How I Plan Work Using Claude Code & Parallelized Agents

Let's cover how I would plan using cloud code. Welcome back to the channel. My name is Parker Rex. I led tech for a startup that sold for 23 million bucks. And after that, I've been all in on AI. My background was in product management, product design, and now been in the engineering world for few years. So, let's talk about it. Whoa, big video coming out. What's he saying there? We're not sure. There's a lot of stories we're going to cover and then we're going to go into the planning. This is part of a daily upload. So, usually I do about 5 minutes of news, things that I'm coming across that I find interesting with some takes, 5 minutes of going through questions that anybody has in the community across both channels cuz I have two channels, a daily upload, this one and a main channel, which I do bigger videos on, and then some strategy stuff, which will be the cloud code. Let's cover a couple of these stories. First of all, this is a pretty telling tweet because if you haven't been paying attention, there's a little bit of beef going on between the folks at Meta and the folks at OpenAI. So, if we go back in time and we look at the young Zuck, let's see if we can't find. Actually, I just want to make a little timeline. Generate a timeline of Mark Zuckerberg from when he launched Facebook to every key milestone along the way, including acquisitions and big investments. Cool. So, let's spin that one up. See what it comes up with. But what I'm getting at here is he's had quite a run, but a lot of people will rip on him and say, "Look, this guy made Facebook and then he had to buy everything along the way." Now, it doesn't matter how he got there, he's way cooler or more, sorry, not cool. He's way more innovative and successful than both you and I, unless you're just like low-key a trillionaire, but probably not. And so, he's doing something right. So, I don't want to hate on it, but there's some truth there where a lot of the innovation was via acquisition. So, while that's loading up, this is in a reference to a video where he was basically playing like 5D chess with how he's dealing with his recruitment for Open AI. And the way that he does it is he goes on a podcast with his brother, forget his name, Jack, and he's asked about Facebook and apparently Facebook is cutting checks for over $100 million in signing bonuses and $100 million in base salaries for key engineers that they're poaching from OpenAI. Now, poaching is not a new thing, but it is crazy. And he's his reply to that Sam's reply is I just don't think that's good for culture. Which is incredibly ironic because Sam is right. Paying $100 million for top talent destroys your culture. It's not how you attract the best people. And there's Joanie Iive who made the iPhone obviously alongside Steve. He's also critiqued because a lot of people think that the stuff that he did without Steve wasn't so hot. One example might be the Apple cheese grater case. This bad boy. They called the first one the trash can, the second one the cheese grater. There's a guy actually grading cheese with it. Kind of crazy. But another thing to note is that they did a release where they had a $50,000 Mac Studio Pro and a lot of people bought that and it came alongside this XDR Mac display that was like also like $6,000. I think it was included in the 50. But then the next Techtober, they call it Techtober. Every October you have a lot of releases. You'll notice right in time for the holidays. They released a MacBook Pro that was at the same spec of it for five grand. So, one order of magnitude more expensive for the thing that does the same thing and isn't mobile. So, let's take a look at this. Some of the key milestones leaked cash flow statement saying they did a net gain of 5.66 in 2005. So, yeah. Wow. 2005. And then he becomes a billionaire at age 23 in 2008. They do an IPO raised 16 bill then valued at 104 billion. So 8 years in 104 crazy quick math 8 * 12 that's 96 13 yeah cool Facebook joins the Fortune 500 okay here's the acquisition Instagram a billion this one's always referenced for the revenue per employee that's what I'm trying to do I want to maximize on that one with a bunch of these agents which cloud code we're going to talk about How about that, Joe? Are you drinking coffee? When do you watch this show? Do you watch this show in the background? Am I at 1x or 2x or halfx? Let me know in the comments. And then we have Onavo. No clue what that is. Didn't know about that one. WhatsApp. This guy, I could literally just talk about one story for the whole time, but WhatsApp founder review. So Jan, highly recommend watching this talk on Jan. He was an inspiration for me when I was trying to learn what's under the hood with Superbase Realtime. And if you read the docs, they uses elixir I believe. And if we can find it, Superbase Realtime Service. Nope. We want to actually go back to here. And where's the code? Where's the code thing? Boom. Elixir. Correct. So, you would use something like this for telecommunications or something that's like really stable. Why? I'm not totally sure. Why does WhatsApp choose Elixir? reference Jan Kum, the founder, on reasons why you'd select that technology and then also give me a one paragraph explanation on the benefits of using Elixir as well as a table comparing all the different languages you could have used such as TypeScript, Golang, Rust, C++ C, and Elixir. write out the pros and the cons with why it'd be a good or bad choice as a language for building something like WhatsApp. Cool. We'll send that off to Opus. Cool. And I should have done a research actually. We'll just see. Cool. So, that's the first story is just basically this battle that's going on between these two. kind of crazy. We did one story today. That was seven minutes. So, now we're going to go into Q&A and let's see if we even have any questions. You guys should ask questions. Oh, we do have a lot. Oh, I've not answered questions in a little while. Okay. Love the videos. Can you point me in the direction of intro to AI coding for a complete newbie? Your content is great, but sometimes it feels like I get lost. A foundation would be great. Maybe a video outlining all of your procedures, tools with explanations on what each of them do. That makes a lot of sense. I'm typing this on my treadmill, so forgive. Okay, that sounds good. I should do that. Yeah, let me add a I think I made this in notion. I made a little tracker. I was like, how do I make this smarter for content stuff so that it can be more valuable to the listener? And oh, here's this build system that I'm making for VI, which is basically here's the 10 deliberate steps that you take. So, it explains your your question, which is like how do you go from start to finish? So, you have the different phases of capture the idea, how do you capture ideas, how do you trick yourself into being Larry David aka the guy from curb your enthusiasm. How do you pick the stack? And then ultimately there's a chart that I made out of that. But let's go and add this real quick. Content calendar idea video outlining all your procedures. Cool. That's a good one. Thanks. Great comment. answered on my daily upload channel and we'll post this video on this channel next week. Great. Keep up this series. Great series. Hey David, I was doing this series on my website and then yesterday I was like I just need to finish this. So I did and this was all with cloud code actually. So it has this piano. I don't know if you can hear that but it plays the different sounds of a piano and then a moog and then a guitar. So when you scroll over the links, cool has this little thing. Has the GitHub I guess it's an API call and it gets the information there and then it writes two lines, one to two lines on the gist of what the project does and then puts that in there. So it highlights some of those and then just links. And I'm trying to think if there's anything else interesting in here. Oh, this is about cloud code. So, I learned something interesting about it when I was using it for the last 8 days about how to get more out of the product. So, I'll be making a video about that, but I won't continue that series because I just needed to get it done. And so, this is like a half hour and it was totally where it needed to be. It's got the content in there. It's got a blog. It's got stuff about my cloud code. If you notice like the ratio of input tokens on my first couple of days was like oh 60 in 240 out versus 37 in 764 out and I did some of these graphs and whatnot and how it all changed. But let's see other questions. We'll do one more. Oops. Nice. I want to use augment for self-healing content systems. You should hi Evan. Cool. And now let's get into some planning with cloud code. So my plan for the plan is that I'd like to have a system that makes these videos have less work around them. So let's see. Did I delete that? The way that I do it today is I drop in a bunch of links and then I put them on this little board, right? And I'd like to just automate that because there's nothing really that special about a canvas. So, in my head, it's basically I want to be able to text a link or better yet, I'd like to be able to add a bookmark to a particular folder on X. Probably has to be public so that we can reach it without hitting their API. And then I need to have that post onto whiteboard on my website. So that way it's just there and I can just do the whole thing on my site. So if I went to whiteboards, then I'd like to have each one for each day. So that's the idea. Now we're going to do some of the planning around it. So what I'll do is make this a little bit bigger so everybody could see it. What I'll do is I'll CD into dev and then I'll make a new directory. This is just going to be called daily upload whiteboard. Actually, since it's part of my website, I probably need it there. So, now we're in here, and we'll just Let's open that up, actually. Okay. So, I still run claude code on the terminal when I'm doing like small things where I don't necessarily need to see what's going on, but I prefer to use it inside of cursor when I need to see what's going on, if that makes sense. So, now let's make this a little bit bigger. And I'll typically have up to three of these going at once. But for now, we just need one Ultra Think on a plan on how we solve the following problem. I have links that I add to a public list on x.com/parkerx forward slashlists/daily upload q. I want to make sure that's actually the right thing. So we can go to my public lists and then let's see. Let's make a new one just so we have it. Daily upload Q. Cool. This is the spot where all my bookmarks go. So, I can scrape it into my whiteboarding software on my site. Cool. That's fine. I'll create that. And then we'll hit done. That's always a bug for whatever reason. So, this would be hopefully publicly available. No. Waiting for a post. Maybe I need to add one. So, let's go into my bookmarks and let's grab this and we'll throw it into this list. How do I add it to a list? Do I need to add it to Oh, I think list is for people. So, can I How could I do this? Can I make public bookmarks? I don't think so. Daily upload Q. So, can you make let's just ask we'll think of we'll let them actually plan it. So, if I have bookmarks and then let's go into here. I have this. That should be the list of bookmarks. I don't think I'll be able to hit that. Now it wants to log in. So maybe I have to use something like a headless browser with my cookie in it. That's one of the things. So I'm just like mentally noting that. But let's at least grab the list idea and then we can go from there and see. Cool. I'm not sure that you can do individual tweets into lists. So, this would be one option where we have a whiteboard that fills with creators because that's how the lists work. But then we also want to have bookmarks so that we can have individual posts. So an example would be the following URL which would require O for my account which we could probably use a O off token and some sort of headless browser. Okay, so we have both. Now let's grab a bookmark and let's see. Let's go in here. the following URL. This is the URL for a bookmark folder which requires my O. Cool. So we have all that context. The desired workflow is that the user can either A add a bookmark to the specific bookmark folder or B add a user to a list such that anytime either A a new bookmark is added, we are listening for that or B anytime a new post hits a list, we're listening for that and the link is then taken for the post and brought into either a SQLite database or a Postgress database, whatever's quickest, probably SQLite, and then posted to an Excaliraw whiteboard for that specific day. This will be housed with a whiteboard for each day on the whiteboard route within our next.js app. Excaliraw is open source. So we do some sort of fork of that and create a new one if there's not one for that day. So it' be for/whiteboard slash square bracket open ID close square bracket slashpage.tsx. Let's see if it gets that. Okay, now it is a little bit off. Close enough. Okay, so now we want to make sure that we're putting all the right keywords in here. I made a video or I didn't make a video about this yet. It's not this one. It is about how to get your prompts better with Claude. So, you want to ultra think your way through this. Use parallel agents to investigate these issues, then come up with a plan. So, ultraink triggers the highest thinking level within your model. Parallel agents create sub agents who report to one main agent saves on context and then plan forces it to go through a plan. And you can also use plan mode, but I just do this because this is what Boris, the guy who literally created the software, says to do. And just so you know, this is how think works. So you ask it to make a plan. We recommend using the word think to trigger the extended thinking mode. These specific phrases are mapped directly to increasing levels of thinking budget in the system. Think, think hard, think harder, ultra think. So use ultrathink. And so in that case I want to make sure that I stick in the parallel part. Okay. Ultra think on how to do the solving problem. Okay. Oh no. Few use parallel agents to analyze the codebase and come up with a succinct plan with options to accomplish this. Cool. So that's going to run and this will take a little bit of time. So I'm going to pause. So as you can see it's starting to output this. I didn't even have to pause that. I should have just kept it going. But immediately it sees the prompt obviously and then it's triggered thinking. So now it's giving me everything back. So the user wants me to create a system that monitors Twitter/X lists and bookmarks when a new content is added to either the specific bookmark folder or a specific list. The system should capture these and store them in a database SQLite or Postgress. I choose SQLite for small stuff like this. But my only concern is that if I want to be able to draw on the board, then I probably need off. So the system should capture these and store them. Yep. Then display them on a whiteboard with one whiteboard per day. The whiteboard would be at this route. Yes. Complex task involving scraping or integration. Probably scraping. I don't know. Let's see what actually let's do this. Hang on. Let's make sure it's still going. Yeah, you can do this. We are trying to decide how we want to work with the X API or do scraping ourselves. And I want to know what the latest pricing is for the number of reads on a free X API account. Cool. Now it's doing this. See, as you can see, there's a bunch of things going on, right? It is being done by parallel agents. Then, oops, we have this the whiteboard. It's using parallel agents. Analyze the codebase. It's got a task that it already did of doing that while it was also researching while it was also analyzing this Excalibraw thing. So, you can see that both of these are still not complete because it doesn't have that. And then, let's see. Check for versel config. Yeah, you sure can. I don't have one. Now, on the right hand side, based on my search, here's the latest. The free tier gets 50 to 100 reads per month. That can't be. I know they said it was bad. Now, let's just check X API developer. Okay. Developer portal. Let's see. I get 100. It is 100 a month. That's horrible. So, we can't use that. Oh, maybe. Yeah, maybe if I did. No, that wouldn't work. Quick math. How about that math? So, we're going to have to scrape. So, what I'll do is I just kind of want to see if it comes back. should come back with the same thing and then give me a better plan for it. But we have the to-dos. Set up the database with Triisle. Implement the monitoring system. Boom. Web scraping. Hit new. Integrate the Excaliraw for whiteboarding. Create the daily one. Autogenerate it. Build API routes for Twitter content to whiteboard pipeline. Now I'll think of succinct plan. Okay. So here's our plan. Let's take a look at it. implementation for Twitterx to excali draw whiteboard system based on the codebased analysis. Here are your options. Web scraping. Yeah, cool. Verscelron. No, I guess we'd have to do that though. Yeah, we would because we don't have something that is longstanding. Here's one of the issues with Nex.js if you didn't know. Because it is not a dedicated server, it's on lambdas. Verscell uses their sauce is basically wrapping serverless functions. It still has a server but it's on demand. And so what they do is they basically can't listen. So you need a separate HANA instance or express. And so we have this that full access to the bookmarks and lists without API restrictions. No monthly API cost. Good. SQLite with drizzle. Yep. We'd use playright. Yep. Cron job every six hours. No, we don't want that. We want it instant. So, we need to have it listening. We don't want option two. Let's get past that. So, it would be X to play, right? This is a weird looking thing. Set up this. Okay, we're going to end up changing this. So, we have another piece of work that requires something similar in the fact that we need to be able to listen to things. You can see the name of this is it the following relative path. So, let me open up Bayer. This is something I was working on yesterday, which is I wanted to transform Bear note-taking app into a full CMS for the website with automatic syncing. So, when I added a tag for #prex.com or nested tags, then as soon as I moved it into there, it would publish to the site. But because it's on next, it doesn't have a listener. It wouldn't be able to do this. So I need a dedicated server such that it can go in and read the SQLite database of bear and then bring it in. So I wanted to at least see like what it would take if it if we were to set up a hano instance because then we get unlock for both of these. So I'll grab the relative path. Cool. then weigh the implementation costs and time comparing using versel cron because we'd want it to be more frequent than 6 hours. So we need to figure out the pricing for that. The rest of the plan looks good. That's pretty much it. This video is almost at 30 minutes, but this is how you go. And then you end up with a really solid task list. This is how I plan in Cloud Code and it's pretty exciting. I'm also about an hour out from a weekly call that we do with our folks in VI. I don't have a Discord open cuz there's too many people talking right now during the video, but I need to go prep for that. If you guys are interested in joining, you should do it ASAP. We've got engineers from fancy companies like Microsoft and Google. Shout out Dan. Now I can say that. Thanks for that. And yeah, like the video if you learned one thing. Make sure you subscribe to the channel. And also you can follow me on Twitter. That's what one of the people on our Discord was like, "Dude, you need to make a little popup that says follow you on Twitter." So you can go and see ahead of time if we do the lists that are public. You'll be able to just follow those lists. Oh my gosh, they just made this. I need to add that to his bookmark trigger. And shout out JD, the boy. This guy's like the NAN genius. He's actually doing a master class on MCPs for VI, I think next Thursday. So, if you sign up, you can go and learn from this gangster right here. Eight years of experience. That's twice as much as me. Okay. Bye-bye. Have a good one.