Auto-generated transcript If you're not using your entire weekly quota of LLM tokens, you're gonna get left behind. This is a screenshot of my weekly clot limit, and you can see over here that I've used like 95% of it. And these models are so good, they have changed my workflow so much, and I'm really excited to go through some of the wacky shit that I'm doing with these models, because they are so friggin good. And I'll be honest, I'm feeling the AGI right now. Like, the things that I get these models to do now is insane. And so really, if you're not using these, if you are not using your entire weekly limit, you're probably not using these models to the full extent of their capabilities. For instance, this right here is a skill file that I created with Claude to help me view my university lectures online. And so literally what I do is I open up Claude in my Chrome browser. I'll just activate this skill when I'm on the university website, you know, and it's just going to view all those lectures for me. And as you can see, I'm a shitty student. I'm not going to watch those lectures by myself, but I do need to keep up my attendance, right? Claude just does it all for me. Literally, all I did was go over to the website, turn on Claude in the Chrome browser, ask it to view the lectures for me. It did. And then I turned that into a skill file that I can then use repetitively anytime I want to, you know, make Claude watch my lectures for me. And I made sure that the instructions in that skill file are so clear and easy to follow that even a cheap model like Haiku can run them for me. And that's just one of the many ways that I'm using these models. Another one is, you guys probably know, I work on Upwork as a freelance AI automation guy and full stack developer and all that. And let's say I'm trying to search for jobs over here. I can apply up some basic filters over here, like make sure to hire someone or make sure to find someone who has verified their payment info, who has some hiring history and all that. I can apply some basic filters over here, right? But these are still like, you know, 50 different jobs for me to go through. I could go through all of these manually or I could just open up Cloud over here. I could tell it exactly the kind of jobs I looking for and then I would just let it filter out the other crap that I don really need to look into anymore right tried this once and it just filtered out like 50 jobs and of those 50 of those 50 jobs it gave me like eight of them that matched my exact filters filters that don't exist over here in the sidebar or filters that are just based on subtle nuance in language and stuff in the job descriptions I just asked Claude to filter out all those other jobs that aren't relevant to me and just give me the ones that are, right? It went through like 50 jobs, gave me just eight of them to review, way less work for me and way easier for me to apply to jobs that actually match my expertise. And again, since this is something that I'm going to be doing often, I turned it into a skill that even a tiny model like Haiku can run. And I do this pretty often. Like when I've got a repetitive workflow, I try to turn it into a very easy to follow skill that even tiny models can follow. This just means that I can get a lot of repetitive work done, one. But two, also not let it affect my usage limit way too much because models like Haiku, you could probably run them 24-7 and still not make a dent in your usage limit because they're just so cheap to run, right? And so if you can make skill files to handle repetitive workflows that even models like Haiku can run reliably, that's very, very high ROI. What else? What else? I've been doing a lot of coding on my phone recently I'll be honest like I just have cloud code open over here as an example but I barely even open my terminal anymore and I barely open up cloud code on my machine anymore I just like you know open up a remote session and just use that on my phone like most of my coding is being done on a bike or on a treadmill nowadays honestly just from my phone and I can do this because the harness the tools the models everything is so great and work so well together that I don't even need to be on my computer to do coding anymore, which is insane, right? So I'm basically just doing like cardio for like 45-60 minutes a day and I'm doing all my coding work alongside that, running like five different sessions and getting more work done than I would have done if I was coding by hand for like 10 hours on my computer over here I getting way more cardio I doing a lot more coding Everything working out nicely Another thing that I've been doing is using agents to run all of my automation workflows. So if you follow this channel for a while, you probably know that I have a ton of automation workflows. I build a lot of CLIs and tools to handle repetitive work for me. Most of these are deterministic like Python scripts and stuff like my phantom editor cli over here i've talked about this a hundred times already you probably know how it already works basically i can run all these different commands to do all sorts of different things for you know content creation automation and video editing and all that stuff and i've got clis like this for like tons of different things and you know it's very easy for me to go over here and type something like phantom projects process dash dash raw and then pass in a video file over here that i just recorded and then just let it do all the editing and processing for me. Pretty simple, right? Just five words for me to type. I could do that, or I could just have a really tiny skill file that tells Cloud, hey, I just recorded a new video, go organize it for me, go process it for me, go set it up for uploading and all that other stuff. And that's one of the things I did. And so the mindset shift over here is that I'm letting my automation workflows be kind of like primitives for Cloud Code to use. Instead of me having to type the commands, I just build the deterministic workflows like all these Python scripts, so that I have determinism in my automation workflows. And also so that the LLMs don't have to spend as much tokens to achieve the same tasks, because we've, you know, extracted most of the work to deterministic scripts anyway. So I set them up as primitives, and then I just let Cloud Code run them for me. An example of that would be this publish loom skill that I created. So this is a pretty long document, I'm sure you don't want to go and read through this entirely anyway. But what it basically does is, I record a lot of videos, right? And I record a lot of Loom videos over here as well, as you can see, for client outreach and all the other things that I'm doing. And I set up a bunch of primitives that can take these videos process them edit them cut out all the bad stuff you know run some compression on them then upload them to my S3 bucket online do all the other stuff they need to do, and then give me a link to those videos that I can then share with clients myself, right? And to do all of that, I have to, you know, I have to run a bunch of commands over here, right? Because I've got CLIs to do all these things for me, but I have to run like three or four commands to do all that. I could do that, or I could just use this skill, let something like Haiku run it for me, and it'll just, you know, take all of the videos over here. It'll process them, edit them, upload them, and it'll give me the links to them. I don't have to do anything besides just tell it to activate the skill, and it'll handle all of it for me after that. It already has all of the deterministic scripts it needs. It already has all the primitives it needs. It's going to cost very little in tokens because it already has all the script it needs to run. It just makes it way more convenient for me to do all these things because I don't have to run any commands myself. I just let my agent do it for me. And so that's another really cool thing that I'm doing with these models. And my hope with all of this is to kind of just inspire you to use these models more and more because they're getting very intelligent. And I'm not just talking about Opus 5.5. Even the smaller models like Sonnet and Haiku are getting way better at tool calls, at following instructions and skills and stuff like that. And if you're not using your entire limit every single week, you're probably not using these models to their fullest capabilities. And so you should probably think about how you can do that. Because I promise you, it is so worth it. I haven't even scratched the surface of what these models are capable of, and I'm already using up most of my limits every single week. So I can promise you, if you have like 50 or 40 or 30% of your weekly limits left every single week, at the end of every single week, you're leaving a lot on the table. And you should probably try to use these models more to get even more work done. That's all for this one. Just really excited about how amazing these models are and what a time we live in. And I just wanted to talk about that. I'm probably going to make a couple more videos on exactly what I'm doing with that's going to be in the very near future. Till then, like, comment, subscribe, and see ya.