The Adam Sand Report - Thursday, June 25, 2026

All right, hopefully we're live. Excited to have you guys here. If you guys are watching this after, um, I do these lives at late at night, 9 o'clock. The reason I do them late at night is because I fully expect them to suck, and, uh, I, I'm new to live streaming. But what we're gonna be doing today is we're gonna actually like work through loops. I've been experimenting with loops. I haven't been doing like hooks as it were, you know, like where— no, I guess I should— I would say it's not a full loop, right, where you just like completely autonomously just like let something run rampant on your code base. Um, if you're doing that, if you are just literally like, here's a bunch of issues, here's a bunch of goals that I have with my software, here's a bunch of features or epics or user stories or however you're doing it, and you're just saying, oh yeah, run autonomously and like just go make stuff. And if it, you know, like every time the comments from the AI just like feed the next one, you're completely like autonomously running a software over the course of days, um, using loops and hooks, I would love to hear from you and I would love to know, uh, what it is that you're doing to be successful with it. Um, but that's what I am. So I'm going to be doing like what I would call loop light. All right, and so I'm going to walk you through how I do loops, how I use Linear, how I use, um, how I'm using Codex and Cloud Code, uh, and skills as an independent verifier. So essentially one does the coding, the other one does the verifying, and then there's a skill and a loop that we go through and how I kind of build those skills. And we're going to be doing this on an app for the roofing industry. I wouldn't even say it's an app, it's more like an operating environment, but That probably just sounds like I'm trying to be fancy, but what I mean by that is that it goes beyond just being like, oh, I made another CRM. My thing is like, people are using AI today to build the software of yesterday, and I don't really vibe with that. What I'm thinking is like, why aren't we using AI to build the things we couldn't build before we had AI, right? Like, why wouldn't you up your goals, right? So we're building what we couldn't build years ago because we didn't have the technology of today. So like, what is that, right? If you're asking, what does that look like? Well, I'll show you. So to do that, and again, like I said, I'm not the best live streamer, so I kind of have to like do this as I go sometimes. So I usually put my— I have to put my, uh, sort out my screens a little bit better. Maybe I'd be better to be prepared. But I'm just kind of getting, getting used to this whole process. And so I will put my StreamYard over here. And again, if there's anything wrong with the audio or anything like that, help a guy out, throw a comment in the chat, let me know where you're from, share this out if you find it cool. Again, these are kind of like build in public, work with me, you know, do stuff as I'm doing it. Um, but I'll kind of walk you through what we're building. So we're gonna go share screen. Come up here and go on this one. Alright. So what am I building? Okay. So the thing that I see out there is everybody's building the software of yesterday today. Oh, I shared the wrong damn screen. Stop share. Rookie move, Adam. Rookie move. Share screen. Entire screen. This one. So, and again, no shade. I know, uh, Garen, I would consider him a friendly acquaintance, someone I've talked to a few times. In fact, we looked at taking over this project at one point. The whole idea was he'd spent close to a million bucks building it, and we talked about potentially taking over the project to start a software company. And I didn't really, I didn't really like agree with the plan, right? And I didn't— and we didn't even get into the point of compensation or partnership. He just asked me if I was interested in it. I said, you know, it's not really my vibe. Um, I'm not— I didn't really want to do it. And so what do I mean by like every building the software of yesterday today, right? So every roofing CRM, you know, roofer turned co-founder, technical founder, technical, you know, uh, software, you know, VC— it's like menu on the left, stuff in the middle. It's AccuLynx. And I'm not just throwing that like— I'm not— no shade. Like, this is everybody, right? I mean, I'm not trying to, trying to like hurt any feelings or piss anybody off. Like, I like all these people. I do. I like all these people. I just have a different vision for what's possible, right? So this, you know, they've built— they built something cool, right? Like, they got some like 3D stuff here. But again, when you go look at like, all right, so you've built a new software, and if you go like features, right, menu on the left, stuff in the middle, menu on the left, stuff in the middle, right? It's like a calendar, like, you know, It's, you know, menu on the left, stuff in the middle. It's, it's, we're taking the software that we're used to using and we're saying, hey, let's add AI, put my AI on top of it. And it's like, and I mean, of those guys that I've mentioned, they're not, they're not the worst. This is the worst one I found. It's like menu on the left, stuff on the right, but like home, homeboy hasn't even put his testimonials. Like, or like, I just delete this section, bro. Like, just delete the section. If you're going to be, if you're going to be out there selling software, at least delete this. And so what is like my vision for, for this? Like, we're using AI to develop the software of yesterday. So I'm going to be talking— for the people who are just joining in, I'm going to walk you through coding loops, how I run like product management in Linear, how I use Codex and Cloud Code to act as adversarial reviews, and I do coding in loops, and, and have like how I automate the coding process. And if you just like building stuff, this is a cool place. I'd love for you to comment and like be a part of the show. That's why I do a live stream. I do it so that people can come in and participate, and I hope to learn a few things. So what do I see, right? Well, what I see is this. When I think of like where is software going to go, I'm thinking this. Like, why is it that when we think of like the future, this is what we think about? We think that like, oh, software of the future, we're just going to— we're just going to have a generative UI. Like, the Tony Stark doesn't walk in, sit down the computer and open up the— click on, double-click on the app and then open it up and then go to like, go to the menu and go add, you know, add file and drag in the Tesseract and then go open and say click on the other submenu. Because the thing is with software is like, that's how it kind of goes, right? It's like, as software gets more complicated, we go from this, right? And then no shade toward this guy, but then he'll have customers and those customers will be like, okay, I like the abacus emoji, but I need this feature, right? And so then the software gets more complicated, right? And so you look at someone who's a little further ahead, and what do they have? They get, they get a little more complicated, right? So you get more menus, you get more tables, you get a hamburger menu, right? And then if you get into HubSpot, well, then you've got, you've got menus, and you've got, you've got like all of this over here. You start to get leads in here and then you start looking at it. It's like, okay, so then you've got filtered views and then you've got stuff that has to pop out from the side and you've got menus that break into other menus so then they get deeper and then you've got all these filter criteria. So then you have these drawers that slide out from over here and it becomes even busier, right? It becomes even harder. But in the future, this is what we want. We're just like, hey Jarvis, daddy's home. And you're like, okay, I want you to take this, slide this away, move this, add this. I'm not saying I'm building holograms here, folks. I'm not saying that that's where we're going. But I am saying, actually ironic, this is like, is Tony Stark now possible with Vision Pro? I've shown this 3 times today to our investors and to a partner, and I never even read the headline. But like, but like 2 years ago, this is what people were asking. Is this possible? I'm so glad I finally read the headline. I always just type Tony Stark on computer. And this is what pops up when I show this when I'm talking to people. So what is the future of the roofing industry? I think it's this, okay? And it's a generative user interface. If you're building software, this is what I think we're asking for. So we like the environment of Claude. And so underneath, underneath this agentic layer, there's a whole bunch of work. And the reason that, as I say, I'm talking to investors is because when you look, when you show investors this, like what actually makes this agentic layer work, It's this, it's event buses, it's Agent Core runtime, it's a finite state machine, it's an event-driven architecture, it's DynamoDB, it's all of this stuff. I mean, like, this is something I'm working on for a few, few years just waiting for the AI to get to the point. And now it's, you know, hundreds of commits, right? We're talking, um, 29 branches, pull requests out the yin yang. Like, this has been a lot of work that's gone into this. But investors don't care about this. They want to see how does it change— how does it change people's lives, right? And so on top of all of this infrastructure, right, you have to have something that people use. Now that doesn't mean that we can't have the classic app, right? It doesn't mean that like we don't have to have that. When we sign into this, it actually— oh, I wonder if somebody's coding on this right now as I speak. And it's breaking off again. I, I don't know. So if I go to the classic app, we have the same layout. We started there too, right? But we never went to market with it because it's the same old, same old. But when you're working inside a user interface, let's say you want to— you're over here talking about inside sales, you're answering the call, right? You answer the phone, you shouldn't have to type the customer's name in when, when they say something, when the phone should just be live transcribing. I can see the live transcript of me talking on here. Why is it that my Zoom or my live stream, I can see a live transcript of what I'm saying, but you're telling me that I still have to type in the phone number when I'm on the phone with a customer? And when the address pulls up, I have to go manually into Google Drive— or sorry, Google Maps— to see like, where is this house? And then on top of that, for like roofing specific, we already know we have all of these diagrams about how leaks form and how roof and roof should be constructed. And, and, and, and we should be able to live diagnose the leak on the call. And when the customer says, oh, like, here's my email, we can just confirm it. And if we say, hey, we have an appointment available with Sarah on Tuesday at 10 AM, does that work? We can just say yes, and it just sets the appointment. Like, why do we have to type all this stuff in? And then if we want to add something to it, like we want to say, hey, we want to see the weather, we can just like show me the weather for this week, and— or we can, we can speak to it and it brings up the weather. We want to drag a job on, like, that's all simple stuff. But like being able to see the weather on the calendar, being able to see like a deeper look of the weather or even the radar, that's cool. That's what I want to see. And furthermore when we're looking at a mobile app experience, this is a menu that won't exist in the live view. This is just so I can demonstrate it. But like, as an owner, right, we already know that the agents can do the work if you set up the proper runtimes, the proper Event Bridges, the proper orchestration rules, business primitives, Task Set knowledge, all this like gobbledygook of how you handle state control and all this kind of stuff. Well, if you build all this, right, Well, that means that the agents can do stuff, but there are still points where you need a human in the loop. So I had this idea, like, literally, um, literally a year ago. If I look at my Slack and I go into this, um, if I go into this Adam's Brain channel, if I go and find the Adam's Brain channel and I scroll, this is a, this is a thing that everybody had to deal with for a long time.. But I had this late night thought, April 18th, 2025, late night bad idea. Love HubSpot, but what if in the future there was a CRM that could be disrupted as a decisions I have to make, the decisions I have to make CRM, an AI-first CRM. And I thought, what if that was possible? And so that is like what now today we finally have the technology to do, which is where it's like, okay, an event happened. Like every business is just an event-driven business. That's how you become AI native, is you don't start with agents and automations, you We start with events. What are the events that lead to success in my business? And then what are the open loop things? So like the things that like happen over a long period of time, or closed loop, like I do a thing and it does a thing and I do a thing, it's a very closed loop process. What are the probabilistic, like sometimes you do this, and what are the deterministic, every time I do this? What are those loops and workflows, right? So if you have an event-driven business and you understand that from an AI native perspective, Well, then all of a sudden when an event happens, usually as a business owner you only really have a couple of decisions. You either do this or that, or you talk about it. So how can we create this beautiful thing where we say, okay, we are 88% confident that this is what we should do, and so we can say, yeah, draft a response, right? Or no, we can just dismiss this. We got a bad Google review, No big deal, right? Or we can chat about it and say, um, this lady was crazy, just write a nice, a nice thank you and apology, or tag Bill to phone this lady or this person and find out what happened and try to rectify it. So you can say another decision, right? So you can tell it what you want to do and you can be done. Same thing with sales. It's like, hey, there's a lead that came in. It's 0.4 miles from you. Your calendar's available. Do you want to accept this new appointment? Yeah, sure. And we go to the next one. Hey, there's a sale proposal follow-up. Do you want me to write the follow-up for you? Here's some information about it. This is what I'm thinking I'm just going to say. I'm 88% confident. Okay, keep going. Same thing with the field. But then even if you're offline, if you're in— like, if you're not off— like, if you don't have service, that state of decision lives inside the machine And then when it comes back and says, hey, did someone else do something with this? Should I still do it? Because someone else could have made a decision. So this is what we're building, and this is what we're gonna talk about today. So I hope you guys like, like, like, I hope you guys have, like, if you guys dig this, I'd love to hear your comments. Now I'm gonna talk about like how we, how we do it, okay? So what I tend to use is, is I'll use, I'll use Codex. So I like to use Codex, and then I pull Cloud Code up from here to do independent reviews. And so I have a couple of— and so what I— my, my loop is every night it goes and looks in our— and this is, again, this is new. I've only been doing this, this loop on a semi-auto way for 8 days, but now is the next live stream. So this is what we're talking about. And so if you're, if you're in the chat, if you're here throw like, throw something in the comment. Let me know who you are, where you're from. I'd love to have you be part of the show. Um, but I have this automation that will run every day. And so if I go back to my automations, it's like the Knightly Foundation Platform PM Plan. So it goes through, reviews the linear project, and then it says like, what is the next issue that we should get on board with? And let's write a sub-4,000-character prompt and like in the goal format. So Codex has these things called goals.. And so we will pull up Linear for a second. Okay, and so in Linear is kind of like the project management tool. I like it because it's cheap, $16 a user. Okay, and so anybody, anybody who's in here, all of their issues, but I'll go to this project. Okay, and then in this project, you know, there's kind of like the main project, right, um, information. But then there's issues related to it, right? So it's just a project management board, that's all it is. It's just a list of things. But there's all of these initiatives. So we have features on the backlog, we have features that are kind of like considered a to-do, there's things that are like on deck that we're actually working on. And so I say, okay, like, we've got— we've been tackling some of this. Where, um, where are we at in terms of like all of our features? Which is the best one to do? And so we listen to Codex on this. So I go and say, okay, this is the one that ran 18 hours ago. I've had a busy day, I didn't get a chance to tackle this. So what it's doing is it's basically saying, hey, I've read through all this, this is where— this is the state of the workspace, this is what I think we should do next, is tackle this one right here. Now, did I actually tackle this one? I didn't, don't think I did. So this is the goal prompt Litter wrote for me, so I just take that. I'm just going to start, and I like— I, you, I typically read it, and I, I don't want to bore you guys on a live stream, but if you aren't going to comment, if your fingers are broken, then you're just going to have to deal with it. So, goal: implement ADA 175, Canvas, add Golden Prompt, and visual regression harness and roofing business partner foundation platform. Start from fresh origin. Yep, after PR 41. Let's make sure— did I merge 41? Yes, I did. Yeah, okay. So start a new branch, build Durable Canvas verification harness, reuse generalized ADA 174 browser proof, cover proposal auth behavior, add practical visual screenshot evidence. So yeah, so screenshot its work, preserve existing Canvas behavior, do not invent arbitrary generated UI, keep high stakes. Because yeah, one thing I said is like when I'm demonstrating this, I do not want one of like my boundaries is they do not want what I call the Nikola truck. So for those of you who don't know, a little fun story about the tech industry. A long time ago, Nikola was a company that was building an electric semi, and so they took this video of this semi truck. Now, who knows what happened? If you know what happened, put it in the chat. Waiting to see if anybody knows. So this video of this truck driving down this road, it was actually rolled down the road. This is a road on a hill, on a slope, and they basically got the truck moving and then just let it roll down the hill and come to a stop. And that was how they got a bunch of investment. Absolutely crazy that it worked, but that's what they did. So that's why it's saying like, don't invent arbitrary stuff, right? Um, okay, so it's going to have its evaluation criteria. It's going to say this is how you know when you're good, and these are questions before inflation. So should the harness be CI ready immediately? Which 5 prompts are canonical? Live call scheduling. Okay, should screenshots be committed under demo evidence or generated as untracked artifacts? Okay, so now what I like to do is I like to take my, my provided goal prompt and I go to what would be considered a thought partner. So I have Hermes agent right here, and I think I've actually been already tackling this in this because I was, I was going to tackle this. No, I did not. Maybe ADA. No, I did not. Okay, so we're gonna go, Marcus. Um, I want you to review the following goal prompt that was provided to me by Codex for today's loop initiation and help me address the 5 questions based on everything that you know about the project. Help me come up with the right questions and rewrite my goal prompts so it's Codex ready. Like that, like this, add a code block, paste that in, get rid of that, and then go. And so now what I like to do is I like to say, you know what, I'm not an engineer. I'm not an engineer. There are going to be questions that this thing asks, and I could, I could ask inside chat, right, about this. But what I want is I want a bit of an adversarial review. So I go to— I have several, um, so I have OpenClaw, Hermes, OpenClaw, these three agents in our Slack account. And I say, hey, I want you to be my thought partner on this. And so it's going to review Linear and it's going to help me answer this question. Was pushed. Who said was pushed? Oh, it was pushed. Yes, the truck was pushed. Yes, absolutely, right on. It says LinkedIn user, so I don't know who you are, but, um, good to have you here, uh, Mr. or Mrs. LinkedIn user. It's good to have you on the stream, and I'd love to know what you're working on or what you find interesting about this or what you want to see. If there's a question you have, um, I want to be able to, uh, like, yeah, I do these live streams late at night because I fully expect them to not be that good because I'm new to live streaming. And so if there's something that would make it better, I want to hear about it. So I guess while that's running, what I run— so that's the automation that runs, but then, and I'm not going to do it in here, but there's another skill that we have called Contract Completion, right? And this skill, when it runs, right, it does this, right? So after I run the goal prompt and it completes its job, right, it then runs this cons, uh, it runs this contract completion. So it says, okay, work item, what is the claimed outcome, what's the diff evidence, what is the acceptance criteria mapping. So how do you map that criteria that was considered acceptance criteria, what is the evidence that you did it, what is the validation that was done, um, Shabazz from St. Paul, used to HubSpot. My man, good to see you. Um, what is the external side effects risk register? And then what is the status? So this is the skill that I run after I run the goal prompt, right? So after I run it, what is the— this is how it completes contract. Then from there I have in here I go Claude In here we have another skill that is the independent verification skill, which does an adversarial review of the commit, the branch, the PR, the linear issue, and says, hey, based on what you thought you were going to do, based on what you did, this is what I think you should do. So this takes the risk out of it. So essentially, it's a— it is— again, I do— I'm being honest here, I don't have it like running on an autonomous loop. I don't say, okay, um, Codex Automation, right up here, Codex Automation, I want you to go and review, right, and then post a comment to, to Linear. And then maybe Marcus, go review the— go review the issue's, um, goal prompt and then refine it and then Codex run a hook that as soon as the comment's posted from Marcus, run the goal prompt and then as soon as the commit's done, run the completion contract and as soon as completion contract's done and the PR's pushed, have a, a Codex or a cloud code do the independent valuation and then repeat until good like a grep tile code quality thing and then push the PR and then run the thing again. That would be super cool if I could do that, but I don't trust it enough yet. I don't trust enough yet to go full auto. But what I do like is that this becomes very much like supervising the work that's being done, which allows me to go work on other stuff such as the design system, which we'll get into a bit if you guys are down with that, if you guys have questions about it. We're going to come back to Marcus here and see what is it, what does he say? Still nothing yet. So, so that's, that's kind of my, that's kind of my loop, right, if it were, right? It's, if I was to type it out— oh, actually, and there is, there is one more. There's a skill that I have not yet put in to an actual skill, but it is the post-merge implication node, right? So then it adds a node to linear saying What is now true? What is the ambiguity? Did this remove? Which linear issue was unblocked? Which issue should be rescoped because of what was learned? What risk did this introduce or leave behind? What should the, what should the next coding agent use as a source of truth and recommend the next issue? That is the final thing I do before going back and saying, review the project, get me another one. If anybody has something that they think I should be doing, that is more experienced at this than me, I'd love to hear it. But that is how essentially I do the work of an engineering team and, and work with people who punch way above my weight class. Like, if you come in here, this guy Jose, he is 10 times as good as me, um, and he is the, uh, he is the, he is the genius behind a lot of this, but This is how— and again, the nice thing about what I'm doing here is I'm not doing this to make a software to sell, right? I'm not doing this because I want to sell software. I'm doing this because what I actually want to do is we're building a, um, a company that, that will have its own system, kind of like a Delta Airlines, right? And so while we're waiting for Marcus to come back I'll play you guys this. Now that means I might have to give you guys some audio, and you guys can, uh, hang out and watch this with me.

Roof under fire before the rain even begin, but something already moving. Eyes in the wind. Welcome, welcome to the future. Welcome to the future. Quiet work, steady flow. From the storm to the star to the street where you live. This the future. In the future, don't quit. Kitchen table going warm like no pressure. Metal over asphalt, that's value, that's measure. Yes, in the gold thread leap off the screen, fly over the town like a signal unseen. In the opposite lands where the angels don't sleep, take a leap, drop the nets. Every promise they keep. Phone buzz, a home owner. Your book gets set. Colin Huff feeling self-made, broke a sweat. Welcome to the future. It's already here. A goal line moving while the sky's still clear. No one had to chase it. No one made the call. One signal, one system is handling it all.

Welcome to the future.

Watch it move, watch it glow. Welcome to the future. Quiet work, steady flow. From the storm to the star to the street where you live. This, this, this, this, this the future. In the future, don't quit. What we got today? The canvas assembled. A satellite ping make the whole sky tremble.

Image come down, crew move —on time with the 3 days out!

Reroute the line, split the order shop and supply! Go beam through the building watch it fly! Forklift conveyor, the truck on the road every hand in the network already know no-no-no-no-no-no-up-up! Roof clean go in the evening sun permit closed warranty done five stars filling one by one knock two to finish the circle spun they never had to chase it We'll go back to coding in a moment.

Welcome to the future. All right, so that's— we got our goal prompt back. So we will, uh, hop back in to actually working. But, uh, appreciate you guys hanging on, watching my video. I'm just— that's me getting to get a chance to actually be creative. That video, 100% generated with AI. Never edited a slide, never— you open CapCut. That's a crazy thing about today, is that like you can literally just It's crazy what's possible. Okay, so Codex came back with the answer to our goal prompt, so it answered our questions, answers to the questions. So again, I like using AI as a thought partner, right? So if I can, if I can use it to say, all right, I want your— I want you to review my goal prompt and see what you would do Connections, current main, okay, so, okay, so then again, I'm gonna go even more YOLO mode than I usually would because I don't want to bore you guys, but that means I'm usually— I want to read these whole things, like these whole prompts, right? So there we go. I gotta actually pull it from this other screen because it might actually have been too long. Yeah, it was truncated. So we're going to go back up. One sec. There we are. Slash goal. Okay. So now we'll get rid of this. I'm going to get it, go back. So we're tackling an issue here in linear, right? And we're going to go back to Codex and over here, start a new one. Actually close this. And so we're going to start here, extra high. You know what, just because you guys are here, we'll even pay for the speed. Okay, so now you gotta always do this, set a goal. So what that means, this is a long-running task, right? Like, this means this is actually going to run um, oh yeah, it's going to basically compress its context and, uh, or yeah, compact— it's go through compaction but do it in an intelligent way. So we're going to let that run. Okay, and so now this is where we are in the loop. The immediate next step is it's going to do that contract completion skill, right? So for you, those who are just hopping in, If you want to see what that is, when it's done, it runs this completion contract, right? So it fills this out, and then when we do that, we're going to pull up Claude Code and we're going to run the independent— we're going to run that skill, and then that is going to allow us to loop around and get this done. Again, you don't have to be an engineer to do this, but you should probably understand why you're doing what you're doing. And then we have our last thing, which is for readiness, and we'll use our thought partner if we want while we're doing it. So we're going to let it run and it's just going to code for us. Now, to keep you guys entertained while we're doing this, what I have here is our design room. So Now I have to not make this full screen. So what this is, is essentially saying we're building out the design system, right? And so what is a, what is a design system? Again, if you guys are just hopping in, Love to hear where you guys are from, but what is a design system? So when you're used to me, you might be here used to hearing something about like a brand kit, right? And a brand kit is essentially the way that you can give agents the ability to stay on brand with anything that it, that it builds, right? Anything that you're building. This might actually be one of you sending me a DM right now. So this keeps your— like, this is so that anytime I want an agent to design something for our company, it uses this brand kit as a way to understand how it's allowed to use logo, how it's like this. It's got the logo kit. You can come in here, they can download our font kit. So it has all of our— or all of our icon kits. We had— we got these custom, custom icons, right? So everything is isometric because When you look on here, we have this stack concept, so it's isometric stacking branding thing. The idea is like, all right, well, we want everything to match that. Similar to that, when you have a software, you want to have all of your UI components and everything like that also in that same way. Right? Go here. We wanna say, okay, we like, this is one improvement we made. Like we said, okay, these dots weren't really good, right? But these lines right here, they kind of made it nicer, right? So we were kind of saying we're just trying to tighten up the design system. So I'm just showing you guys this because this is like another part is like in addition to functionality, we also have to think about front end, right? We also have to think about the front end and making this look nice. So we're going to merge this Canvas Design Room so that we can pull it in somewhere else. Because I wanna pull this in a Cloud Design. I've actually never done this before. So we're gonna experiment with it. But let's check in on Codex. So again, it's running, it's got its full 5 steps. This is a very long-running task. It's going to burn hundreds of thousands of tokens, right? Like we're looking in here, it's already burned up 86,000 tokens, 67% left of our context window. It'll automatically compact itself. This could run for 20 minutes, this could run for an hour, right? So that's the beauty of— you get to work on something else. And so One thing that I am working on is, can I use Cloud Design to pull in our design room? See, so this just says the shimmer, because you wouldn't believe how much actually goes into like a design system. You also have a motion system. Like if you look, there's companies out there that have like, it goes slow, then fast, and then slow, right? And there's other people that it's like fast and then slow and then fast. Right? Like the way your swipe animations work, there's so much engineering and so much thought that goes into it. Like, why is it that the whole like Instagram Reel, uh, TikTok, uh, Reels kind of experience is so addictive? It's because you only move your thumb just a little bit and it slides the whole screen up. So you flick and it like moves the whole screen up. The amount of effort required to go, it's like, it's not direct input that actually takes the whole— you have to move your finger all the way up to move it. You just get to do that kind of flick. It's not accidental, that was intentional. Like, they figured out that if, as the video is going up the screen, if the next video starts playing a little bit, you might get interested and get hooked on the next video. Like, so much of design matters. And so I'm trying to say, all right, I want to actually get even more con— like, even more convicted in our design system. And so I wanna say, alright, we have to bring this in. And so I wanna say, hey, this, this PR is merged. Okay. This PR has merged. Can you pull it into, into this screen so we can make more design decisions? So it's on— still on step 1 of 5. So see here, it broke it out into steps: verify a fresh base, inspect Canvas Pro surface, design deterministic ADA 175 fixtures, assertions, and to implement harness browser group. So it's taken our goal prompt, broken into these 5 things. And this is what's so incredible about today is that like while this is running, you can be working on something else. I can also be working here with Marcus on the design system. Like, you can see, um, I went in with Marcus yesterday and I said, hey, I want to get better at our UI and UX. So I pulled in a bunch of videos that just said, hey, like, companies, I guess, think about this motion system, right? Most teams have a design system, colors, topography, spacing, components, all documented and all governed. But one decision is left unmade: motion. Motion is how your product moves, how it feels, and how people remember you. Google, Uber, Cash App, they all figured it out. They started defining their motion the same way they defined everything else. But for most teams, easing still lives in a project file, so timing lives in someone's head, and guidelines live in a PDF nobody opens. Your motion is inconsistent in every— see, like, some people get their motion goes faster, some it's linear, some goes slow and then snaps, right? And that has implications on how the hand moves. So it's like, hey, we— I got— I gotta think about this, right? Like, I— like, we don't want— we want to build software that feels like art, right? Like, that's enjoyable to use, that, that feels like it has polish, that's not just another like, hey Claude, build me a CRM. And so I got it to break it down into these steps. So this is what we did yesterday, was like some hardening on this to say clean it up. And then we went through and did our goal prompt and like made it better. And now we're on the next stage of it Which is like, okay, I actually want to, like, build the design room and pull this in. And so saying, okay, let me check. Let me read. Okay. Got it. Now the real module state primitives, I can write it faithfully. The graph is set. Okay. And so it's building this out. And so I can work on this while Codex is working on this. And then if you ever have any— if you ever want to have fun, you can always bring your pet. Codex has these cool little pets that if it's on another screen and you're distracted, you can have this little guy right here, and this guy will let you know what's going on. So you can be working in Claude Code, or you can be working in Claude Design, and this little guy is like working away, and then when he's done, he's going to, uh he's gonna like make a little— he'll like, he'll like jump up and down, it'll go green. If you click it, it refocuses, right? So if like I'm here, when I click on my little pet, it brings me back. So little fun way to like make the, uh, make the whole process a little more enjoyable. See, even this, like I learned yesterday in one of those videos, so like there's a reason why softwares do this, is this gives you the illusion of like the illusion that something is happening. So it makes waiting a little less bad. Same thing, why do they create this little pet? Like, why does this billion-dollar company go through the trouble of making this little pet? They do it because then you can send a reply in here, but you can also easily get back. It makes it fun. How do we make— like, how do we make the tools that we use? Don't build software, make art. Make things that people enjoy using. Is the premise. Because I can see what happened here is that there is— if we go back to our, our, uh, our canvas here, if I go back to our classic app, which I don't know why it was kicking me out. Okay, there we go. So we have our light and our dark mode. So this is like, this is where we started a year ago. Right? It was like, okay, we need to build this agentic layer and we need to have a way to orchestrate it, test it, run it. It's not intended to be a CRM, it lives on top of HubSpot and Zooper, right? But now I can see what's happening is that when I did the design prompt, what it did was it pulled in some of the design decisions from back here. So that's why it's saying that we have more fonts than we actually might have. Right? It's like it actually has— I think it has too many fonts, right? And then when I look at this, I'll probably say, hey, I feel like this needs more spacing, right? Like, this isn't great. But what it's doing is it's saying, okay, well, we have a state. Like, when you think of like a great software, right, if you think of something that's really great, you have different states. You have a loading state, you have an active state. Like, I'm pretty sure if I go into Cloud, when I click in here, right, there's things that will happen, right? Like this suddenly turns to orange, right? And so there's like little different states that exist. Like before I start typing, it has these. That's why when you look at our canvas version, right, it prompts these. But these originally were very imbalanced. You can see there's a hover state there's kind of like a— this is the default state. Same thing right now when you click in here, there isn't much of an active state other than once you start typing, this is here. And, you know, we have this here to kind of say pre-populate it, right? And we have this here. But right now, like, what I'd like is I'd like for when you click in here, it would be nice if this didn't just have a cursor, but if it glowed or something, right? Like, how do I make it so it like gets a bit of a glow underneath so you know that you're in that field, that you can visually see that. Like, that's how we can say let's build something that actually feels like a delight to use. And so it's still rendering our whole design system, like our weather. Like, right now we don't have our radar, um, in here yet because it's one of our issues that we still have to build. Because even in MapKit, right, if I go to owner and I go back to online and we'll say reset, we are able to use MapKit, but I'm still fighting to get Radar on top of this and I'm trying to find a way to make it a little less busy. Right? So there's different map tiles we can use, and the goal is to put Radar on top of the Apple MapKit. That's one of the things that I'm working on figuring out and like setting— like, what is the default zoom, right? Like, should it be— oh, like, how do we work with this, with this map in a way that— because I can zoom in but I can't really zoom out, right? So what is the What is the process from that, right? Do we have to put a button on here? Because now that I'm zoomed in, you know, I can zoom in, but like, how do I make it so that you're not dragging the card while you're playing with the map, right? Like, we should have buttons for that. So that's another issue that we would load up. It's a very small issue, but it's an issue nonetheless. All right, let's check on Codex. Actually, we don't have to even look it up. Oh, see, look, read goal objective, verify done. So we click here. And it's done. So it says, okay, add this deterministic canvas. We're going to get rid of our skill here, and it feels like it's done. Now, did it do its completion contract? Verify, verify, done. As a builder self-check, not independent verification. Used up half a million tokens. This is what it did. So we're going to say run this skill, right? So I'm going to turn off— well, now we'll leave speed on, we'll continue. We're probably going to burn some money tonight. Should probably go check what my usage limits are today. And again, for those of you who are hopping in, like, if you're— like, I would love to hear where you're from, what you got going on. We're just trying to make this as exciting as possible, to not deliver like the usual fluff, right? Like, I don't want this to be like the fluffy kind of content where I just read the model card off a new model release and say like, GLM 5.2 kills everything. No shade towards all the people who do those. I, I enjoy those videos, I watch them. But like, I want this to be like, hey, what are we— were we actually building. What are we actually doing? What are we working on? Okay, no, we're actually doing tokens. Okay, so see here, now it did its contract completion, right? Well, it's about to be done, right? So we have in here, we have— okay, so work item, this is what it did, this is the claimed outcome, this is what it changed, this is the acceptance criteria. These are the validation commands. 107 warning. Hola. So what I would do is I'm gonna pull this out. Copy. And then we're gonna pull this up. Make this bigger. So please spend some time on the 107 warnings. And now we're going to let it run and it's going to do its independent evaluation. Let's come on back over here. So one thing I noticed about our design system is we have a lot of these round-edged pills, but if you look at Foundation Project, we don't have round pills, right? We have square edges that are like rounded edges or rounded corners, not fully round pills. I don't know if I like having that. I feel like it breaks up the shape too much, right? And so, yes, this is a lot of Sean's actually booked. So how do we improve? Like, then the thing is like, how do we improve this? Still— oh, this is a loading state, empty, error. See, it's like this error is a little bit too small. Fresh, stale, lock, thinking dots, excuse me, reveal, shimmer, selection, pulse. So I can come in here and let's say, what do I want to fix? What are we going to fix first? Okay, so we'll come in here and we'll say markup. I'll say like this. Say I feel like there needs to be a little bit more spacing between the word and the number, or I guess you could say how you would define the headline and the content, because that there isn't enough pixel spacing to make it so that the number doesn't feel like it's almost touching the word. We need to have a better strategy for how we give this enough white space so that it's beautiful but not making it overly big either. I'd like some ideas as to maybe perhaps justifying the number to the other side or making the text smaller. I'd like to see a couple of options. Let that run. What's our pet doing? It says Canvas component warnings are trivial, no functional impacts, or is there non-canvas files? Let me run the git type case. It's focusing on those errors. So again, this is Cloud Code just doing a review of what we did. Let's see what Cloud's putting together. And this is just kind of like a creative playground to say, you know what, I want to know, I want to be able to tackle the— so, so now see here, so it has our, our design system, it has a version 1 that we had, right? That wasn't right. So then it came here And now it's giving me some options. Yeah, see, this is current too tight. Okay, so this one made it bigger, keep it stacked, add a real 12-pixel gap and a touch more padding, number stays bold, simplest, most editorial. Option B, split to the side. Dial the number down to 22 pixels. So again, reads without growing the tile, commas good for dense boards, hard line divide. Oh, I don't— why is that? I'm curious about option D or option C. I do like the hard line divide. I do think that adds a sense of polish. I do have something I'd like to see as another option, which is a hover state. That when you're looking at something like this, you're looking at revenue month to date. That's the information that you want to see. But then couldn't there always be like a little, um, information icon following our brand standards? And then when you hover over the information icon, it could pop up like a little card that actually shows not just the increase of 8.4% versus last month but maybe multiple data points, like this much over last month, this much over the same period last year. Like, it actually could deliver a little bit more information even because it's just a hover-over card or a card that you click. So for those of you that just hopped in, because I see the number counts going up and down, If when this is done, right, when it finishes doing this adversarial review, the next step for me is then saying, okay, it's going to give me a response and I'm going to want to know, like, what did it think? And again, I'm moving a little more quickly, like, with you guys here because I'm not wanting to bore you with the, the details of reading it. But let's say it comes out and says there's a couple issues. I would take a couple of approaches. One, I might chat directly here with Claude Code, or I might come back to my thought partner, which is the Hermes agent inside the project Slack channel to say, hey, what do you think? Right? Or it could come back and say it's done. And when it's done, then we just do a— run the merge skill, right? Or we close out the merge, and then we can go and pull the next issue and we run the loop again. So it becomes goal prompt, usually has some questions, thought partner, answer the questions, run the goal prompt, contract completion, independent review, discuss with thought partner, close the merge, merge, post-merge checklist, rerun, rerun the analysis, take down the next issue. And this is a way to feel, um, a little less burdened with kind of making sure that you follow the right details by using skills and hooks and loops and, and goal prompts and thought partners. This is how I'm coding these days, and I'd love to see what you guys are doing or if you guys have any thoughts. Okay, so option E, insight on hover. See, now I like this. So it's like, what if we just— oh, so then we could say I do like this. I do like this. Could I see option E with no green arrow showing the increase versus last month but still having the hover option? So essentially you just are seeing the revenue month to date and then having the hover. I think also you should have two buttons on any information card. One should allow you— and this can be function that we have to build into the backend, but you would have, in addition to the hover over information, maybe all the way to the right, you'd have two things. You'd have a kind of microscope icon, again from our brand standards, so it has to have the same line thickness and everything. Microscope, or maybe a magnifying glass. The concept is like zoom in, like what is in this $182,000 revenue once a day? Like, what are the deals? So you— so the generative UI would then be triggered from that. So the canvas would then open up a zoom-in of a table of what actually makes up that revenue. Then another one would be a little chat icon where if you click the chat, it would then bring that element ID or that like something that says, hey, this card is what the user wants to chat about. And that chat icon would bring it down, it would bring a reference down to the chat box that would allow you to then chat about that number. So I know that that's backend function, but I would like to see that represented here so that I can eventually take that and build a linear issue to make that functionality possible. Independent review is almost done. I can tell because I start to see the table. Verify done. Proof scope work is uncommitted. Cannot move to done. Okay. So it's saying that we're good. I think. Can you still make warnings? Okay, risks not verified. Oh, I like that. References a card. Okay, last bit of feedback. Now that we don't have that, that increase of 8.4, I would like option E to go back to having the increase of 8.4, like that arrow pointing up or down, right? But I would like it to be justified to the right. And I would also like the zoom in button and chat button to be available Actually, no. And I believe it needs to have the same white space and padding so that it follows a logical design style. But add that increase back in, and then there's the zoom in. I love the zoom in, I think that looks great. We need to have clear rules around like how that table is displayed because there could be more columns requested, and so there needs to be more of a design system around what a zoom in card would look like. So let's expand upon that. Right here. So then what I'm gonna do is I'm gonna get it to commit its comments to Linear from Claude, and then I will get it to commit its comments to, to Linear and GitHub. And then we will merge this PR, and then we can go run the post-merge skill, and then we can run another loop to see what we wanna do next. So now we've taken it from this card, which felt smushed, right? Now we've got this nice space. Right, we got this much better spacing and we got our increase of 8.4%, which is much nicer. And then we have our zoom breakdown where we get— if we click on the zoom, it should pull this. And then we have— we still— and think, there's so much to think about, like what does that zoom effect look like? Is it going to make the card bigger or is it going to be a modal on top? Then we chat, we know what this is going to look like. It's going to pull this down. Chat icon attaches the card's element ID to the doc as a reference chip, so the next message is scoped to this number. Well, it looks good when I hover the mouse over the Option E card. I like the way that looks. The, the breakdown card on the tile to the right of it though doesn't look the same as the zoom-in card, so I think you need to correct the design there, and then I think we might be ready. Please complete that first. Again, for those of you who are hanging out, I appreciate you guys being here. I appreciate the support. It does good for the channel when there's multiple people in the room hanging out. I would love to know what you guys are actually getting from this. Okay. Says I got this message. Sorry, I know I got to be more entertaining. We got to like Alex Finn clap and be super high energy. That's why I know my live streams are not as good as like Alex Finn's or anybody else's, because I know I'm not being like, let's go, let's go! But I gotta get better. I'll be— I'll promise I'll try and find a way to be more entertaining. New version ready. Okay, so we have option E, merge. This looks good. Chat, there we go. Ask in chat, zoom in the breakdown card. Oh, I love it, I love it. This looks so much better. Okay, there's a multiple. Okay, zoom in card, the rules. Okay, we're gonna go, um, hi Max, let's spend some money. Yes, please merge this into the main, put it onto the main design system page. And put clear rules and explanations so any agent that reviews this design system understands all the decisions and how to reconstruct these across multiple places inside the app. So where are we at here? Still working. Check out merge. Two things is proves— oh, verify done. Perfect. —unless tail sign-off is confirmed. Green against design and work on main. PR 43 is marked ready and merged. It's clear to move into EC1 series. I want me to save a short memory note. Yes. Okay, so let's go into I've marked the PR ready to review. I wanna make sure that we don't have any messy branches or anything like that left over behind when we're done. And when we finish this loop We're gonna run them post-merge, and then you guys will have got a chance to see a whole loop inside of an hour and a half, 90 minutes. Not bad. Okay, so, um, We're going to come here, just wait for this to be finished. Let's go back to Cloud Design while we're waiting. This is what I love, just the ability to like come back in. Okay, so perfect, it's good. I don't love this. Don't love this at all. We're gonna have to fix that. We're gonna have to fix that. And added this fresh. I don't know why it added that, but this is why it is work in progress. But I don't love this. It didn't fall— created this big number at the bottom, which we don't need. We don't need it to make a big number. And then this feels like it's smaller. But we can reopen this as a knit. Oh, and we got our green here. So merge, confirm merge, come over here, skill that I have not yet made. Come back over here. So now this is back on design system. See, so now we have this is how the card works. This is what the zoom-in card looks like. Now, again, we have to fix this. And then we have a reconstruct anywhere. Now, it's gonna— let's see when it checks the design system, it finds that issue. Now ironically, in this video it actually talked about this. It actually said— where was it? Say this one. No, that's not the video. There is a video that I watched that talked about— it was like about vibe coding, like common signs of vibe coding. All right, so I won't waste your time. Okay, so it did his merge, made his comment. And so then now we have officially completed our loop, right? So then we can go back and again, I usually run this as an automation every night at 3 in the morning, but I can come back and basically copy this and run this automation now, or I can just click run now, and then it's gonna go pull me my next issue, right? Rather than waiting until 3 AM. So we're going to rename— copy. I'm going to rename this. This is something that I'm trying to get better at. ADA 175. So I can actually remember which ones these were. And now we go back in here. If we click on this and close this, we can see here, move this one to done, and we're set. So that's our loop, everybody. So thank you for being here. I hope you guys enjoyed this live stream. If you guys learned something, put something in the comments, in the chat, let me know you enjoyed it. I will be trimming this video down probably at some point and, uh, turning it into hopefully some shorts that you guys like. But thank you for being here, and we'll see you guys on the next one next Thursday, 9 o'clock. I might do these lives earlier in the day if people start commenting and saying they like it, then I'll do them a little earlier.