S2. Ep33 - Which AI Agent Is Right for You? A Beginner to Expert Breakdown

Katie: Welcome to the AI Automations for Business Podcasts. We're your hosts, Katie and Null, 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. Hi, hello and welcome back to another episode. Hi, I'm Katie. And as always, I'm I've got Noa here with me. How are you doing this week, Noel?

Noel: Oh, I'm doing great this week. How are you doing? Yeah, good.

Katie: I'm almost ready to put a sweater back on, I think.

Noel: Yeah, we've had summer. Yeah,

Katie: after how many heat waves? Was it five?

Noel: Four, four. Four, five, yeah. Yeah, I've had enough of that.

Katie: Yeah, I'd enough of that. Time to put a jumper back on.

Noel: Certainly.

Katie: Yeah, we're British. we're happy with that.

Noel: Yeah, exactly. Comfortable. Yes.

Katie: So this part of the podcast, we always let you know of any updates that are to do with AI and automations that obviously are relevant to business. So no. Any updates that we need to be aware of this week?

Noel: Yes. Well, there's one big one, which is making quite a few. I've got a bit of noise in some circles, and that is the stealth release of a new model called Ox Alpha, and this was released a couple of days ago on a platform called OpenRouter, where they have loads and loads, they've got 600 AI models you can pick from, and things like that. So this model appeared on there, and it's basically around like the Fable 5, maybe GPT, 5.6, Sol, kind of level intelligence. So it's very, very clever. And what's also catching people out as well is it's also multimodal. So you can input text, image and video. The and video part is actually quite different to what most others do. Because with Claude, you can't put video in. That's no idea. It can't, which is really frustrating sometimes. Just like, just watch this video and tell me about it.

Katie: It is really annoying because sometimes I'm watching a TikTok and they're talking about products and I'm like, but like I don't know, like I didn't catch the pronunciation or I don't know how to spell it.

Noel: And what I would love

Katie: to be able to do is just stick it in, Claude and go, what are the products listed or mentioned in this TikTok video?

Noel: And it comes

Katie: out and tells me and where I can buy them. Now, probably by this time next year, that will be the norm.

Noel: Yeah, I'll build that this afternoon with Ox Alpha for you. Yeah, I mean,

Katie: it seems almost ridiculous now that you can't do things like that when we can do so much with AI.

Noel: Yes, definitely, yeah. Yeah, it's getting there. We're getting close to the, you know, but the thing with this model, though, is there's, there's so much speculation because they've given out a ton of usage. So I think they've allowed something like 200 trillion tokens a day to be used on this model completely for free for like the next week or so. And I'm just having a look on Open Router's website at the minute. And in the last four days or so, they've only used up about 20 trillion tokens, which is still a hell of a lot of usage all for free. So yeah, there's a lot of people trying to work out who it is. what it could be. I think a lot of people are thinking it's going to be a GLM model from ZAI on the Chinese labs. But, yeah, looking at it.

Katie: You do think it is, because I've got suspicions.

Noel: Oh, so I've got a couple of big suspicions. You go first. Don't you dare pick one of mine?

Katie: So I actually think, and I think I've got it with this one. I think it's Miss Gaila in the library with the candlestick.

Noel: Yeah, could be. Tell me I'm wrong.

Katie: Tell me I'm wrong.

Noel: Well, if you're right, wow. Should I go with my options of what I think it might be? I mean,

Katie: go on then. If you must.

Noel: So I've got a few contenders. So I've got three. My one that kind of makes sense would be Google. So when you're looking at like giving away trillions of tokens of usage for free, you kind of got a look at who's got the most servers, who's got all these data centres around the world that can have that capacity.

Katie: And also you can afford to give away that many things. Exactly.

Noel: Yeah.

Katie: Yeah. That is actually quite a lot of money.

Noel: It would be, yeah. I love a lot of money. Well, it depends how much they put the model up price-wise. When they release it, we'll find out. But, yeah, I think, like, Google, they haven't released a model that matches anything from, you know, Opus 5, GPT 5.6. They've had a few little releases recently with a couple of little flash models, which kind of hit the low end from OpenAI. Anthropics. So potentially it could be one of theirs. But looking at the odds, not many people think it is. But I think that's probably a sound bet. My other bet, which is a little bit more outlandish, is this could be the update to Haiku. So this could completely disrupt all of the API usage for businesses and things like. They'll all switch overnight if it is to Haiku or would be five, because this hasn't been updated since last October. So, yeah, it's well overdue. But yeah, if it could squeeze into that level intelligence for a cheaper price, that would disrupt a lot of things. But I think that also is quite unlikely, unfortunately, that it is anthropic. And mainly because, you know, some really clever people have gone through and looked at like this from a technical perspective, and they reckon it's one of the Chinese labs. So I did a bit of research, and no one's talking about it yet. So if I've got this right, then, you know, I would like this to be noted and etched in stone somewhere.

Katie: Okay, you've heard it here first. Exactly.

Noel: So I think it's from a Chinese lab called Shao Mi, and they're behind the Mimo models. So you may have never heard of them, but they're kind of cheapish, kind of models. They're pretty good, actually. You know, they're quite intelligent. And they haven't released one since April 2026 was their last update, I think. So, but the kicker, as Chuchy P.T. would put it, is they've got a habit of releasing models without telling them who it's from. So, yeah, their last two that they, you know, released secretly were called Hunter. Alpha and Healer Alpha. Now, what are the odds that they've also released something called Ox Alpha? I mean, just piecing it together, you know, yeah.

Katie: Okay.

Noel: Going to go all Colombo on it, you know, I don't know.

Katie: Yeah, okay.

Noel: But yeah, that's, I think that's probably who it is. Okay. But yeah, that could, yeah, I think that could be quite shocking, I think, actually, if it is them. and they've got this video input as well. I think that could get quite a few people excited about that, but which I'll wait and see. I think we're, well, I say we're about a week away from finding out who it is, I think. So, yeah, we'll have to watch this space.

Katie: Watch this space.

Noel: Yeah, but if I've got that right, for.

Katie: Bragging rights, right there. Exactly, yeah.

Noel: Cool, yeah.

Katie: Okay, so any other other? other updates that we need to be aware of for this week?

Noel: There is one other, but we're going to talk about that in our main part. So that is a release of Grokbot. We're going to go in a bit more detail into that in a moment.

Katie: In this week's episode, we are talking all about the different types of AI agents, when to use them, and also the skill level that is required for using these AI agents, which I think it's really important because we can talk about, you know, all these different AI agents and then, you know, you could then want to go and use it and then realize actually it's above your level of competency. So it kind of then makes it really frustrating, I think, when someone's really hyping up something and then go, actually, I don't have that level or, or it could be the other way around. Actually, it's really basic. And it doesn't do what I want it to do. Yeah,

Noel: you kind of find out when you give it a go, don't you, sometimes? Yeah.

Katie: So hopefully in this episode, we're going to just kind of point you in the right direction. So you're not wasting a lot of time as a business owner. Because obviously time as a business owner is very precious. Absolutely.

Noel: Yeah. So I guess the first section we'll go through are agent builders. And like we mentioned a minute ago, there's a new agent builders come out recently within the last week or so. And that's from XAI. So this is GROCbot. And it has its uses, but it has a lot of drawbacks as well. So very briefly, what Grockbot allows you to do is basically correct. create little teams of agents to go off and do work for you. You can connect your apps. You can have skills. It's the usual kind of thing that you would expect an agent to do. But the big drawback I have for it is $200 a month. It's a basic. So it's like, well, as a business owner, am I going to waste my $200 to find out It doesn't do what he want it to do. Probably not. Yeah. It's quite cool how it works. There are some, you know, interesting features where you can have your agents go off and chat to each other and go off and complete things. But, yeah, I just, I don't know. It gives me the wrong sort of vibes. But it's that kind of, you know, it does allow you to, like, create agents from plain language. So, you know, it does have that, ease of use in those also. So you don't need to be, exactly, we don't need to be super technical to say, I need an agent to create me some marketing material. We'll go off and do that. So, so that's great. But yeah, there are other things in there as well, you also need to describe what the agent does so other agents know what it does. So that's not something that's within, you know, that's created automatically. That's something you have to do. So if you get that wrong, then your experience of Grokbot will be, you know, far worse than somebody that's, you know, really good at describing those sort of agents. So, yeah, not sure about that. But there are other builders as well. So obviously, I've built one called Clyde, which is fairly similar to Grokbot. Obviously, with Clyde, you can connect in any AI model you wish. So we're not tied in because like with GropBot, you're tied into whichever GROC AI model. It doesn't, you can't even pick which one you want to use for each agent. You just, you just do, you know, you get whatever you're given. Yeah, yeah, it doesn't say which one it is or anything like that. So it could be even their lowest model.

Katie: Yeah.

Noel: Yeah, so we don't know. But yeah, we're Clyde, you can pick any. You can use Anthropic, Open AI, Google, you know, any model whatsoever. You can even use Ox Alpha if you wanted to right now as an example. But yeah, like with Clyde, you know, again, it's really easy to, you know, get on with. You can just describe what you need to do in plain language and it will go off and build it. So, you know, there is a market where there are other tools out there that also do very similar stuff. So that's kind of one end of the agent builder market where you can do. describe and it will go off and then do that, build those agents for you. But then there is the other end of the same builder market, which is where apps like make.com, Zapier and N8NCit. So for this one, you would need to be a little bit more technical, mainly because you've got to build these agents manually. So, you know, you've got to create like the scenarios, plug them together. You've With that one, yes, they do, again, very similar stuff. They can connect into all of the applications that you use in your business, but it's going to be a bit more of a technical leap to get you to that point. And yeah, like system prompts, that's all stuff you've got to do yourself. So that's kind of a skill in itself is getting those prompts right and things like that. So, yeah, those are the two ends of the agent builders. but yeah but I would say like some of the like the Grop bots and I think there's like relevance AI like most of those will just keep you in one particular area that will use one model they don't really give you much flexibility so when assessing those do check that out first things like that.

Katie: Good tip.

Noel: Yeah so these are I would say like the builders are kind of like an easy way to get in I would say if you want to get into the intermediate level. I wouldn't say they're super beginner, but they're in that middle ground. They're not super technically either.

Katie: Yeah.

Noel: And I guess the next one after that, and I think this one actually has the most risk. When I was thinking about this earlier, and it's like agent systems, so you've got things like Claude Co-work, and you've got ChatGBTBT work. Now, again, they're single agent systems, they'll go off and do stuff. You can connect all your tools in. It's already built into your subscriptions, which is great. You're already paying for it, so why not use it? I guess is one argument. But the big risk I see with these is, you know, these companies, they're going to be looking at how many people are using this. How much of their resource are they dedicating to this particular service? And is it worth their time? So as we saw with Open AI, like they got pretty ruthless. Like they chopped Sora, the video generator. Like they got rid of that. They said, that's a waste of time. We're moving away. So who's to say in six months time that track GPT work just disappears because nobody really uses it? So, yeah, I think that's probably quite a big risk in terms of, you know, for business use cases. You know, you don't want to put loads of effort into building something. And then it either completely changes or goes away overnight, which these sort of things tend to do. But again, they're really easy to get into. What you need to do is connect your apps, which is really straightforward in the, in the applications. You know, you can just ask for it to go off and do stuff, and it will do it for you. So, you know, very, very, I would say these are the beginner, super beginner friendly. Easy to get started. Yeah.

Katie: Great. Hmm.

Noel: The third tier to this is kind of like the self-host. Now, this is like super technical. I'm not going to go into great detail in how you would do it, but it is possible to do it. So you could either host on your own computers, laptops. You could host it on your like a virtual server somewhere. And you could run things like OpenClaw. There's other agents like Hermes. that you can run remotely. So these are all, they're not going to have like a subscription cost per se, but you'll pay for your server usage. That's kind of where it gets you. But yeah, these are really open. You can pick whichever model you want. You can tweak them. Most of them are open source. So if you want to jump into the code and make it better, you can do. That's up to you. But yeah, I would say these are like the super expert level, you know, of building agents. And some of them aren't that particularly great. I mean, we've been through OpenClaw earlier in the year. And yeah,

Katie: and hence why it came about that you then created Clyde.

Noel: Yes, exactly. Yeah. I was like, this isn't safe. I mean, the biggest drawback for some of these, especially if you install them locally on your laptop, is these apps would then have access to all of your files. So if you don't want that to happen, then you know, don't go down that route.

Katie: Yeah. Yeah,

Noel: especially if they've got sensitive client information or personal information, your agent will go off and find it. These agents are incredibly clever at finding stuff these days.

Katie: And you've got to think really, like, all of the things that we do on our laptops, like we just think, oh, we'll just do, you know, we'll just do this. And it could be like buying car insurance, buying travel insurance.

Noel: Is they going to

Katie: know all of that information?

Noel: Yes.

Katie: You know, it could be that you, you know, people are sending you emails about a trip that isn't work-related. It could be like, you know, a French trip or, you know, like a hen or stag do, which I think is a bachelor or a bachelette party. other parts of the world. So then it knows information about that as well. It knows your location and, yeah, all those things. And, yeah, I just think we don't realize actually how much information we have on our computers.

Noel: Exactly, yeah. Because I found this when I created Clyde now. There's also an open version of Clyde. And when I was building that and having it on my laptop, like when I was early testing it, it was like, oh, I can't find out how to do this, but I'll run this terminal command and then save this file to your desktop. And I was like, oh, what? No, no, no, no. It's before I put all the guardrails in, so it can't do it anymore. But it just goes to show these agents, if they have access to terminals on PCs and laptops can go off and do whatever they want, which is, yeah, not ideal. Yeah, especially for businesses. Definitely not idea.

Katie: Yeah.

Noel: So, so yeah, those are the kind of like the three main areas of different agents to build. So I guess like if you're like a beginner, you're not really sure how to get into agents and things like then. I would say look at the agent systems first. You know, have a play around with co-work or chat your BT work and things like that and kind of get used to how to connect the tools together, you know, connecting your accounts into an agent, and then, you know, seeing how an agent would go through and use each of those. So most of them will say, well, look, I'm now going off to Salesforce, I'm getting this information. I've now got the information. I'm now doing this with it. And it shows you that breakdown of what's going on, which is really, really helpful, especially as a beginner to understand how these sort of things work. And also, it's also a good way of actually spotting things when it does something wrong as well. So you might see it do things. Hey, hold on. You could have done it quicker this way. And you can teach it that as well, which is kind of cool.

Katie: Yeah.

Noel: So yeah. So beginners systems, then I would then look at builders, like in your intermediate level. So, yeah, I would definitely look at like anything like Clyde. If you've got the money to have a go on crop bot, then you can do. The other tools are out there as well. I think Relevance AI is another one. There's tons and tons of agent builders that you can pick from. But again, they're a little bit more technical. You will have to be a little bit more hands-on maybe on some than others. So, yeah. But again, the same sort of processes apply where you can connect everything in, which is really cool. But obviously, the other builders from Make Zapier and NAA, that's where you're going to have to learn. I guess we make's a little bit different with the latest release of Maya, which is, yeah, so you could ask their makes agent to go off and create you an agent. So yeah, it will take you through it all step by step so you can see it being built and you can then learn how to then build them for yourself in the future using Maya. Which is pretty cool. It is, yeah, it's really, really awesome. So yeah, I were, yeah, that's kind of the intermediate, like I said, did. the experts, probably not for most people that listen to this particular podcast in the self-hosting. So, yeah, obviously, all of these have all different prices as well, which is also worth considering. Also, when looking at these, you know, check out which AI models you can use. Some businesses out there, you may be restricted. So, you know, your business might be like, we're only going to use Gemini. We're not going to use any other models. So you're like, right, well, I need to go off. and find an agent system that I could use that also has that option to use those models. So you stay compliant and things like that. So, yeah, it's always worth keeping an eye on. And I guess the other big thing as well, when looking at these sort of agent systems is kind of, you know, is there, what's the usage limit? But how quick am I going to, you know, hit limits and basically can't do anything thereafter? So that's also a very important consideration because you don't want to do a ton of stuff on Monday and then you can't touch it for another week. That would be really frustrating. Yeah. I think anything on the anthropic side, you will run into those limits pretty quickly. So like co-work, I would watch that. But that's why I've built Clyde in a way where it was like a bring-your-own API key and your own AI models. So you control that usage. cost over time so yeah so yeah those those are the the main points and yeah like i wouldn't be afraid of trying out agents you know agents have been out for cool well haven't long have been out now for about 18 months i think we've been having agents for so they're only getting better as all and they'll be easier to use in the future no doubt as well

Katie: so if if someone hasn't um tried using an agent for say six months. Would you encourage them to use it, try using one again because they've changed so much in those six months?

Noel: Yes, 100%. So, I mean, when agents first came a thing, it was NAM were the first ones to really bring out an agent. And they had all kinds of flaws and that kind of turned a lot of people off. So, yeah, things have moved on exponentially in all areas since then so yeah definitely give it a go

Katie: okay great okay and if you've got any questions that you want no to answer you can either email ask hello at makeautomations.a i or come and pop them in our free LinkedIn group which is called AI Automations for Business anyone is welcome to come and join that group and yeah you can put a question and we'll try and answer it for you maybe on the podcast or if you would rather we can just keep it to the LinkedIn group but thank you so much for listening to this week we really hope you find it helpful and we will catch you next time for another episode very soon