How to Delete AI Tool Anxiety in Coding (Stick to the Primitives)

This one tactic has kept me sane as AI is just pile driving the industry. Especially as software engineers, we're always looking for the next thing, the next thing, the next thing. It gets overwhelming. And so I've spent a ton of time, like the last 3 years, just looking to figure out how to use AI as best as I can. My background is in technology. I led tech for a start that sold for $23 million. and then I've done startup stuff since consulting all sorts of tech and I want to dive into it. So the gist of it is yes there's stuff that comes out every day but what are the things that we can do as developers as builders to stay ahead and the gist of it is you pick tools that aren't going to get destroyed and you pick primitives and patterns and workflows that can benefit from all these updates. not get destroyed like I said. So, first, yes, we have all these updates from Google coming out. This is a daily upload channel and I'm going to be changing the format of this a little bit. It's going to be more build centric. So, I'm not just a news parrot. I can have news be automated and brought into our private Discord, which is something we're working on now, or I'm working on now with our Discord road map and a bunch of different agents or really just bots that are doing that for us. So, I don't want to be a bot, but AI is going crazy, especially with Google. You guys can go and watch. There's a nice 10-minute video that Google put out, but AI mode was finally released with Google search. The visual search is exploding. Their agent shopping mode is scaring every marketer on the planet. Every SEO person is very scared. They have better models. They did not put out a new one for text generation, but they have diffusion coming, which is exciting. It's nearly instant. You can go and check out all this stuff. But you can even see their smaller models are nearly as good as 37. And then if you go to artificial analysis, they have a bunch of benchmarks on this. So we can go and see where things are at. Now 25 flash in terms of speed is obviously destroying, but the diffusion one's supposed to be like five times faster, which is crazy. And you see where things are at. You can go I'm not going to talk about that anymore. You can go check that out for yourself. But what it does mean is Gemini diffusion to me is a tell on generative UI. And if we go back a little bit, there was Nex.js United Airlines demo. Next had put out a demo where you could stream in the ability to book a flight. And it was interesting. This is a flight tracker. It was interesting because you could see how UI would be brought in as you're doing stuff, but I would never use that. But what this does mean is we went from let's say on the term of Gen AI or Gen UI sorry generative UI. I think the V1 looked like this where you could have a chat bubble and you'd say weather and then well technically it should be over here but you'd have a chat bubble and then it would stream in a component. And that to me isn't special. That's just Zod validation. If you're using TypeScript, that's how you would do it. It's a tool call plus ZOD validation plus an LLM that needs to determine what it is. So, I actually ran into this problem personally when I was building my own software where it's not deterministic. It's super fluffy and you don't want any of those when you're building software. You want it to be objective on a lot of the things and then offload certain workflows to AI, not everything. But I think that that's this demo is going to be part of the hype. And so in this case, this was a component that would stream in based on whatever command it was you were doing. So you'd say what your goals are and then we' break them down and create the actions and the to-dos and the tasks and just threw them across the board. And then other things like, oh, you wanted health information and if it knew that you wanted to talk about health information, it would pull in health information. Or another thing with this where it's, oh, you're generating a goal plan. Depending upon what the conditions of the query are, it would show you different components. And so I think that was where we were about a year ago, maybe a year and a half ago. Pretty cool. But I think it's much bigger than that. I think what diffusion allows us to do is there'll be some MCP equivalent protocol where in the future I just want to imagine where as a software provider let's say you're the SAS and then you have over here you have the user right and the user they may get access to five features because those are the ones that they've built. So let's say the SAS has these five features that are baked means the front end, the back end, they're all there. But there might be other features that the SAS could technically have because the API, let's say in a use case of a YouTube scheduling uploader where you go, you upload the raw video as soon as I finish here. This is what I'm building. So I'm using that use case. I upload the video to the React drop zone and then it kicks off a flow where it generates thumbnails. It generates subtitles and chapter markers and descriptions and blog posts, all these things. Let's say I have all those five built that I've built as the person building the SAS. But then there's 12 other features that are not yet done. So other pieces of information in an API that maybe I've done the backend for but I have not done the front end for or maybe I just didn't want to do the front end for yet. So in that case if we had these 12 other features where let's say transcription to other languages. So, transcription, we have different languages. If that's just something that I didn't put in the front end because I don't want everyone to have it, it could be, and this is almost like a feature flag, but that would mean that the SAS provider had to build it. What I'm getting at is you could have functionality that is not yet done on the front end that you could make available for the user to then generate UI around it. And I think that's pretty cool. where if they wanted to, then you could say, "Do you want to add features to it?" And I think the benefit to this is that you'd have to put guard rails around it. This isn't a totally baked idea. I'm just thinking out loud here. The idea would be that you really get these Lego blocks where it would show you what it has API support for and what it has tests for. And I think that that's where it goes where you could hit the add button and or you'd see like the list of them. So you'd have the API support has like a check mark or whatever and then you have a list of all the different features that you could add and that would break out and then allow that user to make the software more personable. And for the user that's great because they can make it look like they want it. I think that it would look something maybe like this where let's say again in this case of the transcription software we had feature one two three four and five but then if there's some way to edit it maybe it pops something open like this and then they can just say in plain linguish plain English exactly what they want and then they could move things around and this is maybe 10% of the way there. I just think that this is where it goes. Something like this where from a high level, hey, we noticed that this SAS can do more and the developer of it allows for you to do. Do you want to do it? And I think it would increase the retention of the user because that'd be just novel and new and interesting. But I think that's where it goes. That's what the diffusion stuff unlocks to me. You can see this is what I was speaking about earlier where it's going to transform the way that we interact with the web, the way that we interact as developers building this stuff. Someone's going to make a protocol for this. The first version, like I said, was this where it's a tool call and then you're passing this in. This is a lot slower than just looking at your phone or clicking on the top right of your Mac. So, that's on the new stuff on other things that are going on. found a really interesting repo. If you're interesting in in cyber security and all of that stuff, this guy is worth a look. He rebuilt Wapalyzer and open sourced it and made it into a cle or CLI tool. What Wapalyzer does is it analyzes the technology stack of any given URL. And why this is interesting is because you could just do Wapalizer space URL- I and then URLs.ext text and you could pass it all of your competitors. You could pass it any one of the SAS products that you're interested in building. And it's not perfect, but it does a pretty good job. So, you can scan single URLs, multiple URL, scan with author authentication, which is important because you can't get into the product and actually wapize something without doing that. And he's got a bunch of other interesting repositories in here. He's worth a look. other things that are going on. I'm building a bunch of Discord automations so that I can go deeper on the things that I care about and not spend the time putting tweets on this board and telling you stuff that frankly I'll miss things. So, I need to make the product offering better in our Discord where you'll have different feeds. And I was testing this out yesterday. Okay, the first one that I'm making is called Yoda, but I had made one or sorry, I brought one in that someone else had built. It's the founder or not the founder, but this guy that runs this dev mode Discord has all these watchers where you can see when new pages are added to different models. So, he has Google lab, Gemini web, cloud web, copilot web, new models, app updates, webatcher, onboarding, all these things. And I just found his GitHub and then found that he had this bot that was called I think it was Li Lyra. Yeah, there it is. And it used Gemini 2.0 to make small versions of Rag essentially. It didn't work very well. I think there's room for improvement. There's only 34 users that have it, so it's very new. But it got me thinking. One of the problems to solve when you're running a community that has lots of messages is if you come back after a day at work even, it's hard to catch up. And then if you're coming in after a week, it's even harder to catch up. And that's the nature of chat. And I think there's going to be certain channels where we want to have roundups where you can have if there's an automation called the news bird that's putting in all the things that matter from trusted sources, then you could come in here and just say either have it on a schedule where every week it will just do a roundup of all the things and the interesting pieces of news with the prompt or a user can actually ask Yoda uh what's going on in the AI world and set a duration for that. And then ultimately it's just more helpful. It provides a better member experience. And this same pattern can be used for a lot of different things. So I think that for prompting a lot of people share their workflows. So the creative workflows really loud, dev workflows really loud. And that goes back to what I want to talk about next, which is workflows and how do we stay ahead. So when I look at the new tech that's coming out from VS Code, I get really impressed. The fact that they're open sourcing is a huge signal to every developer, huge signal to all the VS Code forks out there. Got to be a scary thing for them. Their latest release, and by the way, open source that comes next month. We have some people with some info from some certain companies in our group. Got confirmation there. I'm sure they announced it publicly, but the latest release allows for instruction files locations and just instructions in general. So, it's following this pattern where we've seen this before. Everyone has one. There's claude.md for cloud code. There's client prompts that you can have, cursor rules, all these things. This is their version of cursor rules, but it's done better. And if I show you a basic implementation of this, this is going to help my dev tooling and every person that is developer because I had originally thought, oh, I'm going to make a quick CLI that just goes through the prompt chain. But now I don't have to do that. I don't have to ever do that again. But I did luckily learn how to write basic CLI stuff over the weekend. But when I look at this now, you can create a GitHub instructions directory and that instructions directory has a instructions syntax so that it knows it's an instruction and then those instructions can be brought in natively from your context. So you can see all the different instructions here. If I click on create a new instructions file, it will ask me where I want to go. I can click instructions. Then I can name that. So this is cursor rules done better. And you can also see that it has the glob application at the top because it's native markdown. It's just vanilla markdown. You don't notice any performance issues. Cursor was using MDC. And all these small changes just get away from the ability to have a performant app. And I hate to say it, but I've stopped using cursor as much, having used it since it came out, because I think that they're at a point now where they're adding so many things. They're under a lot of pressure, background agents trying to come out, all these different things. But I've just had so many times where it's crashed. And then I jumped to Zed and Zed, I ran a trial on it. I enjoyed using 3.7 Sonnet with their tune of it, but it was also crashing. And their kernel pattern for Python's a little bit confusing, too. VS Code's now at a place where their product velocity is on fire. And I'm excited to be using it. I'm looking into all the different things. I'm checking out, okay, even when it comes to different profiles, making sure that I'm using different profiles for different things so that it makes the app run more performant. And if you're not using profiles, there's a bunch of templates in there, but it allows you to select which llinters, which extensions, which tabs you want to have, the whole layout associated. So, right now, I just have my default one, but then I have a Python one, which is important for me because I want everything to just be dialed in when I open that up. There's different tools obviously they use for different stacks, and it makes it a lot easier to get going. So, that's another thing they added, the instruction files. They did a video where they did a deep dive and I can just pull that from our co-pilot channel where I did a breakdown of one of their videos. But in short, they launched a Devon competitor in their co-pilot on GitHub. They released huge updates to agent mode so that the co-pilot is so deeply integrated it can fire off separate panels. Example attaching test frameworks. So it will look at your Python work and it will say hey do you want to add this chain and that's even without instruction rules. They give some recommendations on models. So they say 2.5 pro from for all planning 04 mini for Python. I had made a video about that before that was the big difference in performance is if you're using Python then 04 mini is great. And then 3.5 sonnet for writing front end. This was however filmed before into the new flash. They also dropped launch and task support for agent mode, which is really nice. And then I already mentioned this one. Here's a screenshot of it with the configuring test button. That's the source. And then yes, you have the three core modes of ask, edit, and agent. I think it's actually more intuitive than the competitors, which is crazy. And you can also instruct it to keep a running change log. So it will update change log.md every time it makes a change creating a history of its actions. Very similar to memory bank which I think is great and I'm just really excited about everything that's going on. I think this is great. Another one that I really like is if you put in tag chat participant right here, then you can see all of the different extensions that have native commands slashcomands for VS Code. So, for instance, I'm a Fast API noob and I want to have the latest Fast API documentation without using an MCP because I'm allergic to MCPs. Someone should sell me on MCPS. I don't get it. It's they're so janky. With this extension, you want to look at the last release date and make sure that it does have the latest stuff. So, in this case, yeah, this one does, but you can I guess it's a few months behind actually, but then in that case, you can just do an at symbol and then get direct information from them. So, that is amazing. I actually misread that date. So, I'm annoyed that I I I misread that. But there's other things that I'm thinking about with context. Context is a big problem to solve. I'm going to be giving Augment a fair shake in the near future. I'm going to be talking about them a fair amount and see if they've solved that problem because I still think that's one of the few problems that has really not been addressed at all. But I'm really really optimistic about everything that's going on here. In other news, I am also changing the things that I'm doing with the community. Considering doing a rename because of the confusion around the word vibe. A lot of junior developers or people that have vibe coded think that they are engineers because they've made a basic website and then they get upset when they need to learn stuff. And then on the other end of the spectrum, you have a lot of developers who don't know when to give AI the wheel and when to take the wheel. So they just don't give it the wheel. And then they're upset at the people that claim to be engineers. So reconsidering this, reconsidering the content strategy as well. So less focus on just day-to-day news and more about the learnings of applying the tools as an engineer and target audience being people that are working at companies that are engineers or they're engineer adjacent that are trying to leverage AI more or people that are very new to this space but are serious. They recognize the challenge. They recognize that they need to learn. And then the third cohort that I've noticed is business owners that are willing to put in the time because they understand the value gained using AI. So, we're going to be doing that, changing the positioning, and then as part of the offer, the Discord will be beefed up with a bunch of the automations so you can always know what's going on without having to watch six hours of YouTube videos or me every single day. And then on top of that, a bunch of the other ones I mentioned around the contextual chat bots to help keep up with things. and then a membership site. And membership site will have the directory and other things that connect into Discord. It'll be single sign on with Discord so that you don't have to deal with pesky different login. And then the last part that I'm most excited about is actually just the ability to open source a lot of these different repositories because we have builders that are excited to build things. We should have repositories where we can collaborate. So I've set up different project templates. Very simple. looking at the way that I think Taskmaster did a really good job with the way that they're doing open source. I've also looked at cal.com the way that they do open source stuff but we'll be using these project templates issue templates and then the following will be a part of the open- source stuff. So we will have map which is a B toC multi-agentic platform. We will have parts of echko which is B2B content automation. We will have a prompt library. Oops. Nice prompt library where I'm looking to also have some way that when a prompt is added to GitHub that it actually runs the outputs for it across different models. And then as this grows and as we have more and more projects that people want to collaborate on, we'll be beefing up the support for the organization at the organization level. we'll have code rabbit or whatever it is that needs to be a part of it so that it's a better experience and then that plays back into the content where I'll be talking about the projects that we're working on. It'll be open source in that anyone that is watching one of these videos can go and view all the information. They can make comments on it, but they won't be able to contribute. But then anyone that's a member can contribute. And I think this is the way where it actually brings people together. It actually builds community and then ultimately people can chase that outcome of a lot of us want to build SAS products and want to make money with these things. So I'm really excited about all this stuff. If you learned anything this video, make sure you subscribe to the channel. You should like the video so that you see more of them in the future. Feel free to check out Vibe with AI. The link is in the description. And I'll see you in the next one. Bye.