If you watch this channel, you know I spend a lot of time testing local models and running benchmarks on hardware. I love that stuff. But, here's the
reality. When I'm working on a project that doesn't have strict privacy restrictions, I still reach for the frontier models and cloud services for the
heavy lifting. And I don't think that's going away anytime soon. So, today I'm walking through the
AI tools and subscriptions I actually pay for and use on a regular basis. But, before I do, just a little disclosure, I use all these tools, so I
reached out to three of them to see if they want to sponsor this video, and they agreed. So, for the sake of transparency, the ones that sponsored,
even though I generally use them myself, I'll have a little indicator scrolling across the bottom of the
screen. So, you always know which one is which. Let's kick things off with the heavy hitters. First, the one I've been using the longest. This one
always has a special place in my heart. It's GitHub Copilot. This was my entry point into AI-assisted coding, and it's still there in my IDE every
single day. The inline completions, the chat, the way it just fits into the flow of writing code. Now,
it's got agent built in. It's one of those tools where I notice the most when it's not there. It's not flashy, but it is reliable, and that counts for
a lot. It also gives you access to a whole ton of models from Google Gemini, Claude Sonnet, Claude Opus, GPT, and even local models. I've made videos
about that on the channel. Next is ChatGPT. No surprise there, right? This is my general purpose
go-to. I even have it set up on my iPhone to be this button right here. It just takes me right into it. Research, brainstorming, breaking down
problems, working with images. If I need a quick answer, or I want to think through something out loud, this is usually where I start. The ecosystem
around it is massive, and for a lot of everyday tasks, it just works. I used to have a pro subscription,
which was $200 a month. It was useful, but not $200 a month useful for me. So, that's where Claude came in. I'm on the max plan right now. By the way,
you can probably see from my open tabs at the top the mix of ChatGPT, Gemini, and Claude that I use. Claude is where I go for some longer context,
more nuanced code review, especially writing. Claude handles that really well. I don't think either of
it is better or worse than chat GPT. I just think of them as a little bit of a different tool for different jobs. Even pasting each other's output
into each tool to see what one has to say about the other one. So, those are the three core subscriptions that have been at the root of my work for
the last few years. Now, let me get into some of the other tools that have worked their way into how I
actually get things done now. See, there's a lot of different models out there. Paying for all of them is just a pain. This is chat LLM Teams from
Abacus AI. And I ended up using this for a pretty simple reason. See, I do use a lot of image, video, and speech generation. And especially for video,
trying to remember what model does what best is not that easy. And I definitely didn't want to pay for
a bunch of separate subscriptions just to test different video models once in a while. So, with chat LLM, I can go into one place. I can jump between
models like Claude. Yeah, they're all in here. Gemini, Grok, even some of the open weight models like DeepSeek, R1, and V3.2, Qwen 3.6. Oh, that's
new. That wasn't there yesterday. We've got Kimiko 2.5, GLM 5. I can do all these without juggling all
that separately. And what I like is that it's not just text. I can use it for normal prompting, sure, and then compare output side by side, but the
image generation is really cool. Because I have access to all these: GPT Image 1.5, Nano Banana, Seed Dream, Flux, Qwen 2.7, and video generation like
Sora 2, which hopefully will stick around for a little longer. We've got Kling here. We've got Veo,
Qwen 2.2, and Qwen 2.5, Seed Dance. I like to try a bunch of them when I want to test an effect or rough out a visual idea. That's what actually
solved the workflow problem for me. It gives me options without adding overhead. There's also Deep Agent. I've actually shown this on my channel
before. This is their agent that can build full stack apps with APIs and payment support. Sure, as developers
we can do all that ourselves, but this also helps presentations and reports, which is something I'm pretty happy to hand off to somebody else. 10
bucks a month is pretty crazy considering how many separate tools this would otherwise use. I'll leave a link to this down below. It's chat.abacus.ai.
All right, now sometimes I don't want a broad AI search results. I want answers based on the higher
quality things that I already saved and trust. So, for the past year or so, I've been using Recall as the place where I save stuff that I actually
don't want to lose. Yeah, it's mostly YouTube videos, but also notable articles and other things I know I'm going to want later. This graph view is
kind of cool. Let's click on Nvidia. These are all the different connections in my knowledge base about
Nvidia and all the different articles and videos that I've saved. So, I can hop around. Now, what made it stick for me is that it's fast enough to use
from both my computer and my phone. So, if I come across something useful, I can just throw it into Recall and keep moving. Then later, instead of
digging back through tabs, bookmarks, or old messages just to remember where I saw something, it's
already there, summarized and organized. I can chat with videos. I have a reader with links to the video. I have detailed summary, short summary,
connections to other related things in my knowledge database. And what's interesting now with Recall 2.0 is that it's becoming a lot more than just a
place where I save things. The big feature here is agent is where I can talk to my own saved knowledge
using this Recall selection, the whole internet, or both of them together. And that matters because sometimes I don't want the general AI web search
giving me a bunch of junk. Sometimes I want to limit the conversation to Recall, which is the high quality things that I've already saved and trust.
So, for example, I could ask something like, "What's the latest in AI hardware from AMD?" From my
knowledge base. This goes through and tells me all the details that I need to know based on everything I've saved with links. If there's something
newer that's not in the database, I can switch to web and do a follow-up. What about the Ryzen AI 400 series? And now it expands beyond my knowledge
and enriches it with the web. Nice, there's a lineup. I also like that I can choose the model I want
instead of being locked into one. And the new extension looks pretty cool, too. I'm usually not a huge browser extension person, but this one might
actually make me change my mind because it brings so much of the workflow right into the browser. Like one-click summary, take my notes, chat
specifically with my content. So for me, Recall has already been a useful capture tool. And with 2.0, it
starts becoming something more of a real knowledge engine around the stuff that I've already collected. And it's got API and MCP access. It can
connect into other workflows, too. If you haven't tried Recall already, check it out for free or use my link down below. Make sure you use the code
AZIZK25 for 25% off the uncapped version. Now here's one that's less about chatting with AI and more about
using it to solve an actual production problem that I have. As you know, I make videos on this channel. And one workflow I've been using lately is
with Waymark Video API. And for me, the big win is not really the UI. Sure, you can paste the link to a published video here. It'll find clips for
you. It can even generate shorts. All that stuff is great, but it's the fact that I can program against
its API that really makes a difference. So I bring this up in VS Code, pull in my already published video, and through its API, I have Waymark Video
generate candidate clips from it. Give me some JSON results back, the timings of each clip, and I let the Waymark Video handle the AI part that
figures all that out for me automatically. Then I just feed that into my own pipeline that turns those
results into JSON and builds out a Premiere Pro project. And that gives me a lot more flexibility than doing everything manually. So here are all the
clips on the timeline. I can adjust their timings. I can modify the clips, add graphics. If I want vertical format video, no problem. There we go. We
in video found this for me and using their API I further processed it into my own automated system
where I generate these shorts. I export the ones I like, tweak them however I want, and publish. And honestly, the most annoying part of doing
something like this is not transcription or summarization. I already have my own workflows for those. But it's going back through older videos and
finding the moments or quotes that are actually worth resurfacing. What's been useful here is that I can use
their find moments API to search for the kind of scene I want, like a strong quote or a reaction or a key insight, and it can surface those moments
directly from the video. And from there, I feed those results into the clipping workflow and generate short from clips that I can review and refine
inside Premiere Pro. That's where this API has been useful for me. It saves me a lot of editing time and
it helps turn one long video into multiple reusable clips without the boring part of it being fully manual. Even though I do like to edit and do fun
things with editing, this gives me a lot of starting ideas. All right. So, I already mentioned Claude and ChatGPT as core subscriptions, but both of
those have expanded into something bigger, dedicated coding agents. Not telling you anything new. I
know you already know this stuff, but just for the sake of completeness. These have honestly changed how I work more than anything else on the list.
First is Claude Code, and I swear I've never had this many terminal windows open. I was kind of a late bloomer actually on this one, and I only
started using it a few months ago. But once I did, it clicked fast. My go-to use case is starting things
from scratch. If I'm building a new project, scaffolding something out, or prototyping an idea, Claude Code is where I go. Describe what I want,
iterate on it in the terminal, and I can go from nothing to a working project surprisingly fast. Basically, for greenfield work, it's been a
game-changer. Then there's Codex. This one I jumped on immediately when it launched. And for me, Codex lives in a
different part of my workflow. I use it for everything tied to my existing Git repos. Pull requests, refactors, working through issues on hosted code
bases, it's really good at understanding the context of a project that's already built out and making targeted changes. This tool kind of sits between
Claude Code, which is entirely terminal-based, and something like Copilot, where it lives inside
your IDE. Now, here's the thing. These tools are pretty much interchangeable at this point. You could absolutely use either one for either job. And
the way I split them is really just personal ergonomics. It's kind of like being an old dog doing the same trick. Once you build a habit around a
tool, you kind of tend to stick with it. So, Claude Code for new stuff, Codex for existing repos, that's
just the way it works for me. Your split might look completely different, and that's fine. So, that's the list. GitHub Copilot, ChatGPT, Claude, and
the coding agents that extended them, plus a few of the tools that have solved real problems in my own workflow. And it's pretty simple. I run local
models when I need privacy or I'm experimenting with hardware, but for the actual heavy lifting, the
daily driver stuff, these are the services that keep earning my money. What does your setup look like? I'm curious. Drop a comment down below and let
me know which AI tools are you actually paying for and what stuck around for more than just a month. I'm always looking for something I might have
missed. And if this was helpful to you, you know what to do. Here's a video on doing some local model
stuff, and I'll see you next time.