I Rebuilt a SaaS in 2 Days with AI - Here's My Learning Framework

If you're not growing, you're dying. If you're not learning, you're not getting better. You're just literally decaying. It doesn't matter if you're just getting started. Or if you've been doing it this for 3 years, 5 years, 10 years, you need to stay on it. So, the traditional advice is, oh, just learn by doing. But what does that actually look like tactically? That's what we're covering today. This is a daily upload channel. It is May 30th. We cover a couple of news pieces and my takes on them so that it's evergreen. I don't just do little news parrot stuff. Then I cover some questions and then we get into the concept that I just mentioned. So first of all I've been talking about augment a fair amount because it's the first one that does end to end ideiation to testing to CI/CD to core integrations with tools that you love like GitHub and mermaid and notion and linear and all these things. So I just wanted to call that out. Second thing is Gemini Pro obviously really good. This guy Farsa he ran buildspace very popular program in San Francisco. RIP build space, but he made this thing that basically counts his shots made and crazy. And it's a oneshot in Gemini. I asked him if it's open source. That's what he typed in and I gave him the code. Crazy. And I made this meme. I thought it was funny because cloud code can be overzealous and oh, we definitely fixed this. And then I'm staring at this deploy thing just being like, please. And what I did with Augment in two days was go from building out the original version of VI 2. VI is our community which will be paired with a bunch of SAS. We are launching a membership site that has a bunch of things in it. You can see the earliest versions are here. I'm just please ignore how ugly this is right now. But the scaffolding's done. Python back end with fast API have the Google ADK in it as learning paths. bring your own key prompt library by popularity, model, input, output, intent, actions, the member directory, so you can see if people are open to collaborating in projects, what their skills are, where they're lacking, the integrations, so you can connect things to make your profile better, and then settings, all that normal stuff. So, this is functional in that all the third party services are there and the endpoints and whatnot. Now I need to go and connect them and make it not look horrendous. But yeah, because of augment I was able to basically go from this setup where I had Astro front end. Wanted to play around with that using Netlfi for serless functions cuz I had never used them before and then discord bot services in the client layer which then spoke to Superbase for O Superbase for the DB. So managing stuff like payments, membership status, user info, all that. Stripe API with their web hooks for payments and a Discord bot that kind of is like the policeman for membership but among many other things like getting summarizations on the message threads using vertex as a search index. Gemini AI API yada yada. So fair amount of stuff very simple project though and then moved it over to this. So now the back end has discord bots that are integrated into fast API. We have the fast API back end that does all the AI ops and will also speak to any of the external services. This is a one piece that I'm really excited about which is hotel open telemetry. Shout out Lindsay. His last name is Rex. He's a member but he's a DevOps G. And he was telling me all about these self-healing agentic workflows and I was like whoa that sounds bananas. And he's you got to check out Dino. It has a built-in, but you don't have to have Dino to do this. So, it will be piping down information through a Discord bot and pinging me and private channels when there's things going on or things I want to know about. Postgress, not much change there. And then instead of Astro and Tanstack Start, now it's Nex.js because of the LM experience and the ability for remote agents and the team that I'm building out to be able to do that. And I could, no matter how much I tried to guide them with documentation and examples, it's just better at writing next code. And then you have engine X reverse proxy running on Debian and that's got some containers. Yeah. So that's all exciting. Next thing I came across was just the product requirements and product management role in general is just totally changing. So, they were just pitching this project basically, and I want to just see how it works so that I can make one little Discord bot out of it that does a lot of these things for you as a test. I think of Discord just basically as like front end for agent stuff and then you can test it there really fast. And that's why I built this out the way that I built it so that anyone that is developer in VI can go and make an agent and it'll sit on there. There's actually an endpoint where you can go and create one for the bot registry which is cool. And then last thing I came across I thought was interesting was this data set marketplace. You can basically it's like firecrawl on steroids plus data sets and then all these tools. So, it's called bright data and you can scrape as markdown, scrape as HTML, have sh session stats, and then pull web data from any of these things without having an API key, which is really nice for X. Cool. So, let's talk the actual content of the video. The main lesson that I've learned is when you're trying to learn something, you need to do it by doing. A lot of the time that can sound challenging. So, you want to treat your day like you're Larry David from Ker Enthusiasm or like you get annoyed by really small things and then you write those down. And is there a tech solution for this? They might not be commercially viable, but it'll teach you the practice of coming up with ideas and then you have a big note and you go through those and you start to try to build things and eventually you'll come across one that you want to do. And I highly recommend checking out these lists that I just organized cuz this is how I do it. I have not filled all of them out, but you can just scroll through my stars. What I do is I just saw someone else that had organized this way and I thought it was a good idea where you have the AI engineering stuff. So in here you have a vision agent. If you wanted to go learn how those work in Python, you can do that. If you wanted to see how the easiest way to deploy agents, models, rag, pipelines, and more with no ML ops, and no YAML, you can go check that out. Awesome cursor rules, fast API, MCP might start sneezing. I'm allergic to MCPS. And then there's a bunch of courses. So this decoding ML has a bunch of stuff on how you would make like a second brain assistant course. I'm going to actually move that out of here and into learning and then the open AI agents Python ADK by Google. So you can check those out. You can also check out the awesome list of awesome lists. So if you're familiar with awesome lists, then you know what I'm talking about. You can see every single NAN flow. There's like thousands of them in there. Or a bunch of opinionated Python frameworks, libraries, softwares, and resources. all the prompts for the client community, all the leaked different prompts. So, when you're trying to learn how to prompt, you can go and find these. A lot of them could be wrong, but then some of them are mislabeled as well or clawed. They didn't leak it like they just put it out there so you can just see it. Awesome. Tailwind CSS, awesome fast API, best of JS, that's another good list. You can mine some of them from me, but then also build your own lists. Like this Discord one's pretty good. That's how I was learning to do this. I'm thinking of building bots out that do things similar to this like Meridian project which I've spoken about before but it cuts through the noise and basically delivers a personalized brief because I think that's a better way to deliver news than to get let's say like a quarter hour update which I've also seen and yeah you basically just categorize them by this and then when you want to go and learn something if your problem matches to a similar technology that already exists then you can just pop one of these open and start asking questions with it with the chat with Copilot. It's a great way to learn. And then of course you should just pull this codebase down. What I do is I have a developer folder. And in the developer folder, I just organized this so I'm okay with looking at it. It was really bad before. But in developer folder, I'll have a tmp. So I just removed it. Like I said, there's nothing in it. But either temp or tmp. And that'll be full of these code bases that I then go and I have them side by side. And one hack too is if you're having problems with implementation, you can literally just drop the codebase that is your reference codebase for the concept that you're trying to learn inside of the codebase that you're working on and then use augment or whatever your code tool of choice is to run the comps on it and see what are the concepts that I'm missing and then have it generate a list of the concepts that you don't understand. That's a big piece is be humble with this. You're not trying to brag about what you know to an LLM. You're trying to learn from it. So that's what I recommend. You can see like some of the ones that I keep in here, the ones that I still want to look at. So it's like morphic. This is a project I haven't updated since November, but it taught me a lot about like multi-aggentic flows because it was open source perplexity. Then you have FFUF, which is empty right now, but this is a it basically allows you to programmatically use Golang to look at a website in every single route that is public and see it. And that's like a cool thing. I'm like, how does that work? And then you peel that open and take a look. We have other ones that are interesting in here. This site auditor project that I built. This was because I had done client work and I was like, man, that took 10 hours. I bet you I could build something that's a bunch of scripts that does the audit for me and then generates the different site results. So you can do stuff like that. And you basically just again you take those problems you have in your head, you start writing them down. Make it a practice to write them down. And then you go and you do just in time learning. Ask lots and lots of questions. Just treat it like the 24/7 tutor that it is and you'll be off to the races. If you like this video, make sure you like it. If you want learned just one thing, make sure you subscribe to the channel. This is a daily upload and I will see you in the next one.