands. So this is a CLI and, I am going to try to use the, actually LLC me mention that if you'll notice that this is a Orbital, obviously, but down at the bottom you see these Staff. This is tmux tabs and I have my own configuration for tmux. For example, if you're used to if you know TMUX, you don't necessarily have seen this menu because this is a menu that I custom of, because I can't remember all those crazy control keys that it wants you to remember. Orbital LLC me give you an example. So here is endless list that shows the ones that are on this machine. This is actually not my development machine. I had called write something to to sync everything from my development machine over to here. My development machine machine has probably 25 projects as opposed to these. And then another command is, endless task next and this is telling me that, wait, you're not in a you're not in a registered project directory. So it doesn't currently have the ability to bubble up and go up to see what the the project directory is. Is it an apparent? So I'm gonna go back to that, and I'm gonna show you those are the next tasks that the best algorithm I've come up with thus far tells me what's next for me to do. The, I can show, for example, recent recent 5 Industries. Recent limit 5. So I can show, like, what has what are the things that Dives most recently worked on. I can show a specific, task And I need to type it correctly, of course. So that shows, you know, that particular task. And then there's another task that I'm gonna show. This time, I'm going to tell it to show me the child tasks that come with that task. I'm going to stop on that for just a second. Dives AI help you keep track of the status? Really good, really good question. How about I do 1 more command and then I will segue into your question, which was an excellent 1. So when you're working with AI, of course, you don't necessarily want all the verbosity. So I have the command line tool uses dash LLC for switch a switch to give the llm just what it needs without all the stuff that a human would want. In this particular case, I don't use titles because I asked LLM, do you need titles? They're like, well, I don't need titles. So we took the titles off of it. So, we were just looking at that 1 task. Right? This 1? And I'm going to, switch to the endless session. 1 of the things that I do, I don't it's not supposed to be a good thing, but, I haven't seen that it's causing me a lot of trouble, is I tend to to use long running sessions. So I'll have a session for endless. I'll have a session for each of my projects, and then I'll just I'll work in each of those sessions. I have a max plan, so it's $100 a month. That's but I never run-in almost never run into, situations where I've run out of run out of time, so be that as it may. I'm going to say now Staff e 7752. You'll notice the e dash, that's just a convention I came up with. And, Claude said, you know, Jira uses that same Consulting, so that's a good convention. I'll it said, I will understand that. And the reason I chose that, I I first started using pound 7 5 2, and then I went to another project and I said, start using endless and, and then look up this this project. And so it goes to GitHub and looks for the issue on GitHub. Like, no. No. No. Pound Dives not pound 7 5 2 does not mean a GitHub issue. It means an endless issue. So we we came up with this convention. And I was talking about dogfooding. Right? So, I've totally used it, here and this is saying, error, no such command detail and that is because there may be a bug in here. This is, let me, okay yes, the bug is that we just changed it from from detail to show and the instructions still say show. So, if I had time, I would actually show you how how to resolve that, but, right now, I'm not. This task says to ask me, what should can you read this? Is this big enough to read? This task says, what should the heading text be? Needs attention, I'm gonna say make it backlog. So this I'm not making it clear what I'm trying to do here. 7 5 2. The purpose of 7 5 2 is, notice that this says needs attention, but the rest of them don't have doesn't have a label. So, I wanted to tell it backlog. However, and this is this is kind of the key thing and I'm just gonna stop it because I didn't want it to keep going. I'm going to say let's make the, the needs attention. Sorry. Let's make the needs attention, box configurable so that it does not necessarily start expanded. Create a task for that. So you were asking about the AI. Does the AI help you? Yeah. So the answer What's that? Read the tasks and do the fix for it. So the answer is that it is and you'll notice that there's errors and that's because as I've been working on it, I mean, this is a this this this Project 10 10 days old. I keep making changes and but you'll notice that it is recovering from it and that's a key takeaway. The tool is, I didn't really expect it to work, but I figured I'd try and it's working beyond my imagination, beyond what I could have thought it would do. It self corrects. It figures out what it did wrong using this tool because because it helped me figure out how to build the tool and it self corrects and starts doing those things. So, as you can see, it created a new 1, E778. And, if I go over here, and I, I'm sorry. The the glare on the my thing is so bad. I can barely see the screen. There. It was, what were we on? I'm sorry. The glare is killing me. So now it's going and doing the, the E752 as well. So what I wanted to illustrate there was, and this is this is what was causing me so much trouble before I I built this, was I would be in the midst of of working on a plan, and, I would of I would realize I need to tell it to do this. And so I would ask it to stop and work on it because I was afraid if I didn't stop then and have it do it, then it wouldn't, you know, wouldn't be able to do it right. But I kept going down rabbit holes, and I kept losing track of what I was working on. So now I can just say create a task and then go back and do the other thing. I have a whole bunch of other things to show, but I think I've run out of time. I wanna give some takeaways. You know what I didn't get to was? What it's not? It's not a memory system like beads. I had a friend of mine, I showed it to him, or I told him about it, and he's like, oh, what's a memory system? Everybody's building those. I'm like, it's not really a memory system. Yes. It is a memory system. But beads is for, the AI to remember things, not for the user to remember things. The key point of this is for the user because AI, the mem you know, everybody's like, oh, AI AI can't remember things. AI can remember things so much better than I can. The problem I have and why I built this is so that I can remember things. It's also not a knowledge tracking system which is the same Steve Yiggies. No, Andrew Carpathier did a wiki thing and so my friend was like, oh it must be like that wiki thing. I'm like, no it's not like the wiki thing. What do you mean by lightweight? Lesser than lightweight. Just to do list. Can to do list go running? It is. A lot of what I intend to to do, because you saw there's lots of tasks, is to automate as much of keeping track of that as possible. So using hooks for that and so on. The key takeaways that I wanted to leave with with which was write command line tools. You know, and and I didn't write any of the Python and all of this is written in Python. I planned to port it to go, but it's all written in Python or almost all of it and, but write command line tools because then you can you can cut down on the token usage and you can do all the the determinative things in a command line tool. And then dog food when you're building it. Aggressively simplify because when I first got into doing this, Claude came up with this elaborate schema for parsing markdown files and into all these little records and I almost gave up on the project and was like wait a minute, we just need 1 record for every plan, just 1 record. And that really opened the logjam. And the final thing I'll say is that your asset is not the code that was written, it's the design of the thing that you built because nowadays AI can rebuild from your design in no time but it takes a long time to figure out what the design is going to be. And if you are interested, I'm going to be building it for my own use. It's totally open source. I don't have a revenue model for it. Maybe somebody could suggest 1, but, I'd be happy that people would use it. And if you are interested and wanted to give me feedback, I would love that too. So I'll leave that up for a second.