This AI Setup Will Outperform My Old Engineering Team

So, we've got Taskmaster, we've got Klein, we've got this memory bank thing, we've got Windsurf, we've got Windsurf Next, we've got what Ader, we've got Copilot, we've got all these tools, and they all kind of do things differently. Some have strengths, some have weaknesses. Welcome back to the channel. My name is Parker Rex, and on this channel, I share the tools and tactics that I'm using to grow my AI businesses, both across SAS and services. and we talk about things that are going on. So AI stories and some hot takes. So we try to keep it short and snappy. The first one today is Zuck coming out just saying, "Hey, look, like within 12 to 18 months, most of the code is written by AI. It won't just be autocomplete. AI agents will set goals, run tests, find problems, and write better code than top engineers." Now, they also released a MetaAI app. So they're going on the personal AI way of things like where it's like this is your personal one. You can tell it anything you want. Don't love that. I know he's doing a big rebrand and he's got his shades on and I do think those are pretty interesting because not only is the audio awesome, but just the form factor in general. And if you're familiar with the story, they had a lot of trouble fighting mobile basically because they get kind of neutered when iOS comes along and does a big update. So now they're trying to own the new platform and a lot of it will have AI generated apps on board. I gave these a test over last weekend and the audio is bananas. They're still a little clunky, but we can only expect things to get smaller, lighter, better. And just to note that this is specific to their code bases. So I think that's going to be a big trend too where it's like your tooling internally will be trained on that codebase for your business. Next up is on cursor. So, I don't really love MCPS. I think that they're inconsistent. I think they'll get better over time, but we're seeing this trend happen where you can actually use an MCP to talk to cursor. So, when you download this plugin, and I'll be showcasing this actually in a three-part per week series where I do live builds on my main channel. Right now, you're watching my daily upload channel, but I'm going to be doing 20-minute episodes of things that I'm actually building in my own businesses and how to use them. But in this case, this would fulfill kind of the last part of the design phase. So design's going to be at a huge premium very soon. We're seeing huge on uh basically ramp up in new developers and everyone's kind of using the same stuff. So it's not great, but that's why design will matter more and more and more as time goes on. And that will fill in the gaps for that part. So now onto the bulk of the video. So we've seen a lot of different ways of working over time for software. So you had waterfall was a really popular one where it's like each step falls into the next and when you go to do something well you have conception. So that's the idea phase and then you initiate meaning like okay we're going to get this thing going and then you decide that you're going to do it. Then you analyze the market. You try to figure out like who would the customer base be? What's the unique value prop? All those things that are required for business. Then you go design it. After you design it then you start building it. then you test, then you deploy. And that worked for a while. And then we came into agile where we kind of combined some of those things, got a little bit faster. You also saw lean and agile and all these different things. There's like a Google 5day sprint which they wrote a book on and there was a popular firm that you can go and do a fiveday thing with. I did it once with our team at Delivery Dudes and it was good, but we didn't do it again. So that tells you it wasn't great. Then this all ties back to Agentic tooling by the way because I think each one fits a spot. So then it came shaping and the shape up methodology from DHH and Jason Freed over at Base Camp or 37 Signals and you can see that you'd start with this like raw idea that's based in a problem and then you shape it up a little bit and then as you go you bet on the work. So they work in six week batches. You could do six one week, two three week, three two week, you get the idea and you'd go and actually do the work. I talk about this a lot with the ideation phase where product management is actually at a premium as well as design and development's kind of going down because those steps before are not automated and because your ideation your ability to think differently for product strategy is just so so important. So, when I look at this, it's like when you're coming up with a new new idea, this is basically like the PRD phase. And everyone on the internet's like, "Give me the new prompt that's going to solve my, you know, world hunger or whatever." And it's like, "No, no, no. You still need to think." And I say this over and over, but like if you have an hour to work on something, you should be spending at least half of it thinking about what this thing is going to do. Why does it need to exist? And so, that comes into modeling your prompts. So in the case of shape up every time we go and we build something and I've actually brought this into how I prompt but you're basically starting with why. This is the big question to answer where it's like if you look at like Simon SK that's a good example but like he talks about that where it's like why are we doing this and you narrow that problem down and I recommend that you solve problems that you yourself have especially if you're learning how to code because you won't have to go and talk to some external person because you are the person that wants it and you're going to know the problem deeply and that's what they also look for when they're hiring people to or when they're vetting people for Y cominator in the 15inute interview it's Why are you doing this? Did you push yourself to the edge of what's possible to figure out what the problem is and then how to solve it? That's what a good pitch is. So then the solution phase is then when you'd actually like get out like a fat marker and you're like drawing some stuff, not that much detail. That's important because you want to get the rough idea that then ties to the solution. You don't want to get married to the design. Then you have a time box. This is irrelevant now, but still good for if you're doing team stuff. I think at a higher level, not when you're giving it to an agent, you don't want to mention a time box. But when you're thinking about the PRD, it's like, well, how much time do I want to constrain this to? This is your affordance, your ability to say, this is what the time is that I'm willing to bet to solve this problem, not the other way around. So, you could come up with this amazing solution statement and all these screens, but if it doesn't fit in the time box, then it doesn't matter. Then you have rabbit holes, which are things that like you just need to kind of avoid, which is like you know about them. They're known as being bad. Big time sucks. And then you have no goes which are like let's not even go there. Um so rabbit hole like the big difference one like between these two is just so you know is like rabbit hole is like you know that you might have to touch it and you're like let's just be aware of that and then no go is like no no no these are off limits. So then that comes into our PRD. So you use a template, right? And you actually go and you write this stuff out. Like you're sitting with a pen and you're thinking about these things. Like what is it I'm going to build and what is actually the big part of the product manager, right? It's like what are we building? Why are we building it? And then how is the engineer's problem to figure out? So in the olden days, but like before all this AI stuff, you'd have me so a TPM or just a PM but a technical product manager and their job and a company. if you're not building for yourself is to go and meet with ops and meet with sales and meet with marketing and meet with customers and you might have multiple customers. In our case, we had three. We had restaurants, drivers, and uh end customers. Then you come together and you make this dock and then you're going to end up with a stack of these things, a stack of PRDs. And then you're trying to figure out which one's the highest impact, like where do we need to go next? And you might be optimizing for different goals. Maybe you're trying to optimize for efficiency. We want to have faster order times. If that's a goal for the quarter, then you go for it. But in our case, for this exercise, then I'd go and I'd meet after I figured out the what and figured out the why, then I'd go meet with an architect and then we figure out how. And that's when it comes into all the systems pulling together different parts of maybe it's like a monor repo, maybe you need a microser, whatever it is, the different files that are affected. and you're doing your best to say, "Okay, don't pull any like while we're in theirs because that will go off the rails." So that's when you're saying these are no goss. So that's how it used to work. Now, although people that haven't done this before, they're confused why their PRD isn't awesome. And you just need to get better at this part. So I recommend like go read shape up because that's really easy to read and really good. But then once you have it, you put your idea or your pitch in between these little brackets. And if you want this prompt, you can comment prompt below and I'll share it with you. I have it tied to native keyboard shortcut. So I just do at F1 and that's all the pre-work, right? So we've basically just talked about a gentic chat with prompts, but then there's all these other steps that we're basically trying to map to different agents in the stack. So it's like you have analysis, you have design, you have construction, you have testing, you have deployment, you have all these things. So each one of these will represent one or multiple agents because they have different strengths and weaknesses. So you're kind of gluing them together. I think that's what's confusing about a lot of these tools as people like expect one sizefits-all sort of deal, but it's not the case. So what I found so far, what I'm doing now, having a lot of fun with it, is basically I do that whole process before. So if I lined it up, it' basically be like pitch step one. Step two, agentic chat with prompts where I'm like shaping up that PRD. And then this is where I'm trying to figure out where it's like taskmaster at the end of the day is basically like the product manager that's sitting there buying the engineer cheeseburgers and like pumping them up and reminding them what's the most important thing. Don't build the Bitcoin wallet. Build this. Make sure this has this ticket tied to it. This ticket needs to go forward that ticket. That sort of stuff. Like that's where Taskmaster shines. Now, where it doesn't shine as much is actually knowing what the architect knows. So, like knowing like just like I didn't know everything about the stack, the taskmaster doesn't either. So, memory bank acts more like the architect of just keeping up to date on like an architect has everything in their brain. And what memory bank does is it kind of emulates that where it's like, oh, I need this, I need this, I need this, I need this. And you'll see great examples of that in this codebase where it's like if I jump over and uh let's say like progress for instance and I can see what's going on and I initiate the chat and I just get all the information about it. I can see like the progress, the project brief which is like basically the PRD, the system patterns and they also have like really good mermaid charts so I can see like what's going on where and these come out of the box a lot better. So, this isn't a commercial for Klein. This isn't a commercial for Taskmaster. This is just me recognizing that like they kind of work hand in hand. And for a minute, I thought that augment would make it into my stack for doing tests. But I don't think so. I think it's just these two because at the end of the day, it's like whoever does the better job of what we used to do then, you know, they win, right? And so each one of these companies is trying to basically fill one of these gaps in what was and still is at most companies the process that they run today. So in order to be on the edge of this and figure this stuff out, you have to have that base knowledge of how this stuff works to begin with. Otherwise, you're just another guy on Twitter that's like looking for a viral thread that's written by somebody that actually hasn't done anything interesting. They're just selling ads. So that's what I think. And I'm going to be testing this out of like how do I glue these together in a way such that someone's going to try to build the one sizefits-all, but when you have a team, especially open source folks, like people that are like really excited about it, which is Klein and Taskmaster, they like push what's possible for that one thing really hard. And I think that gluing them together in the short term is the move. So I'm not sure because right now they're both dependent on the like basically like you have cursor rules or wind surf rules and these are globals that sit underneath. So it's like oh I send out a chat and it's like go got to go check this. And so in memory bank's case it has like a bunch of diagrams and they're connected and it's like hey every time you do something go update these or reference these and whereas taskmaster it's going to say something but a different set of rules. So that's like the big no no. that's like the big thing that's like causing the issues. But there's ways around it. You can totally do it. You can probably tie um them to like keywords or something. It's like maybe even a CLI that wrapped both of them. I'm not sure. But for now, just separating them has been working really well for me. And uh I want to just get them to work better together. So yeah, we've been chatting a lot about this on the Discord, just seeing like, you know, every single day it seems like somebody's coming up with something new. Um, but I've found again like just running client and then in my head I'm doing the taskmaster part, but I want both of them together. I found like client's really awesome just because it has like the browser use stuff where it's like, oh, I wanted to add like a tan stack start as the front end and it struggled a little bit because that's newer. It doesn't have like a bunch of docs that it's trained on, but it does have this web thing. So that's great. And um yeah, that's it for the video. If you guys want to check out the school, it's half off right now. It's going to go to 100 when we launch ivibewithi.com, which is my own rolled version of school. It'll have a bunch of AI features because like how are you going to have an AI community without AI? And uh yeah, it's going to be dope. So, I'm excited about that. Go check out the main channel. Make sure you like the video. I'll wait like five seconds. Look at my hands. Don't look at them too hard. You should be liking the video. Subscribe and I'll see you in the next one.