S2. Ep15 - How to Build an AI Agent (Without Writing a Single Line of Code)
Think you need to be a developer to build an AI agent? Think again.
In this episode, Katie and Noel walk through exactly how to build a working AI agent from scratch using Make.com — no code, no JSON, no command line. They cover choosing the right no-code platform, planning your agent on paper first, structuring system prompts with clear roles and goals, and connecting business tools via MCP servers so you don't have to add 200 individual modules.
They also dig into Anthropic's latest Opus 4.7 leak, the Make.com Grid update, managed agents for Claude Code, and why Katie might be introducing a buzzer for future episodes to keep Noel from going too technical. Plus the golden rule: if you need a conversation, build an agent — if you need a repeatable output, build an automation.
Previous Episodes Mentioned:
S1. Ep 14 - AI Agents vs Automations: What Does Your Business Really Need?
S1. Ep42 - Why Only 6% of Businesses Trust AI Agents (And How to Fix That)
How to find us:
Join our membership over on Skool, where we support you on your AI and automation journey. We share exclusive content in the membership that shows you the automations we talks about in action how to build them. Find out more about the AI Business Club here.
We have a free LinkedIn group (AI Automations For Business), the group is open to all.
New for 2026, you can also find us on Substack, click here to subscribe and get all the latest news and updates from us.
If you would like dedicated help with your automations or would like us to build them for you then you can find our agency at makeautomations.ai
-
Katie (00:26)
Hello, welcome back to another episode. Hi, I'm Katie and as always I have Noel here with me. Hi Noel, how are you doing this week?
Noel (00:37)
I'm doing great, thank you, how are you doing?
Katie (00:39)
Yeah, all good, thank you. Before we get into this week's episode, any updates from the AI world or automation world that we should be aware of? I know some weeks there are like 100 updates and then everyone goes quiet for a while. What is it this week?
Noel (01:02)
So I guess probably one of the biggest ones is there was yet another leak from Anthropic. I think they love leaking stuff on their website.
Katie (01:11)
Do you think they do it on purpose?
Noel (01:17)
I think they must do. It's almost every week.
Katie (01:19)
It's like some sort of PR stunt, so people actually then talk about it.
Noel (01:25)
I think so, yeah. It's to build that sort of interest, isn't it? Because the leak this week was Opus 4.7. So obviously, again, that would be the most powerful thing they've ever done. Well, I suppose it's not what they've ever done. They gatekeep that, didn't they? Because I think we talked about that last week, where it was so powerful. They're like, no, we're not going to give anyone access to it. We're only going to give it to big enterprise companies because it would benefit everybody if they used it.
Katie (01:30)
Yeah.
Noel (01:53)
But yeah, I guess that's kind of the big news. There has been an update actually from our friends at Make.com. They have a tool called the Grid where you can visualise all of your automations in one place. And they released an update this week where you can now drag and drop and resize all of how the Grid looks. Before it was just based on their algorithm. So if you didn't like the way it looked, tough. That was it.
Katie (02:23)
Yeah, you didn't like the way it was designed, that was it. Not very visually appealing.
Noel (02:34)
Yeah, it was more like a messy web that it would put together. But now you can move them around. You can change the size of the folders and things. So yeah, it's been a bit long overdue. They haven't released an update actually since last June for the Grid. So yeah, nice to see there was something there.
Katie (03:06)
Have they implemented my traffic light idea on the Grid?
Noel (03:12)
No. I've not seen it.
Katie (03:14)
Rude. I gave you a cracking idea for that.
Noel (03:26)
We did, yeah. The developers loved that idea, didn't they?
Katie (03:29)
Yeah, until probably it came to actually coding it and they were like, oh god.
Noel (03:37)
So hold on, what was your idea? Was it to do...
Katie (03:41)
So you know on the Grid where you can see all of your automations that you have created and when there's a problem with one of them, it comes up, I think it was a red colour. But I said, well, if you've got lots of automations and lots are in red, you almost want to know which one's costing you the most and therefore go and fix that one first. I said it would be really good to have a traffic light system. So it's like, this needs attention, but actually it's broken but it's not costing you any money. And then a really deep red would be like, this is broken and it's also costing you a ton of money.
Noel (04:33)
Got you. So yes and no, actually. I'm looking at the Grid right now. Kind of. So they still have the big red markers for things that are actual errors. So your automation has failed. But where things are, you may have deleted your credentials or something, they now show up as amber. So we've got two traffic light colours there. We haven't got anything else.
Katie (04:48)
Maybe they did listen to me.
Noel (05:06)
It might have done, yeah. But it's not quite 100% for your use case. But yeah, there is starting to be some additional colours, which is good to see. Makes it easier.
Katie (05:16)
Well, come on developers, implement my cracking idea, please. I'll be asking them in October if they've done it. What was wrong with my amazing traffic light? Okay, we went off on a bit of a tangent on that one. Let's rein it back in. So that's really the only update, yet another in quotation marks leak from Anthropic.
Noel (05:31)
Yeah, they've also released managed agents as well, which is a bit nerdy. It's in Claude Code CLI and API. So unless you're really into your software development, you'll probably never see it or use it. But essentially it allows you to run agents for long periods of time, but you pay for them to host it. So as well as paying for your tokens like you normally do, you also pay them eight cents an hour for basically renting their server to run it. So it's not expensive, but if you've got a hundred of them running 24 hours a day, that could get pricey after a while. But that's developer focused more than your standard business owner.
Katie (06:39)
Yeah, that could start to add up. Okay. Well, let's move swiftly on from that then. Our podcast is not aimed at developers. It is aimed at service-based business owners, mainly solopreneurs or people with very small teams. But if you are a developer and you're listening, then please go and find Noel on LinkedIn or Substack or wherever and feel free to message him because he will talk to you about it. He will get nerdy with you. Thankfully this is not the time and the place.
Noel (07:24)
Yeah, love it. Definitely.
Katie (07:38)
Well, this week we want to talk about AI agents, which we have talked about a couple of times on previous episodes, but quite some time ago now. But in particular, we want to talk about how to build an AI agent if you're not techie. So we're talking no code. We're talking about how anyone who doesn't have a lot of developer experience, tech experience, almost the average business owner who can find their way around things, but aren't super techie. And we're going to be talking about that today.
Noel (08:26)
Yeah, absolutely. There's lots you can do. Like I say, we don't have to be super nerdy like me to do it.
Katie (08:37)
Okay, so Noel, where do we even start with building an AI agent if we are not techie? Because I think for a lot of people, this is actually a really overwhelming thing.
Noel (08:44)
Yes, so it kind of comes down to picking the right no-code platform. I think that's probably where you should start. So if you're in that camp of I'm not particularly techie at all, then I would suggest, and I think what we're going to talk about most on this episode, is Make.com because they've made it super simple and they've made it really easy for you to create and understand what's going on.
So there are other applications. You've got your Zapier and n8n. But with n8n, it's very much like you've got to create connections using JSON strings. That one sentence alone is going to probably put off a lot of people.
Katie (09:41)
Oh, that's overwhelmed me, just that one sentence.
Noel (09:48)
Exactly. Whereas with Make, it's just like, this is the data, I just want it to go there, please. And you just click it and it's done. You don't need to worry about it. Make sort out all the JSON stuff in the background. You'd never see it, which is really awesome.
Katie (09:58)
That sounds perfect. How do we get involved in this?
Noel (10:12)
So I don't think you can get it on their free account for AI agents. I know they used to do a trial, but I think they've stopped it now. So I think you do need to be on the first paid plan to get access to agents. But one thing to note with Make is they've got two different ways of building agents at the moment. So I think one's going to be phased out. They're kind of both called the same thing.
When you start Make.com, you create a scenario, you can search for agents and you'll see those pop up. And there's one called "new AI agent new." So that's the one that I would stick with. But yeah, it's really straightforward. They've made it really easy how to connect things together. And that's kind of where I would start. You don't need to really think about how am I going to chat to this in the future. We can come back to that. But getting the agent there on the screen is step number one.
Katie (11:13)
Yeah, okay, great. Step number one, pick scenario, agent new.
Noel (11:21)
Yeah, pick your platform. It's a nice little setup. They've actually made it quite easy. So if you do select the wrong one, the wrong one's a circular icon, whereas the new one is a hexagonal shape. So you do know if you've got the old one.
And I guess to start off with, when you're thinking about creating agents, you need to think about the key role and purpose of this agent. So what I would say is, let's say you want to connect things to Google Sheets, Google Drive, Gmail, and you've got loads of different business tools and you want to bring them all together in one super agent. I would caution against that. That's probably not the right idea.
What we can do within Make with no code whatsoever is we can link these agents together. So you could have an agent at the start which goes, what are they talking about? They're asking about emails. I've got to go up the email route and then talk to the email agent. And then you get your response back. So I would look at splitting them out because if you start overwhelming agents with loads of different things it could connect to, lots of things it could do, they tend not to work as well.
So step number one is to figure out what you want it to do and then is it worth splitting them out into multiple agents.
Katie (13:06)
Okay, so ideally, how many tasks can you give your AI agent without it being overwhelmed and then deciding that it's not going to work?
Noel (13:19)
So I actually ran into this problem with Clyde this morning, actually. Clyde's a platform I've built as an AI agent, but I ran into an issue where it was like, you've got over 200 tools attached to it, so it won't work. A tool would be like a single integration into a platform. So that could be like create a row in a Google Sheet would be one tool, delete a row would be another tool. So if you connect in Google Drive, there's 50 or so tools in there straight away.
So it depends on what you're connecting in. If you connect in a big CRM, Google, anything like that, they're going to have lots and lots of tools to pick from. So definitely worth splitting those out. But the way it normally works is like an orchestration agent. If you're doing that, they're basically just your filter. They read your message in, they then look at the other agents that are attached to it and then go, right, go down that route, get the response and then send that back to you as the user. But if you're doing something simple, then you don't need that step.
Katie (14:40)
Okay, good to know.
Noel (14:42)
I guess now you've kind of figured out and written down what you're going to try to achieve. The next step...
Katie (14:52)
Yeah, this is something actually that you say quite a lot. Before you actually start building your automation or even your agent, just grab a piece of paper and actually write down what you want the agent or the automation to do. So actually it's easy then to follow along and to build it out.
Noel (15:01)
Yeah, exactly. You're kind of in that sort of process. When I'm doing it, I usually tend to spot things that I've not thought about, or I think, actually, no, that could be a potential problem. So you're not spending an hour building something that you realise won't work and then you've got to unpick it and rebuild it. So hopefully with that little planning phase, you're going to save a bit of time.
Katie (15:42)
Yeah, and as well, a lot of people will maybe be like, well, do I need an AI agent or do I need an automation? It's something that we know a lot of people come across as a problem all the time. They start building out an AI agent, but actually what they needed was an automation.
Noel (15:52)
Very true, yes. Because we did an episode, didn't we, I think it was a couple of weeks ago where we talked about companies that were using agents and they were like, well, we just don't trust them. And we went through in a lot of detail in that one. But essentially, if you need a repeatable output, so you know every time you're going to put the same thing in and you want the same thing out, that's an automation. If you need to have a conversation at any point, then that's where your agent comes in. But I think most enterprise businesses are just going down the agent route when really they don't need to.
Katie (16:43)
But yeah, we also did an episode on that. It was AI agent versus automation. And that was series one, episode 14. And the episode that you mentioned, we'll make sure that we link it in the show notes.
Noel (16:49)
Yep, we'll definitely do that. Okay, so I guess once you've done your planning bit, the next most important bit is to think about your AI provider. With Make.com, they've got every AI platform you could probably want to connect into it. And with their agents, they've done something a little different.
With Make, you can use their own AI, which is based on some OpenAI models, and you get charged extra credits based on the usage and things like that. So it's probably not ideal for long-term use, because you'll get through all of your credits really quickly.
But what you can also do is link directly into things like OpenAI, Gemini, Anthropic, all the big ones. I think even Grok and xAI are in there and all that sort of stuff. So you need to pick an AI provider that you are comfortable with using. That could be from a security perspective if you're using it within your business. And yeah, it gives you the output and results that you're going to want.
This isn't one of the settings where you set it and then you can't change it. You can go back and test it and then go, actually, I really don't like how Gemini is answering this. I'll switch it for an Anthropic model and then it works better.
There's also actually a really good cost-effective method as well. They kind of link into the big AI providers directly, but you can use an app within Make called OpenRouter and that connects to 630 AI models, I think from the last time I counted. They've got all sorts. So they've got your DeepSeek, your Kimi K 2.5, every single model, all different price ranges, all different capabilities. So you could connect in agents from there as well depending on budget and what kind of thing you're after.
So that's really important to test and get right, especially for long-term running agents. And I guess once you've selected that, the next big thing to look at is your prompt engineering and your system prompt. It's a little different to what you would do if you're just chatting with ChatGPT or Claude or whatever. That's very conversational. Whereas when we start connecting agents, we kind of need it in a more structured way.
The first thing I'd put in there is a role. So tell it what it's there to do. And you can then set up goals in there or particular workflows and things. You can add all of that in. So every time the agent fires up it can go, right, okay, this is who I am, this is what I need to do. I'll go and do it.
Another top tip would be once you've done your planning stage, have a chat with ChatGPT or Claude and then get them to draft out the system prompt to help you get started. Because obviously not everybody's got either time or wants to sit down and type out a thousand character prompt every time. They can get quite detailed quite quickly. So I would definitely offload that initially and then you can go from there.
Katie (20:58)
Yeah, so with this prompt, you've got to be really clear and concise about exactly what you want with it. It's basically just giving it very clear instructions and directions, not allowing any room for error. So it's not kind of like, I want you to do this or that. It's, I want you to do this.
Noel (21:13)
Yes, you've kind of got to think of it as almost a single purpose thing. So you wouldn't have it connected to your CRM and then go, what's the weather today. You've got to structure it so it won't go off on tangents. If you did ask it questions that it doesn't have access to, it shouldn't then come back with a response, unless of course you want it to. It's up to you to personalise it how you wish, but I always keep it constrained so I get what I want every single time.
Katie (22:07)
Okay, good.
Noel (22:13)
Yeah, that's really important to get that part right. And again, with Make.com, they make that super simple within their app. So you can go back in and edit it at any point once you've created it. It's not a set once and that's it, which is awesome.
And I guess now you've got your system prompt, you've got your provider. The next thing is to maybe think about what you want this agent to have access to. You kind of look at your business products or your software that you use on a daily basis. So you might just be all on Google. You might have Google Drive, Gmail, Docs, Sheets, all that sort of stuff. What you can do is figure out what you want to give access to, and then you can assign them to the agent.
In agent speak, they're called tools. So it would be like, here's a tool. This tool is then going to allow you to do this particular function. So it could be create a new Google Doc or send an email or look in this particular folder of your emails. They would be an individual tool that it would have access to. It can get quite complicated if you've got lots and lots of different things.
But the way Make show it on their screen and in their builder is, you have your agent and then underneath there's loads of little bubbles for each individual tool. So you can clearly see what it has access to, which is quite awesome. And for each tool, you can also provide a description in there as well and say, you only use this tool to do this particular thing. This is what it's there to do.
I would always make sure that that is also referenced in your system prompt as well. Just so it knows, it goes, oh, I need to create a document. And it goes through and goes, this tool is all about creating documents, I'll use that one. So it won't go off and use different tools. That's kind of important to work out.
But with tools, we can connect individual modules from Make.com, which could get huge if you're doing things especially on Google or even HubSpot, like CRM. They've got 150, 200 different API calls that you could do within the Make platform, which is way too much.
So what I would also look at is do the software platforms that you use have something called an MCP server? So MCP is Model Context Protocol. And essentially what it's doing is, for HubSpot for example, they take their 200 API endpoints and put it into one single tool. So when the AI goes, I need to use HubSpot, it uses that MCP and then the agent sees all of those connections and goes, I'll go and use this one, does whatever it needs to do and then feeds it back to you.
So instead of connecting them all one by one and creating all of those descriptions, you just connect that one MCP server. They've not been around for that long, probably about eight months, I think. A lot of big companies already have them within their enterprise suites and whatnot for their tools.
You could also actually create your own MCP servers on Make as well. So if you have particular workflows, you could create your own MCP server, add all of those scenario workflows in, and then just call that from your agent. You don't have to rely on somebody else to do it. You could create your own if it is needed as well.
Katie (26:11)
And is this something that you show people how to do inside the AI Business Club?
Noel (26:39)
Yes, I do show agents. I haven't created one for the new MCP toolkit because that's not long come out. So that will be getting added in there soon.
Katie (26:58)
Okay, and if anyone's wondering what the AI Business Club is, it's our membership over on the Skool platform. If you search on Skool for AI Business Club, it will pop up and everyone is very welcome to join. Whether you're a beginner, intermediate, or more advanced, we've got different people in there from all over the world and also different levels as well.
Noel (27:26)
So once you've got all your tools connected, I guess the next step is to connect it into something that you're going to use. It's kind of like a chat function. So you need to use some sort of chat app.
There's Telegram. You could even use things like WhatsApp to connect your agent into your mobile phone. Or if you feel very brave, you could vibe code your own little business app and connect those in. That is also possible.
Katie (27:52)
I don't think that's what we'll be talking about on this episode. Let's keep it nice and simple. You see what I'm having to deal with, people. I'm having to rein him in every time. I almost need a buzzer when Noel starts going really techy. To bring him back down to just talking about it for beginners or those who are intermediate.
Noel (28:09)
Yeah.
Katie (28:36)
Yeah, I think because Noel's got so much information and knowledge because he's been in the AI world for so many years, I think he forgets that a lot of people don't have his knowledge. So yeah, maybe Noel, next week I might find a buzzer and I'm just going to start buzzing you.
Noel (28:56)
Oh dear. After I do all of that, no.
Katie (28:59)
No, that wouldn't be edited out. That would actually be in the podcast. So people would be like, what is he on about? Oh, thank goodness. Katie's pressed the buzzer. We're about to find out what that actually means.
Noel (29:16)
Oh dear. Okay, so I'll try and rein it back in. The easiest way I would say to connect your agent to your mobile phone would be Telegram. On Telegram, you can create your own chatbot and from there you can then connect it.
With Make.com, you would then connect the Telegram module into your agent. So you would have the message coming in on one side of the agent and on the output, you would then have the message that goes back to you as the user. Really straightforward to do. It's just three modules, including the agent, to link it all together. That's a simple way to get it on your phone.
But before you get to that point, you could test it within the Make.com platform. They do have on their page, you can chat with the agent directly. On the module itself, there's a little plus button at the bottom. If you hover over that, you'll see a pop out that says chat. That then opens up a preview chat window within the scenario builder. So you could chat away with it and then see what tools it's using. Is it doing the right thing? Is it giving you what you need before you then go off and connect it to an external application? So that is built in within the builder, which makes it really straightforward and easy to do.
Because you will find you need to test and refine. You might get lucky and get it 100% right first time. But for me, I'm always like, that could be a little bit better. I like tweaking. But with the new builder, it's so much easier.
Katie (31:07)
Okay, good to know that you've got to tweak as well. You can tweak away.
Noel (31:36)
100%.
Katie (31:39)
Okay, good to know. And that's how you build out an AI agent, the non-techie way.
Noel (31:44)
Absolutely. It's all drag and drop. There's no scary code in there. I think probably the most complicated part would be the system prompt and everything else is pretty straightforward. If you've never used Make or any sort of no-code platform before and you want to connect things like your applications into Make, when you connect them as a tool, they always pop up with a little dialogue that says you need to create a connection. That basically means either log in or supply an API key for that particular platform.
If you click on it, there's usually a hyperlink in there as well which says, take me to the documentation. So if you're really stuck, you can click that little hyperlink and it will tell you for that dedicated app where to go and how to do it. And all of those documents are fully updated as well. So if you ever get stuck, you can always click on that and it will help you out.
Katie (32:46)
Yeah, or people can come and get help from you.
Noel (32:50)
They could too, yes. I'm more than happy to help.
Katie (32:53)
Yeah, through AI Business Club, which is our monthly membership, or they can book a one-to-one call with you as well.
Noel (33:01)
Absolutely, always here to help.
Katie (33:04)
Yeah. So I would love to know if you are now thinking about giving AI agents a go after listening to Noel's explanation that you can create an AI agent without being techie. I would love to know if you're going to give it a go and even what you're going to build and then let us know how you get on. We'd love to hear that, wouldn't we Noel?
Noel (33:17)
Definitely would, yeah. It's always exciting to hear what people do with AI. Things I've never thought of.
Katie (33:32)
Yeah, I find it fascinating. When people tell us the sort of things they're building out, I just think that is so cool. Or I would never have thought of that. Or I didn't even know that would be possible. It's kind of like when people tell you the reasons why they started their business, it's always so fascinating. And I find what people build with AI and the automations as well as the agents, the bots, I just find it really interesting what people are doing with it.
So do let us know. Even if you haven't been in our membership or you haven't had any help from us, but you've been building things out with AI or automations, then let us know. And we'll give you a shout out on the podcast as well. You can let us know either at hello@makeautomations.ai or you can come over to our free LinkedIn group, which is AI Automations for Business. You're very welcome to join that. Anyone's welcome to join and just have conversations with fellow people who are building out with AI and also automations.
Amazing. Well, thank you so much, Noel, for taking us through the non-techie way to build an AI agent. We do hope you've enjoyed this episode, and we'll catch you next time for another one very soon. Take care.
