#100 - Tmux Tutorial: Boost Developer Productivity Like a Pro | Thomas Bustos

I wouldn't say that cloud, both are open cloud when you're doing deep engineering work, is the tool that is going to make you 10x. I definitely think that you can be extremely productive and inefficient using it, but not necessarily what will make you the most productive. All right, let's do this. In this episode, we're going to cover a few cool things and we're going to comment the path to 10x engineer, which is a poster recently where you can see my stack, everything that I use, some interesting references that I would recommend in a specific use case, as well as spending some time in Tmux so that we can comment on cool processes and ways to use those tools based on your workload.

So starting from the setup, my favorite tech stack right now is this one, Cloud Code, Tmux. You can use NeoVim or whatever you prefer. I use both. Cursor, when I want to be more on an ID VS Code thing, and I'm picking up NeoVim.

The interesting thing here is Yoko, maybe. So Yoko is my operating system, so it has lots of skills, and it connects to another repo, which I call my vault, and it's basically an obsidian, a second brain. And so they interact very well together, and it helps me manage lots of things, including MCB setups, as well as lots of other things. All of this is open source, so you can check it out.

And the cool thing is that if you're thinking about your own setup, I would recommend that video from Miska. I hope I pronounced this name correctly. If you haven't set up T-Max yet, I would recommend this video for you to follow. It's a great reference.

All the rest resources, and at the end of this post, you will be able to see what you need to go deeper into these different subjects. My core goal right now is to show you a few examples and go to this T-Max session and give real examples of things that you can do. So the first thing is that I use a daily or a weekly skill or a monthly. And basically what this does is it's going to take me through a very specific set of steps to manage entropy around my vault and my goals and it connects to my life pillars, et cetera.

So if you take a daily, for example, DailySkill is going to take you through checking the prerequisites, checking which day are we, the states, checking your calendars. If you have multiple calendars, it will check those. And then you have like a cascade. It checks all the relevant files from your vault.

So this is to fight entropy. Then project updates, people network, task system. And then it processes the information, load extension, check current date, read context, gather information. And now it starts building a table, a draft of a plan with deep work and sessions that are in your calendars.

So this is just one example of all the things that I use in my OS. I usually CD into it and start a cloud code session and then I manage everything from that session. So this is a very specific example, designing a UX. I created here a use case where I have an open source repo that is called Reader This is Reader So here I did the simple analysis of the structure You can see that it an HGS app that has some UI components into it Then I use this to generate the best practices table.

And what's interesting about this is that here I asked it to do a simple change of the icon here. By the way, this is Reader. This is the app. You can see it's in localhost.

Right now it's running in the port 410. And what it does is that you upload papers and it's synced with MyObsidian. It has teams, it has settings, it has a comment palette to do all kinds of things. It has a guide, which is basically how to use it, a methodology to read papers that I picked from Latent Space and also somewhere else I can recommend the resources.

So this is Reader. And here you can see that it has the favicon updated, etc. This is a simple feature. But more importantly, when I used Spawn here, what it did was that it created a new session and a temporary .

md file to give instructions to that specific cloud card. But before doing this, it CD'd into the reader repo. So now this one created this session so that it has its own context window as well as directly being into the repo it's working on. this can write notes into .

md that this one can also read later. There are different ways to make them pair like that. And some will work in parallel, some like more as a master in like sub-cloud code sessions. There's like many things that you can do.

But here I used a skill that I call spawn, where it's basically taking a current to max session and going into another one with all the context, which is similar to compacting or clearing, saving everything in .md before. But the way I've set this up, I like to jump from cloudcast session to cloudcast sessions. A very interesting example here that I think is worth mentioning is that if I check my MCPs right now, you can see that I have all of those.

And here you can see that I have Excalibur. And so what I was doing earlier is that I have Excalibur running here. And all of those are questions of UI that I was thinking about. So for example, I asked, okay, we have our PDF reader app where I can ask things about a paper here.

That's great. But like, what other things could I do? Like recommend me. So basically what it did is that it generated a table with ideas and then I asked it, okay, cool.

I like those ideas. And for the flashcards slash conky kind of experience, proposed me a few UX examples. And so it used MCP, the MCP of Excalibur to generate those variations. and then I asked it to go and look at all the different colors that I know of in my code base, so in my system design, and show me the color palette as well as a brand and positioning and a competitive landscape and a matrix of where would you position it or to compare to existing solutions, etc.

And so all of this was generated in seconds. What I would usually do here is that I might just take a screenshot of this one just to iterate on that example Maybe what we can do is that we can clear this conva Also what I usually do is that I copy all the things that are generated on the locally running Excalibur and I put them into my Obsidian in an Excalibur plugin file so that I keep track of everything and I constantly clear so that context aware. So I will paste the screenshot with this direction, generate four new, more detailed wireframes to have a very clean experience from reading a paper to testing yourself.

Think about the user neuron. Create all visuals in ScalicDraw, Xenium, CP. So that would be an example. And so here I can maybe take a screenshot of this so that it has like an understanding of what this looks like and I can just run this.

And so what this will do is that it will help me iterate on the Scandero combat basically and generate IDs. The ideation process is something that I wrote here about, which is you can just use all your MCPs and granular MCPs to get all the transcripts, etc. And then you can do this kind of prompts to ideate on top of what everyone said so that you get directions and you kind of know the intuition of which direction makes more sense and thinking from first principle what's the best user experience based on all the information that I have so far.

Here it's going to generate and using this candy draw and it's going to be working right here. So a few notes while this is being generated. You can analyze repos, you can generate tables of options when analyzing repo to like map what you have in mind versus what already exists. And then you can ask to create schemas in a Scalidraw so that you can visually look at the table entities and all that.

There are so many things that are possible. You can see that this is very context-heavy, so I have some best practices around that. But you can spin up new Cloud Code to work on dedicated features as well as having one big scope with one big plan and having it one-shotted through one session, basically. That's possible.

Generally, Scalidraw is what we commented. And here in a moment, I'm going to share with you how I save lots of context when I'm doing these kind of things. So if we go back into Excalibur, here you can start seeing a representation of your app currently. And so now it's telling, okay, cool, you're in reading mode and maybe now you also have cards.

Okay, so deep reading mode, this is one. Then card creation from notes, I can like generate study card based on the notes that I have in my app. What is the purpose of pre-training in GPT? That would be like a focused study mode where you go through your cards and then what happens during the fine tuning stage maybe that would be for quick review widget this is like a user journey that he proposed here you would have read paper technos create cards to the cards and quick review now from this i can like save that and ideate on yeah this makes sense i would prefer this here what about those considerations can you look at this transcript from conversation that i had with someone so that you can reflect So that gives you a very strong idea of how I usually iterate on those things Now what I wanted to show you is that see now that we at 56 because we generated all of that and in reality I can just get all this context back because all of this is already generated and we don't need it earlier.

So I can just go like here and go back there, restore prompt and voila. Now I'm like at 60 so I just got back 20% of the context window which is great because now instead of asking it to do that I can just remove all of this and I will be able to now take a screenshot like this and be, okay, so now this is what we want to do. This is what we have. I can go there and remove the previous screenshots and do something else.

Okay. So that's like the bonus point. Now, what I would share with you is that if you're building these days, I would recommend that you have an IT, either cursor or whatever is your favorite. Then I would have a CLI.

Highly recommend T-Mux here. I think it's the PowerPoint of this video. Then you can have agents that can run in the background, either you're open cloud or maybe more sculpt agents like Devin that directly works on a code base. And then like being able to connect remotely to your computer with SSH.

I'm not speaking about a cloud bot here and you can see that there is cloud bot here. The reason I added it is because it's a trend. So I think it's just cool to look at, but I wouldn't say that cloud bots are open cloud when you're doing deep engineering work is the tool that is going to make you 10x. I definitely think that you can be extremely productive and efficient using it, but not necessarily what will make you the most productive.

It feels to me more like something fun to play around and experiment with, and that you can definitely get value out of it. But you also want to think about the security and the setup, which is quite consuming and intensive in resources, whereas all of those are straightforward and directly correlate it to your goals. So we've covered everything that I wanted here, which is a pass to the next engineer. We've reviewed how I usually use my Yoko setup, design UX, building features.

Here you can see the creator of CloudBot talking about how he ships code. Here you have all the resources around NeoVim and Tmux if you want to set it up yourself. This is like a map that I would recommend that is useful, especially if you want to go into NeoVim, Vim kind of things, but that's optional, totally optional. And I hope that you enjoyed this brief video on getting your hands on using Tmux.

This is how I would connect to the Tmux session. So currently my sessions, I think that have Turing, have a work one and a live demo, which is where we were. If I just run my attached, this is like where we started creating all the demo things on top of the reader repo. And the reader repo is also accessible here.

So you can just check it out. It has everything into the doc. I hope that was helpful. If you have any questions, let me know.

If you have ideas of future episodes, I would love to know that. And I will see you in the next one. Ciao.