Women: Unlocking AI - Ash Stearn on Building Agents for Everyone

Joanna Shilton (Host): 00:00:00

Foreign you're listening to WithAI FM.

Hello and welcome to Women WithAI, the podcast dedicated to amplifying the voices and perspectives of women and shining a spotlight on the fascinating, important and sometimes unexpected work being done with artificial intelligence. Today I'm chatting with an AI agent specialist who brings a marketing mindset, some seriously powerful tech.

So before we jump into our conversation, let me tell you a little bit more about her. Ash Stearn is revolutionizing AI automation for non technical professionals.

As founder of ashdern AI and a certified expert at Relevance AI, she's launched a woman led AI agent building community supporting women, LGBTQI individuals and neurodivergent people, creating a space where members can learn and build together. As a fractional chief AI officer, she develops advanced prompting frameworks to enhance agent decision making.

Featured in Forbes, Ash bridges traditional content creation with cutting edge AI implementation or advocating for women in technology when not designing AI agents or teaching others.

In her spare time, Ash likes to explore hiking trails in the Italian Alps, traveling and spending quality time with her partner and their chihuahua, Pedro. Ash Stearn, welcome to Women WithAI.

Ashleigh Stearn (Guest): 00:01:17

Hi. Thank you. Thanks so much for having me. Very excited.

Joanna Shilton (Host): 00:01:21

So, yeah, I know it's very exciting. Well, you're my first person that's based in Italy to speak to, so yeah, lovely.

Ashleigh Stearn (Guest): 00:01:27

And I guess I'm kind of unique in that I'm an Australian living in Italy. Huh?

Joanna Shilton (Host): 00:01:31

Exactly. So, yeah, two new boxes ticked off for me.

I love it because we connected and it's on your linked home profile because you say that you, you quote, I teach building AI agents without the jargon and making the complex simple.

And that is perfect for me and this podcast because that, you know, I'm learning and I just think it's fantastic that you know, people that are sort of doing that and helping people understand what AI is. So perhaps you could start with sharing a little bit about yourself and your journey into using AI.

Ashleigh Stearn (Guest): 00:02:03

Yeah, sure. So I guess, my gosh, my journey, it started, well, it really started about two years ago or almost three years ago.

I think that's when like ChatGPT really kind of hit the scene. It was something that people were talking about and you know, I was just listening. I wasn't, I didn't like take to it straight away.

I actually was quite scared of ChatGPT at the time. I was a content writer, so obviously I wrote blogs, I was doing Google Ad copy, so I was a copywriter as well.

And then I moved into social media management. So when you think about, my goodness, you know, ChatGPT, AI kind of, you know, taking over the world at the moment.

It was for me quite, quite a scary thing because Obviously what does ChatGPT do? Well, it writes, you know, it's a language model, so it's right stuff.

And so at the beginning it, I took a little bit to warm up to AI and to chat GPT.

But then slowly, as the agency I was working for, very open minded to using AI, I started to get a lot more curious about it, thinking, well, how can I actually use it to help with my content stuff that I'm doing at the moment. And that's where it all really kind of took off for me. The interest then really started kicking.

I mean I was always, I've always been a creative writer and obviously the thing with ChatGPT or any LLM model that you use, you prompt, you know, you have to write and the prompts you give it have to be quite, quite good. You know, they have to be, they can't just be a very vague one line or they have to be quite detailed. And obviously that was still creative for me.

It's writing.

And then I actually allowed ChatGPT to help me expand on content ideas, content angles and perspectives and how to enhance what I was writing for this agency essentially.

And then it really kind of took a, absolutely took off for me when we decided to go with a very, very, I will be honest, very aggressive repurposing content strategy was way too much. But that's what the CEO wanted at the time. And it ended up being about 120 pieces of content per week.

So and I'm talking about this was strung over across six or seven different platforms. So this was text based content. It was also graphics as well. So obviously chat gbt, woof is it really helps me to do all of that.

And then I was able to do it in about 10 to 15 hours per week. So it was, you know, it was, that's where I really noticed how AI was able to help me just with what I was doing.

I think soon then after that I started to become very curious about other AI agents, like AI agents for example. So that was something that was introduced, you know, you know, around 18 months ago, maybe two years ago.

Back then it wasn't a trend, it was just an I, it was more like a floating idea. And then that's kind of when I got very, very into it, as in I became very obsessed with learning about AI agents.

And obviously that's when I came across Relevance AI. And this was over 18 months ago. So when you think about it, our agents have only just become trending as an AI agent.

Joanna Shilton (Host): 00:05:31

That's why I want you to, for people that have never used or built one, or what is an AI, how would you describe it?

Ashleigh Stearn (Guest): 00:05:38

Right, so an AI agent, when you look at the kind of, the comparison here between like ChatGPT and an AI agent, so ChatGPT, you really, it really is focused on that back and forth interaction. You have to give it a prompt or you have to give it instructions for it to deliver to you.

An AI agent, on the other hand, it's able to make decisions, it's able to do things autonomously, it's able to go outside the realm of a user going in, telling it exactly what it needs to do, when it needs to do it, whereas an agent is able to do all of that by itself. I think the defining factor here, what really makes an agent, is how it can actually make decisions.

So it's able to make decisions and that's really one of the defining factors of what, what makes an AI agent is its decision making abilities, which, you know, ChatGPT doesn't make decisions. An AI agent will make decisions.

Joanna Shilton (Host): 00:06:37

Yeah, yeah, because they've, I, they've recently, it was in the news this week or the end of last week about how ChatGPT has suddenly got a new personality. Everyone's been saying it's sort of, it's been, it's been too nice, it's been too. What is it?

I don't know, it's just, it's kind of gone too far the other way. It's sort of being, oh, you're amazing, you're brilliant, you're wonderful. Thank you for asking that question.

I mean, I guess with an AI agent, do you train it to have a sort of personality or would it learn how to be the more you use it, or do you sort of program it at the beginning?

Ashleigh Stearn (Guest): 00:07:08

So it really depends on how you're going to use an agent.

So if you're looking at the intention of using an agent as a collaborative partner, then the exchange and the terms of output and how it's interacting with the user, you want to at least give it instructions on how it's supposed to respond, what, what kind of terminology it's going to use.

If it's an agent specifically used to write content, obviously then you need to give it specific guardrails in how it's supposed to generate that content for somebody.

You know, it's got to adhere to specific guidelines and guardrails that you're going to be able to Set foot it so you can train it to respond in a specific way or to write content in a specific way, for example, if that's the intention behind the AI agent you're creating. Whereas some people have been using agents more as a backend thing so they might have it in the background working.

So I guess the response and the output isn't quite. They don't need it to have the all use know this is how I want you to respond, you know, or anything like that.

So I guess it's a little bit different in terms of how you're using it.

But yes, you can program it to kind of speak in a specific way or deliver in a specific way otherwise other people don't and they just kind of, you know, that's just how chat the LLM is going to respond. That's how it will respond. So it really depends. It really depends on you. Yeah, yeah.

Joanna Shilton (Host): 00:08:41

And as you said, you were an early adopter of relevance AI. What, what do you think you saw maybe early on that others didn't?

Ashleigh Stearn (Guest): 00:08:49

So yeah, so I think, I guess one of the main things here is I've been able to see how, how the team have really started pushing the platform for it to be used by industry experts, not technical experts. Now it's a really.

And this is why I love relevance AI and why I'm so aligned to that platform is because they're trying to make it so easy for anyone, doesn't matter what your background to get in to start using it, start building agents immediately, like instantly.

And now when I say that, I also want to mention that one of their features that they didn't have before, that they have now, is that you're able to actually invent agents and the other stuff that goes with it, which in this case is tools. So every agent doesn't matter what you build, it's going to have automations.

Automations is just that, that stuff that's going to go on the background, that's an automation and every agent will have that. But relevant AI allows us to even create those just by giving it some brief instructions of. This is what I want the agent to do. Create the agent.

The agent will then be created. You can then create the tools or the automations that the agent will use to be able to complete tasks autonomously. Yeah.

Joanna Shilton (Host): 00:10:14

Wow.

So because I guess everything's changing so fast that yeah, you have to be able to invent on the go because there'll be things that you're coming across that haven't been thought of or.

Ashleigh Stearn (Guest): 00:10:25

Right, right.

Joanna Shilton (Host): 00:10:27

And you have, when we've spoken before, you said that agents shouldn't be fully autonomous. So what do you, what do you, what do you think the boundaries sort of lie. Responsible. Where does the responsibility lie? When building with AI?

Ashleigh Stearn (Guest): 00:10:39

Yeah. So this is, this is what I really mean by, for me, and this is a personal, a personal take on it and a personal opinion.

I believe AI agents, yes, there are levels, so when it's completing tasks, they're autonomous. It's going to complete tasks, it's going to do it autonomously and that's okay.

But I believe there should be a back and forth collaboration after it's finished the tasks, if those tasks need approval, anything like that needs to be handed off to a human.

So I believe that there should be a constant interaction there between the user and the agent so that the agent's just not going off and doing all sorts of stuff. And you don't realize because it's not telling you. This is when very specific guardrails come into play as well.

Things that need approval to be used first before the agent can do the task. So I, yeah, that for me is how I believe we should be interacting with agents.

Whereas some people might set up that agent and just let it do its thing and that's it.

Joanna Shilton (Host): 00:11:50

You wouldn't get a new member of staff and just let them go rogue, would you? And go, oh yeah, just do whatever you like, don't bother checking with anyone. You just do what you think's right. Right.

Ashleigh Stearn (Guest): 00:11:57

Could you imagine the chaos if we actually did that? You know, so it's the same kind of thing, it's the same, same kind of concept.

You know, you're not going to just hire an intern and expect the intern to just know everything from the get go and just let the intern do whatever they want. No, the. And it's the same thing with an agent. I personally would not let an agent have too much autonomy, definitely. In some things, absolutely.

But for some other things, probably not.

Joanna Shilton (Host): 00:12:22

No, because I'm sort of thinking, you know, an AI agent, oh, they're going to help me book a holiday or like my travel or something like that. And because you can give them your credit card details come you. And then I guess they would just be like, oh, I booked you on the train.

And then you'd be like, well no, actually I don't want to get up that early or I don't want to go through that place or Right, right, you need to check. But what kind of, what kind of industries and stuff are you have you Been building AI agents for where does it.

Ashleigh Stearn (Guest): 00:12:45

Yeah, so I think. So right now what I've been specifically building for is definitely centered around research. So and sales.

So a sales process obviously when leads come in or leads are generated, there's a whole bunch of research that needs to go into that. So researching the lead, lead enrichment, you know, generating pre sales core reports on that lead before you have discovery call with them.

So that's one very specific part of the sales process that I love because it's all research based and I love research. I can research anything for about eight hours if I wanted to. But like, so that's one thing where I've kind of honed in on other places.

So other use cases I've been using it for is definitely content production.

So content production in that it generates content, it's repurposing content especially from like transcriptions, podcasts, videos and then it will regenerate that content into so many different formats depending on the social platforms or even web content as well. So blogs, lead lead magnets, anything like that.

So that's another massive use case at the moment that I've been building for specifically because that's, that's my skills, you know, content. So makes sense. But then that I've been seeing other people do all sorts of stuff with, with agents.

So there's not really one, one area, do you know what I mean? There's not really one specific domain where they just kind of, oh you, you know, you kind of just see them everywhere.

And some of them are quite creative as well. The creative ideas I've seen people come up with is actually quite, quite cool. So.

Joanna Shilton (Host): 00:14:30

And then the agents must talk to each other, right? Is that.

Ashleigh Stearn (Guest): 00:14:34

Yes, yeah. So, so how does that work? So just to add another level of like coolness to it. So relevance AI, they, you have the ability to create teams.

So you're creating multi agent teams essentially.

So just like how you would in a department, a human department, you know, you' got certain layers, you've got manager, then you've got, you know, your juniors and everyone underneath that.

It's just like this with relevance AI, so generally how that would be set up is that you can have a manager agent, right, whose sole purpose is just to delegate tasks to the sub agents, right. So the agents underneath it.

And so it has the ability to communicate with the sub agents and understand when a specific task needs to go to what specific sub agent.

So let's say for example, we've got a content manager now underneath this content manager, we've Got a sub agent for social media, we have another sub agent who is responsible for planning and scheduling posts. Right. We might then have another sub agent that could be for, for web content. Right.

So based on whatever the task is being given to the manager, the manager will then delegate, you know what, where that task needs to, who that, sorry, who that task needs to go to based on whatever it is, whether it's social media scheduling or web content. So yeah, it's really cool.

Joanna Shilton (Host): 00:15:57

Wow. So then if you've got, if you are going to keep the human in the loop here, then I guess do you have that in a human structure as well?

And then because then. So you'd have different levels checking the work or is it so maybe just one. Because I guess some people, that's the thing, isn't it?

Is AI coming for your jobs and it's taking it and making you redundant. Because if you can have a whole team but you still want someone to check it. So do you need someone at each level or is that how people learn?

It's going to change the job landscape completely, isn't it?

Ashleigh Stearn (Guest): 00:16:27

Yeah, yeah.

So I think, I think in that specific case, when you're looking at content, well, you're never, even when you create stuff with ChatGPT, you're never going to take it and then post it and if you do, you're using it wrong because you never do that. Right. Same thing with the agents.

It's going to generate the content, you're going to take it away and you're going to make some edits or you can even suggest comments and edits about the content piece and you can get the agent to review those comments and edits from you and it can then, you know, make the changes.

Otherwise you're not, you're going to take the piece away, you're going to make the changes to it, you're going to approve the final piece, that's when after the approval, then the agent will schedule it and post it out to the social media platforms. So you're always going to have that human in the loop. And that's a perfect example of when something needs approval.

So you're going to have those guardrails in there that tell the agent, don't ever go posting content like to, you know, any platform without firstly approval from the user. So, so yeah, so that's a prime example of having a human in the loop. Same thing with emails though.

If you're doing a back and forth exchange, whether it's outreach, cold out email outreach, whatever it might be if there is something that needs to be escalated to a human, an agent is able to do that. So if somebody has responded in a negative way or they've specifically said they want to jump on a call immediately because it's an emergency. Right.

The agent's going to be able to analyze the sentiment of that and understand what's going on and then being able to pass it on to a human and that the human will be notified of when it needs to respond to that email. So it'll give it the email. Human knows exactly what's going on, and then the human will then respond accordingly.

So, yeah, there's lots of examples like that of human in the loop because.

Joanna Shilton (Host): 00:18:24

I guess you've gone from, well, from doing the custom AI builds to doing templates is because. Are you teaching people how to make their own templates or how to. Or. Yeah, yeah. How does that work? What made you change?

Ashleigh Stearn (Guest): 00:18:39

So, so I think the whole, the whole idea for me here was that I wanted to generate a very, a solid agent template that I can quite easily customize to clients.

So this is when, when I am teaching people who want to sell agents, I'm always saying, okay, you got to narrow in on one thing you want to do, one, one process that you want to build AI agents for, and obviously using already the skills and knowledge you already have. For me, it was content production. It just made sense. Right.

So I went in and instead of actually doing a project by project basis for various different kinds of use cases, different domains, that actually caused a lot of stress and anxiety for myself. I'm someone who likes to, to know exactly that the amount of time it's going to take for me to do something and all of this kind of business.

So I then thought, all right, well, the ideal situation would be that I have a couple of agents or a team of agents. It's all a solid template, and I can simply just sell that and then customize it to the, According to the SOPs or the brand guidelines of the client.

Just put all of that in there.

And then obviously you're testing and iterating and then launching, you know, so that's kind of what shifted me from building from scratch the systems to actually just getting a solid grounding first selling that and then customizing on top of it. So that's kind of the angle that I try to encourage people to take if they want to sell agents as a, as a, as a service, you know.

Joanna Shilton (Host): 00:20:15

So, yeah, because you came, you didn't come from the tech background, and it's Those transferable skills, I just say you stick with what you're strong at. So if it is contemporary production, then you can create templates and AI agents.

And is that what's the difference between an AI agent and like when people say they've done a custom GPT? Yeah, but again, is that because, yeah, the agent's thinking for itself and obviously the custom GTP doesn't.

Ashleigh Stearn (Guest): 00:20:41

Yeah, so. So a custom GPT. So a custom GPT is. It's not an agent, but because it has quite a lot of limitations to it.

So if you're looking at doing a custom GPT, usually what you, what you can do with it, you can upload knowledge into it.

So if you've got company information, brand guidelines, all of that kind of stuff, you can actually just upload a PDF file with all of that information, put it into your custom GBT so it has something to actually reference when it's creating content. For example, it knows that with the knowledge you've just put into it, it has to adhere to those brand guidelines.

Custom gbts also have access to external tools, so they can do web search. They can do, they might be able to post to your or schedule posts to a platform that you're using, such as LinkedIn for example. Right.

But it has its limitations in how much it can actually do and it's not actually something that you can, that it's autonomously going to go off and perform all these other different tasks for you and it's still very much relies on the prompt that you give it. That prompt needs to be really solid, it needs to be really clear and defined for it to kind of generate any kind of, I guess, quality output.

With an agent, on the other hand, you can give it one little query, right? And this query might involve various different things like I'm talking about.

Let's say, for example, you give it a query of, I want you to research my top competitors, right?

Gather all the research, identify a gap analysis, but also want to analyze the current posts that they've been doing on LinkedIn for me so that I can say then where my content I can generate kind of slips in the gaps there that I'm going to fill it based on my content.

So it's doing all this research, LinkedIn, it might be going on to Google search and searching up some other high, I guess, high performing results on Google based on a certain topic.

It's collecting all of that together, it's going to analyze it all for me and then it's going to give me back A report of, okay, this is all your top competitors, this is what they're doing, this is the content they're producing, this is where you and your brand can slip in and fill in the content gap there. So do you see what I mean?

It's done multiple things autonomously in order to answer your query, but it did involve it doing a Google search, scraping content of websites, LinkedIn scraping, and it's done all of that, which a custom GBT just won't be able to. It can't handle all of that kind of stuff at the same time. So it's, yeah, it's just next level.

Joanna Shilton (Host): 00:23:31

And you're teaching people from non traditional tech backgrounds. So I think that's. There's just so many opportunities out there.

When you think about the people probably aren't aware of, do you come across, I guess people come to you because they want to let. Or like, how do you find, how do people find that this exists?

Because especially, I don't know, I mean, this is women where they are, especially women. Like how. I don't know. Yeah, how can we get into it? I mean, is this where you've lost your community?

Ashleigh Stearn (Guest): 00:23:56

Yeah. So this is. So what I noticed a lot of what was going on is that there's a lot of, a lot of noise.

And to cut through the noise, it really does come down to, okay, if, if someone says that they're an example expert, how are they showing that they're an expert or are they just an enthusiast? Right. So you, you, you gotta.

I think what really happened for me was that I noticed, even on my own learning journey, I notice all of this noise going on. I focused on a handful, a handful of trustworthy content creators. Heather Murray was one of them, Isabella Bedo was another one.

And I really just stuck to them. And then from there I noticed in, in the space of AI agents was there's not really anything kind of accommodating for beginners. Right.

You know, I think it should be something that is accessible to everyone, not just people with a technical background. And a lot of different perspectives were happening with content creators who were very technical. Oh, here's an AI agent that I built in two hours.

Yeah, it might have built you two hours, but what skills do you already have to build the agent that you could do it in two hours because a beginner can't?

It's just not, it's just not reasonable to think that, you know, so that's the reason why I have set up my community, because the community, what it does is that we actually, we don't just go the theory. So I'm not just presenting. So I've got a mini course in there at the moment that does go over the basics.

But the real perk of my community is that we actually build together. So the whole intention is to not build these complex AI agent systems.

That's where you eventually want to go to, but you have to break it down and you have to break it down and you've got to cover the nuances between how this goes here, what that you know, why is there an error here? Why doesn't this work there?

Because it's super nuanced and if you don't know these little obvious things, but they're not so obvious, then it can be quite frustrating when things just don't go according to plan. So that's the reason for my community just to cut through all of that, just really bring it back to like the fundamentals just with relevant AI.

I don't use any other platform.

Joanna Shilton (Host): 00:26:16

Yeah, cool. And do you, is it, are you finding it's just individuals that come to you or is it actually like small. On small businesses and startups or.

Ashleigh Stearn (Guest): 00:26:24

Yeah, yeah. So I'm, I'm finding a, quite a. Quite a diverse range of people.

So people who want to build them for their own business, people who want, who are working within a company and want.

And the company is actually quite open to them using AI or building agents and they want to learn how they can do it to make them more efficient in their current role.

And they're also having businesses approach me who want agents built for them, you know, so it's, it's really, there's a lot of different people that are coming from, from everywhere that know about AI agents that either want to learn about it or they want to build it, you know, so it's. Or they want somebody else to build it for them. So it's quite a, quite a range at the moment.

Joanna Shilton (Host): 00:27:10

And so apart from Relevance AI, what's your favorite AI tool that you use?

Ashleigh Stearn (Guest): 00:27:16

Oh, there's. Okay if actually the list of AI tools that I use at the moment and it's not, it's not a lot I use. I can't say my favorite.

But anyway, if I was to name one, one, obviously it's Relevance AI because that's just my, my love. But obviously another one I've been trying out is Manus. So I don't know if you've heard.

Manus is another agent that was released just, just recently, maybe the last three or four weeks. That one was, is actually quite good and actually helped me develop a very advanced framework for agents.

So that one I've been using a lot, but I've just noticed that there's, it's now starting to not produce such great quality as it was when I first started. So I'm like, okay, maybe I'll move away from that. But the other look, the other top ones I use is actually Perplexity.

I don't use Google Search anymore. Perplexity is another common AI platform that is just on real time, like search results, all of that kind of stuff.

So if I've got anything that I just need information on, I'm just plugging that into Perplexity and be like, hey, can you just help me out with this? And so I don't go to Google Search anymore. I hardly use it. Yeah. So.

Joanna Shilton (Host): 00:28:40

And do you give like ethical and strategic advice to people as well? Is that all part of the. Yeah, yeah, yeah.

Ashleigh Stearn (Guest): 00:28:49

So that's kind of, that's something that, that just naturally comes into it, I think especially when it comes into AI agents as well. You do need to set up specific guardrails when you're using it.

So for example, in an agent that I've just built, I have purposely drilled it into the agent that it needs to warn a user when. Specific tools, there's two specific tools that it has that it can use and they cost a lot more credits to use them because they're quite advanced.

But it has to inform the user that this is about to cost you a lot of money essentially or it's about to cost you a lot of credits in order to use them. So there are concerns in that area where there is the transparency between an agent and a user, which I think is really, really important.

But then obviously the ethical use as well. In some things that I've built, a lot of the stuff is involved in like scraping web content. So I've from websites.

So I've set it up in a way that it's actually determines whether that content, if you're allowed to.

Because some websites, they've got specific documents on there which I mean you can't just find it, you've got to really search for it on their website that actually says no, I don't. This, you're not allowed to scrape my content.

And so what happens with that is that as soon as that is picked up in my tool, we don't use it, I don't use it. I will go buy based off other websites that allow it. But if something flags it and says, no, you can't scrape this content.

Then I just discard it and I don't use it.

So that's just one real ethical way of how I'm aligning what I'm doing with my content agents to make sure that it's not going into places where people don't want it to go.

Joanna Shilton (Host): 00:30:40

Because people say it's not that AI, it's the people using it. But do you think that if it is going to start making its own decisions, will it just ignore that and go, well, I'm going to use it anyway?

Because you know, they don't.

They sort of try and program some of them to like, oh, AI, would you turn yourself off if you were going to take over the world and like kill humanity? And they go, oh yeah, yeah, yeah. And then they discover that they wouldn't.

Ashleigh Stearn (Guest): 00:31:01

They just carry on going, yeah, that's an interesting, an interesting thing to think about. Oh goodness. It's really. Yeah. I think that's a deep question.

Joanna Shilton (Host): 00:31:15

Probably a whole other. I was going to say, how much.

Ashleigh Stearn (Guest): 00:31:17

Time do we have?

Joanna Shilton (Host): 00:31:20

Yeah, let's not talk about the scary stuff. What excites you most about AI agents? Where do you think they're heading next?

Not down the kind of like doom and gloom taking over the world, but yeah. What, what's the fun, the fun stuff.

Ashleigh Stearn (Guest): 00:31:32

Exciting, you know, it's really exciting.

I, I think, I think AI when you're looking, even with agents, right, and we've seen that now with relevance AI everything is going to be language based. Right. Now you can do, you can code apps through language, through prompting, through even now talking. It's called vibe coding.

You just send an audio, reads the audio and it can create you an app. You know, it's the same thing now with agents and relevance AI.

This is what we're seeing now is that you can just give it some instructions on what to invent and the agent you need and off it goes. It's invent, you're doing an agent. Right. So it. Or creating an agent. Sorry. So everything I think is going to be very language based.

So things like, like coding and now you don't need to know coding in order to code. Right. It's the same thing with AI agents. Beforehand you needed to know coding and now how.

Now what's developed is low and no code platforms like relevance AI that allows you to create agents without coding. Everything is going to be language based. Everything. I think absolutely.

And I think when in the terms of the agent space, I think it's going to become.

It's going to get easier to build agents, but I think the next step is going to be enhancing how they make decisions and their reasoning capabilities.

I think that is, I think when we really start seeing how we can enhance an agent in a way that it's ethically aligned and you know, it's really abiding by the guardrails, but it's also at the same time being able to take the vaguest query from, from a user, but it's able to process it in a way that it's able to ask clarifying questions to help it to make decisions based on, you know, whatever it might be. So really I think we're getting into now just really advanced decision making abilities.

I think that's, that's kind of, that's where I feel like without getting too into it because that's where I'm leading at the moment. That's because that's why it's fresh in my mind. That's what I'm doing at the moment, trying to develop something like that. So we'll see.

Joanna Shilton (Host): 00:33:43

But yeah, what would be one thing then for people that are listening today to take away from our conversation? Especially you know, people might be thinking that AI is not for them or they don't know how to use it or where to start. What would you advise them?

Ashleigh Stearn (Guest): 00:33:57

Okay, so if you're a beginner that's just starting out with AI in general, or even if it's AI agents, you've got to narrow down on who you're following and what you're consuming. Don't get bombarded with all these different tools. I only made it as far as I did.

Coming from writing blogs to now developing pretty complex, you know, agents and all this kind of stuff. And I didn't, I wasn't able to do that if I didn't just focus on one platform. In this case, it's relevance.

AI give my all into it, into learning how it worked and then cutting out all the noise because the tools that are coming out, the advances in the upgrades some of the tool existing tools are having, it's really easy to get distracted. If you're gonna start, start with something like chat gbt first you've got to learn prompt engineering.

Prompt engineering is, is honestly if it's a skill you need, it's that because everything is prompt based. Everything like we just, I mentioned before, everything is, is programmed by language and it is prompting.

That's how these LLMs do what you want it to do. It's through prompting. Learn that Skill, it's key also with agents as well. In relevance AI, it's all prompt engineering. So develop that skill.

It's really important if you want to start with AI agents, it's easier than you think, a lot easier than you think. Now, sometimes I can be posting stuff on LinkedIn that's quite complex, but it's only complex because I love it and I want to share it.

But then there's other moments where it really is easy. It's not difficult. And I think now with what relevance, AI, how they're pivoting right now is to make it so easy for industry experts.

You can just invent agents, you know, it's a massive step up from where it started 18, like two years ago. And, and, but that's thanks to that, I've learned as much as I have. But now it's so easy, I'm like, are you kidding?

Like, why couldn't you have this before?

Joanna Shilton (Host): 00:36:00

So thank you, Ash. It's been incredible speaking to you. I've learned a lot.

So if people want to either join your community or just reach out or find out what they're doing, what's the best way to find you or connect with you?

Ashleigh Stearn (Guest): 00:36:12

Yes, Look, I would say LinkedIn is obviously the best place for it. In there, I've got a featured post that's pinned to the top of my. My profile and it includes a link to my community.

So you can join through the link in the caption in that featured post at the top. Otherwise there is, yeah, LinkedIn, there's a book, a call with me, whatever it might be, but that's how you can reach out.

LinkedIn happens to be my platform, so.

Joanna Shilton (Host): 00:36:39

That'S where you can go all the links in the show notes. That's fantastic, Ash. Well, yeah, it's been an absolute pleasure to speak with you. So, Ashton, thank you for coming on. Women WithAI.

Ashleigh Stearn (Guest): 00:36:50

Awesome. Thanks, Jo. Thanks for having me. It.