Claude Cowork is Taking Over & More AI Use Cases

Ladies and gentlemen, welcome to another week in AI. We're starting to get all of the innovations that they were withholding for us throughout Christmas, New Year's and the first week of January. And I'm here to tell you about them and to try them, including more use cases of Claude's cowork product, their Agentix system. That is the first consumer product from one of the big labs that is actually trying to get things done rather than just assist you in doing things. And then there's a bunch of tools for creative uses, like new state of the art transcription models, a Chinese tool that allows you to move the camera in a scene and generate new images from that, all of that and so much more. In this week's episode of AI News, you can use the show that pulls together all the AI innovations. We test them, compare them, I show you the results and hopefully you walk away with something useful. Or at the very least, you know about all the bleeding edge releases in Generative AI. Somebody pointed out to me that the more time goes on, the more I look like this guy. Maybe like this. What is up with Tech Bros just transforming into this archetype? I don't know. I say we just look at the first story. All right, so let's start by talking about Claude. Now, if you want the full summary and my initial tests for it where I go through two different use cases step by step, check out the separate video I uploaded earlier this week. This is a follow up to that after a few days of usage and I'm going to talk about the patterns that I found while using it so you can get the most out of it too. So first up, a summary of the entire product is basically a friendlier Claude code. That's really the best way to put it. It's a simpler user interface that everybody can use. And a lot of people I've talked to see said something along the lines of this is the first time an agentic workflow feels achievable and feels useful. But in many ways it's just an evolution of what you might already know with Claude or ChatGPT. It is just more proactive and it does more. It does these to do lists in between. And let me now concretely talk to what works and what doesn't. Well, what works is the stuff that already worked. If you really didn't like the connectors to Gmail and Google Calendar, well, they didn't change that uses the same connector. So those won't work in a way that you expect them to just because we Have Claude coworkers work. Now, if you like the Claude Chrome extension and you use that well now, it integrates into here and it's so simple to use that everything you could do there, you will be able to do here. The problem with that is that the Chrome extension was limited in certain ways and that didn't change. Now, okay, I realize that for a lot of people that might be stating some obvious facts, but here's a thing that was not so obvious to me. Anthropics Claude had this release called Skills, where it basically created an entire skill that contained certain instructions or certain guidelines into a markdown file that then you can access at any point in time. And now in this interface, they've become so useful because they're really easy to create, they're really easy to use, they work in conjunction with everything else. Again, this is nothing we didn't have in a while. It just became so much simpler to use. And let me show you a concrete use case that you could do before. Talking about the things that I tried and I couldn't do, but talking about these skills. If you just tab over to Claude Cowork here and then you navigate to the settings, you can go to capabilities here. And all the way at the bottom of the capabilities, you see skills repeatable, custom, customizable instructions that Claude can follow in any chat. These work really well with Claude Cowork, and here in the settings, we can basically create them. So one recommendation that I would have is what worked well for me this week, and I've also seen other people do this successfully, is it's really good for content repurposing. Claude Cowork is really, really good for that. But you might want to start off with creating a skill to keep things consistent. So what I recommend is you just go here to add skill, you say create with Claude and I'm going to create a new skill with our AI Advantage brand guidelines baked into the skill. So whenever I want to work on content repurposing with Claude Cowork, I don't have to worry about brand consistency. So I happen to have a PDF with our brand guidelines where it outlines fonts and other best practices. And all I'm going to do is simply add those social media guidelines and I'm going to say I want the AI Advantage social media brand guidelines to be within a skill. And then as we're in the skill builder here that we got to through the settings, it will just build the this skill for us, which then we can use inside of Cowork. At any point in time. Okay, as it asks questions, I'll add a little bit of context. As per usual, the more specific you are, the better. All right, so that took a while, but you only have to do this once and it's really over delivered. Look, it has voice and tone presets, it has an entire repurposing workflow. I could delete parts of this or just keep it as is, which for the demo I will do. All you have to do here is hit this button, copy to your skills, and then it moves this markdown file with the skills in it into your skills, which now if I tab over to cowork, if I go into my settings here capabilities, you will see the AI advantage social skill. And then if I do something like turning my cloth cowork video into a carousel and I just provide it with the link and then it opens up the window and tries to go into the description, reads all the content here and ultimately from what I can tell here, it actually fails to fetch the transcript. So I would still need to do that manually by going here saying show transcript, toggling, timestamp, solve copying this entire thing. This way I could work with it, but see, these are the imperfections. Like it can't do everything, but it can do a lot as it just progresses throughout here, fearlessly use the skill that I just created. And then it creates an Instagram carousel with proper AI advantage branding. And in this case it doesn't even have an image generation API, so it can't create different slides. If you want to get creative and feel comfortable with this type of stuff, you could add a custom connector to a MCP that connects it to an image generation model. But there it is. Here are all the prompts that it wants me to run in an AI image generator. Here's the caption. And it even created a little artifact that is not great. It did its best. It really lacks the ability to create images here. But you know, research, preview and all of this will become so much more seamless soon. You won't need these custom setups, but this is what you can do right now. And to round up the segment, I want to share two more things. One of them is I tried a thing where it looked through the entire community we've been building over the past few months. I haven't been very vocal about it here yet. We'll talk about it soon on the channel. But basically we needed to go through hundreds of posts and look for the mentions of a specific name. And because something changed, we wanted to update it. I figured, okay, Claude Cowork could be perfect for this. So I let it do it. But as per usual, the Chrome extension wasn't perfect. It was pretty good. But sometimes it just got stuck and I needed to reprompt it. And after reprompting it like eight times, it did like a third of the task manually for I think two out of six spaces or something like that. So it just kind of worked. But that's not because Claude Cowork is not great. It's because the Chrome extension is not perfect and it kind of struggled within that application. But to abstract away from that and just round out the segment, I would love to say that the thing that this is really good at is batch processing. Something if you have a lot of repetitive work on something, whether it's lawyers with 80 different documents, they have to read through and summarize and create a new type of report import from it, or you keep repurposing content from format A to format B, well then just create a skill and use that together with Claude Coworker to do that consistently. Or any other knowledge work that is repetitive and where you have dozens of repetitions. That's where this friendly version of Claude code, or you could also call it a business process automation agent, really shines. It's already useful, but you need to get a little crafty. But it will only get more useful and user friendly from here, and we'll keep an eye on that. And hey, if you're enjoying this video, make sure to subscribe. We do this every week and if there's tools that really stand out, we create dedicated videos just on those. Okay, back to the next story. For this next one we have Scribe V2 from ElevenLabs. So if you're not familiar, this is Elevenlabs transcription tool, meaning you give it some audio and it turns it into text. Now, scribe v1 previous version of this obviously was already the state of the art model in transcribing on many dimensions. And they just released Scribe V2, making it even more better. This is mostly something for builders, but also for content creators. If you just want flawless transcription for many video files, this is the one to use. Yes, there's an API for developers, but also you can just try it in Elevenlabs studios, which is easy enough for everybody. So let's just try the demo. It will auto detect the language. So let's throw some curved balls at it. First of all, okay, that just works. But secondly, let's use some words like chatgpt, Claude Gemini, Xai. It got all of those right. Not bad. What happens if I switch to Deutsch? Where does a Kennen desert Jaffwammar nichtme of Englischen. Okay, it made a slight mistake there. It said canon instead of a canon, but that's fine. It's back in English. Okay, switching to Slovak mid sentence was a bit heavy. Maybe let's try one more time. Dobry j panie Krasnoicutum. That worked okay. Impressive. Arguably even better than the previous model. Just feeling wise. The benchmarks are obviously better and this thing has every feature that you could imagine. All the audio formats, super quick latency as you saw, and also features like speaker detection and all that. So if you need to turn multiple voice files into text, this is probably the most reliable way. And you could just batch process them here. And if there's some specific terms, you can add them here because it just has the stuff in a dictionary and a lot of AI terms obviously, because I got those right. But yeah, not subscribe. So next up, we have a new Midjourney release. It's a iteration on their anime focused model. And bear with me, because this thing generates visuals that are so impressively beautiful that you should see them. What we basically did is we ran our test prompts that we used to do. Now we do them less because everything turns out similarly on all AI image generation models. And we ran them on NI7. That's the name of the new anime model they have. Look at some of these results. It just has this like super unique esthetic where it's more anime esque than anything else. Seriously, look at this image. When I'm creating a presentation or something creative. I think I personally would actually prefer this style to, well, every other photorealistic result you will get here by default. I mean, to be fair, we're literally prompting for a cinematic still. But because this is an anime focused model, you will get more of that for portraits. That's not what you want to use it for. Logos, same thing. But if you lean into this anime capability, you might just be pleasantly surprised. And I just wanted to bring it to your attention. Our team member, Hayd, who always loves to explore and create interesting things with these models, shared some initial examples of one of his signature styles. And honestly, this is just one of the most unique looks you can get out of AI Midjourney, the newest anime model. Great way to differentiate yourself from all of the other AI generated stuff that most people can identify as such by now. Okay, next up, we have open source tool that we've seen iterations of in some creative tools. But this is Quen Image Edit 2511 and it allows you to do 3D camera control to generate new angles of an image and it demos really well. So let me show you. Okay. I just have this little AI image generation of me playing paddles, something that I've obsessed over for the past year. And here's the idea. You get to control these camera angles. So maybe let's do a more top down angle generate. That's the same thing. Maybe let's rotate all the way around. Yeah, there it is. Maybe it's just a bit laggy. I mean, that's not perfect. It's not even a paddle racket anymore. Maybe one more attempt, a slide switch. Okay. Yeah, I think it's just lagging behind a little bit, but you get the idea here. Is it perfect? No, nowhere close to perfect. But this is a hard challenge with glass and everything. I mean, the anatomy is fine, the net looks almost identical. You really need to start nitpicking to find the differences. Now when we tested it on a more realistic image, it actually works really well. So yeah, just a fun thing to play with. And I thought this was an interesting interface that I wanted to show you. Okay. And next up for this week's quick hits, there's a bunch of stories that are actually in the same theme and that is basically the big companies going after AI plus shopping. For example, we got Gemini Shopping, which is very similar to ChatGPT shopping, where they partner with different retailers and online stores. And then they turn the Gemini app into kind of a new interface for online shopping. Now obviously all of this stuff is problematic because you want to unbiased opinions. And when they have these partnerships and they kind of direct you to certain brands over others, well, that's not really a neutral recommendation, is it? We'll see how this develops over time. This is just getting started and there's a lot of competition, which is good for the consumer. Then Google also introduced something they call the Universal Commerce Protocol Open source framework that allows AI agents to handle the entire shopping journey. So maybe in the future when you combine that with something like Claude Cowork, it can kind of figure out what you need based off your email, your calendar, your messages, maybe your smart fridge and whatever, and like order new groceries and send messages to people and just organize things and do things. It's going to be really interesting when a lot of these technologies kind of collide. We'll keep an eye on that as it develops. And then also Microsoft did their own AI commerce feature called this Copilot Checkout. Again, this is a gigantic thing that is aimed at purchases being done through chat conversations. One big story that I just wanted to point out quickly is Apple will be partnering with Google for their Gemini AI. So you'll see Siri powered by Gemini, which is really what people just want. They just want to talk to an AI assistant and get things done across their phone, across their life. So we really getting close to it. I know we've been talking about that for the past few years on this channel and all across the Internet, but agents are materializing in a whole different way now, as opposed to two years ago where it was more like an idea and people tried to build automation flows that broke all the time. And now it's just becoming intuitive and it's starting to ship in products that are available to not just millions, but billions of people. It's going to be an interesting year. There's also OpenAI for healthcare that was now announced. This is different from OpenAI Health that we covered last week. OpenAI for Healthcare is their hospital and healthcare facing product that helps them make better decisions and gives them information, whereas ChatGPT Health is the consumer facing piece that helps us as consumers make better health decisions. So all of that is happening and some of the early data on that is actually very promising because the fact remains that standard for AI to match is not perfection, it's the human error rate, which is actually quite high, even amongst doctors. And using AI in the process has already proven to be of benefits to the entire process and we're only going to see more of that in healthcare. And then finally there's an Update from Google VO 3.1, their flagship video model. It now has things that people are requesting like native 4K upscaling and you can easily integrate ingredients into videos. And just a little pro tip, a lot of marketers are starting to use this product that we covered previously called Photos Flow, where you can bring in ingredients really easily and then mix your products with different environments and then turn them into product marketing videos. And all of that just became more powerful with this update. If you already use this, this is a welcome addition to your toolkit for everybody. Not using it is probably not a reason to start doing that. All right. And that's pretty much everything I have for this week's episode. I hope you found something that was interesting to you. I think this year is going to be acutely practical with all of these agentic apps actually being put into practice. I'll be here covering all of it. And with that being said, my name is Igor Pagani, and I hope you have a wonderful day.