S2. Ep29 - Set Up Claude Once, Use It Everywhere: Projects, Skills & Cowork Explained

Speaker A: Welcome to the AI Automations for Business podcast. We're your hosts, Katie and Noel, and we'll be discussing how you can use AI for your business, along with the latest news, updates and automations to help you stay ahead of the curve, allowing you to grow and scale your business more efficiently. Please be sure to subscribe and I hope you enjoy this episode. Hello and welcome back to another episode. Hi. Hello, I'm Katie and as always, I have Noel with me today. Hi Noel, how are you doing?

Speaker B: Well, um, I'm doing absolutely fabulous again this week. Just very, very busy with everything that's going on. How are you doing?

Speaker A: Yeah, all good, thank you.

Speaker B: Good.

Speaker A: Um, so we always start the podcast off, um, with any news, any updates in the AI and Automations world. So now what are those updates for this week? What do people need to know?

Speaker B: Well, I've had to write them down this week because it's been quite a few. I had to keep a track of them as I was seeing them. But I guess the first big one is, I know we've all been, you know, lots of people have been complaining about Fable 5's usage and it's not been very good. And at the end of last week, uh, uh, Anthropic decided, hey, here's Opus 5. It's cheaper than Fable and it's more powerful than Fable and it won't go through your usage anywhere near as quick as Fable does. So, yeah, massive announcement. I've been using it lots, uh, in the past few days, uh, since it released. I can definitely say it's very, very good. It's a lot quicker than Fable 5, I was finding, especially like coding projects. It was taking an age to get through things. Whereas Opus 5 is accurate and fast. So just what I wanted. Um, so, yeah, really, really awesome release. Obviously, uh, that's better than any other AI model currently on the market, uh, which is pretty awesome. Well, apart from Mythos, but we can't see that. So, no, we'll forget that exists for now.

Speaker A: Okay. It's not there.

Speaker B: Yeah, exactly, you might as well. Uh, there are other. So there's two kind of related things. So this is kind of like with OpenAI and anthropic, maybe getting things slightly wrong. So I'll start with OpenAI. Uh, they were testing their new Frontier model. Uh, they had it in like a closed off, uh, environment and they were, they tasked it to, you know, show off its hacking skills. And this agent was like, I know what would be pretty cool, you know, to show off my abilities as I'm going to break out of this sandbox and I'm going to go and get things off of the Internet to help me do a job. And essentially hacked the website of Hugging Face, which is, it's basically an open source, uh, website for open source, uh, AI models. Uh, and it went off and it grabbed loads of stuff from there, then bought it back into its sandbox to help it finish the task. So yeah, the engineers OpenAI obviously hit the kill button and we're like, well, this shouldn't have broken out. So it's obviously very good at hacking. But yeah, yeah, very interesting. But I've seen that when I tested Clyde when I first created the open source version, you know, it has like its own, uh, file repository. So you don't give access to all of your files on your computer. You know, it should only access this certain working folder. And yeah, silly me, I forgot to give it write access. And yeah, Clyde managed to work around that and then save it to my desktop. And I was like, ah, that shouldn't be allowed because I hadn't really put the guardrails in at that point, you know, and it was like, right, they need to go in now to close it down for you. They're very clever, these models, getting around things, which is, yeah, interesting. Um, the other one on the similar note from Anthropic is. I don't think most people know about this one is, you know, you know, when you create like, um, I know, like an artifact or maybe you created like a, an app within. Flawed. Within normal chat, there's like a little share option.

Speaker A: Yeah.

Speaker B: Uh, so you then get uh, a URL and I could give you the URL, Katie, and you can look at it and use it, whatever it is. Those links have been indexed on Google.

Speaker A: Oh,

Speaker B: yeah. So they no longer get indexed on Google, but they have been and they have been for, uh, some time, I think. So there's like people creating, you know, little apps that maybe has API keys in there. There's been things for crypto, wallet keys and addresses that have been, you know, found in these, these links, all sorts of personal information, you know, you might like, like for us, you know, we could plan a holiday, you know, uh, and it could have, you know, how much money we're going to take, what currency, where we're going to be, where and when, you know, and I could share that with you, but anyone could see it, which is.

Speaker A: That's literally like my worst nightmare.

Speaker B: It is, yeah. I mean, I've never used the share feature, thankfully, but yeah, I was like, oh, I didn't quite read it, that it would do exactly that. So, yeah, I think they've stopped it now. Thank you.

Speaker A: Okay. I was going to say is there a way that we can like prevent this? But you think that, uh, it's been solved?

Speaker B: Yeah, they've stopped indexing those links, uh, on Google, uh, which is the big problem, I guess. But yeah, I just.

Speaker A: So really, it's just be careful actually. What go. What data still goes into that?

Speaker B: Yes. Uh, the safest option would be to download whatever the artifact is or whatever that chat is and then send it. Yeah, yeah, don't use that.

Speaker A: Share it directly from.

Speaker B: Yeah, yeah.

Speaker A: Okay.

Speaker B: I thought it was quite an interesting problem, that one. But anyway, an interesting, uh, problem.

Speaker A: Understatement.

Speaker B: Yeah. Uh, there is a new image model coming soon. So this got me quite excited because before we went all silly about Nano Bananas and chatgpt, uh, images and creating model figures of people, um, the leading model was from, uh, Black Forest Labs called Flux. And that was excellent at uh, photorealistic images. And it was really one of the first ones that was actually very good. Uh, but they're going to release in a few weeks, I believe they're going to release Flux 3 and I've seen the previews and it looks incredible. So it's not just going to be images, it's also video and video with audio as well. So yeah, I'm really excited to see, um, that one coming out. I think it's going to be pretty, um, awesome and probably kick things like Nano Banana to the curb a little bit maybe. Yeah, it all looks very accurate. Like the, the I get with most things these companies will cherry pick the, the samples and whatever, but I've not seen any crazy sort of videos and stuff with things going wrong. So. Yeah, be interesting to try that one out.

Speaker A: Mhm.

Speaker B: And um, my last bit of news is about Substack.

Speaker A: So I was going to ask you about this because we had this conversation a couple of days ago and I was like, surely we need to mention this on the podcast.

Speaker B: We do. Yeah. I guess if no one's ever heard of Substack, um, it's a social media platform. Uh, you can have like your mailing list in there. Uh, you can earn, uh, revenue from paid subscribers. There's like a, uh, social media feed where you can post notes and things like that. So it's very, you know, it's very writer focused. Is, yeah. Is their main sales pitch.

Speaker A: Yeah. So like you can't post like, like you wouldn't create videos to post on substack. It's all about writing. Yes, you can go live, uh, and you can host like calls with your subscribers. So you can have free subscribers, paid subscribers, annual subscribers. Um, but yeah, it's mostly um, a text based platform and it will show you different people based on the interests you have, um, which I really like.

Speaker B: They also have a really cool algorithm as well which allows you to grow a lot quicker than you would do on say LinkedIn as an example. Um, but the reason we're talking about it is because they're the first social media platform that I know of that have now added in AI text detection on all of their written content. So this is something that was released last week and basically in the uh, the Notes feed you can hit on the little three buttons, uh, on a post and then you can then go right scan for AI text and it will give you a percentage score. So if that note is less than 100 words it doesn't bother, it's only 100 words more. Um, every written bit of content on the platform will get through the AI detection. Um, but what people are finding is now this detection isn't particularly infallible. You know, it does make a lot of mistakes. I've posted stuff which just comes straight out of Claude and it's gone, it's 100% human. And it's like, well that's not true because you know, I've got the receipts for it in my CLAUDE account. Um, or it could, you know, score things which are AI, which aren't AI. It's something which has been handwritten. So there's, there's a lot of upheaval about it, uh, on their platform. And yeah, I don't know, I hope other platforms don't go down this sort of route because it, it kind of penalizes those that actually write the content manually. Uh, those talented writers that are on the platform, like some of their stuff has been, you know, put out as AI generated and they're like, well it isn't. And there's no review process either. So the score is the score. You've either got to rewrite it and hope for the best or you just go right with it.

Speaker A: I just think as well like you, you've just got to say to yourself, well actually, do I enjoy reading this content? Yes. And then therefore you've got to think, well, how important is it if I know if it's original or if it's AI? And then I just think, make up your own mind from it.

Speaker B: Yeah, uh, but as ChatGPT would say, here's the kicker. Um, when they announced the, uh, update, there was one little bullet point which I spotted in there and I was like, that's interesting, because what they're going to do is at some point they're going to put in an option so readers in their preferences can tick a box and say, I do not want to see any AI generated content.

Speaker A: Oh, uh, gosh.

Speaker B: Which is fine. I mean, that's a personal choice. But for those writers that are being penalized, their content's got to be hidden. So, yeah, yeah, it just all seems a bit, you know, self defeating because, you know, people are earning revenue on the platform, which is paying substack in royalties. Um, and yet this content is going to be, well, technically, I guess shadow banned really is a way of putting it. You know, it's just gonna silently disappear and never get read.

Speaker A: Well, not, not everyone will be ticking, you know, no AI. But the thing is, is what is heartbreaking is the, the people who are writing everything themselves, but it's being flagged as AI. And then there's people who will then say, I don't want to read anything from AI. I feel like that's the problem. I think they need to substack, need to make sure that whatever program or software they're using is actually working 100%, otherwise, what's the point of having it?

Speaker B: Yeah, well, I had to chat with Claude about this and basically there's no, there's no way to get it 100% accurate with text. It's impossible. You know, they are fighting and losing battle because these models are getting better at writing humanized content. So it's like, even today these algorithms aren't good enough to be accurate. And it's quite interesting on the website of the provider, they have a bit about education. So I'm sure there's people that listen to this podcast, have got children. So I think it's kind of interesting that their homework could be graded through this system because they target education, universities and whatever to check the work. But if you get a false positive or false negative or whatever, then they could be hauled up and said, well, you've written it with AI, when they actually haven't. Um, so on their website they say they're 100% accurate. It should be believed. And then in the FAQ on the same page, it goes, yeah, but this shouldn't be the sole evidence for a disciplinary. It's like, well, what is it? It's either accurate or it isn't accurate.

Speaker A: Exactly. I feel I, I get where, what they're trying to achieve. But until actually the technology is available to make sure it is 100%, then I think, yeah, don't bother. I mean the, what's happening on substack right now, I'm not even going to believe anything. I'm not going to be ticking. No AI. Um, you know, I'll, I'll leave it all open because yeah, if we're not 100 sure, then I could then be missing out on reading blogs or uh, you know, other things from people that I enjoy following and you know, reading their uh, things. So.

Speaker B: Yeah, exactly.

Speaker A: Yeah. Okay. Interesting.

Speaker B: Yeah. Fun week. Yeah.

Speaker A: Okay. So this week we've actually had a question from long term listener and friend of podcast Fee who is all the way down in Perth, Australia and she asks how can you set up uh, Claude skills for a membership that includes all the different areas of running a membership so like social media, emails, content, community, customer service so that you're not having to re. Enter, uh, the same information over and over again?

Speaker B: Great question. Yeah, yeah, I think it's one that you know, every business owner will come up against when using AI. Um, so the easy, the easy question, the easy answer, uh, I guess for this is using projects. So within Claude and ChatGPT there are projects you can collate and gather all that information. Um, you can then have chats within that project. So they're all hyper focused on exactly those areas that you wish to not constantly repeat. I get that. It's very annoying, isn't it? With AI, you've got to re, go over the same point you did last week. Um, so yeah, the first step is setting a project. Uh, within these projects you can set instructions as well. So uh, you can see that as kind of like the system prompt for an AI agent. So if there's anything you want it to focus on or don't want it to focus on or do, that's the place to put in. I would spend quite a bit of time in those instructions because that's just going to save you lots of time in the future. You know, you're not having to rehash things every time. Um, if you do find that you, you know, you are coming across things that you don't like, you can always go back and edit the instructions. Uh-huh. So any future conversations, uh, uh, are a bit more accurate. Um, but yeah, projects definitely the way to start. So you, you could have, you know, I guess for me like I would see like a project, uh, in My business is kind of like, you know, it's a project for one client and then another project for another CL of thing. But you could also see this as like certain verticals within your business. So it could be your marketing projects, you know, your substack project, your community project, you know, you could have multiple setup which are hyper focused, um, and you can use uh, at any time. It does mean you do have to switch in and out of projects to do stuff which isn't, that's just a user, user experience thing. But yeah, if that's what you need, that's the way to go. Um, and I guess to make it even better uh, is to use Claude skills. Uh, I can't remember the terminology in ChatGPT off the top of my head right now. I don't have a ChatGPT account anymore so I can't go and check. But um, you've got skills within claude and basically they're just kind of like your um, your process instructions, you know. And your project can use those skills to get you the things that you need right first time every time and you can describe them, you know, you can tap away and say, well look, this is the skill I need to create, uh, this is what I need to do. Fine, it will go off and create it and you can then use it in the project. Uh, but one thing that they released this week actually, um, is a thing called Recorder Skill. So for me with creating skills I always find it quite difficult because you're having to type everything out, uh, or dictate it somewhere and you might forget things and ah, things like that. Whereas with Recorder Skill it will record your screen and you know, you could be in a document, you could be in a spreadsheet and you could go, right, well this is where I get this from, this is where I want this to go. And it takes screenshots as it goes and transcribes it and creates steps. And then once you've, you know, recorded what you need it to do, it then creates a skill based on that, which I found incredibly useful. You can use it in the browser as well. So if you want to click around websites to do stuff, it will watch you on that website and then create a skill based on that. Yeah, so really, really powerful uh, to use skills within projects. So that's kind of how I would tackle it. There's lots of other things you can add in. You can add in documents for context, uh, you can add in links which you um, maybe to like sales pages and things like that or Copy from sales pages, all of that can go into projects. And then you would never have to reteach AI ever again. Uh, how to do that particular function. So that's, that's how I would tackle it, I think.

Speaker A: Yeah, Okay. I think, I think that's a good, ah, solution. And I love that you can, you know, take Claude through your screen and documents. I think that's really helpful.

Speaker B: Hugely. Yeah. Ah, yeah. I didn't know I needed it until I tried it out. I was like that. That's actually really awesome because we're only human, aren't we? We're always going to forget something when we're typing something out. So. Yeah.

Speaker A: And is that available on free accounts or just paid accounts? Great.

Speaker B: Uh, question. I think it's available on all accounts for Claude. It's not on chat. G. I think you've got to have at least the minimum. There's the go plan, isn't it? It's the $8, uh, plan. So I think you need that to get projects. Um, so, yeah, but also actually within projects in claude, you can also use Coworker and have scheduled tasks as well. So if you've ever used Cowork, where it can access your local files and, you know, do complex tasks, um, you know, autonomously, that can also be done within a project. So you can have all of that context and the automation to go with it, uh, which is really awesome.

Speaker A: Okay, good. Well, we hope that answers your question. And, uh, if you've got a question that you would like us to answer on the podcast, please send it over to us via email. Helloakeautomations. Uh, AI. And yeah, we'll try to answer your questions on the podcast, but that's it for, uh, this week. We really hope you've enjoyed listening to this week's podcast episode and we will catch you next time for another episode very soon.