Table of Contents
Takeaways
- Understand AI agents as digital employees:
- Autonomous, proactive, and capable of complex tasks.
- Unlike standard AI (e.g., ChatGPT), agents can independently execute and adapt without constant oversight.
- Distinguishing between tools:
- Custom GPT: Tailored, context-rich AI, but needs continual input.
- Traditional Chatbots: Limited, rule-based interactions.
- AI Agents: High autonomy, able to proactively make decisions and handle complexity.
- When to use AI agents vs simpler automations:
- Make/Zapier: Simple, predictable tasks.
- Relevance AI: Suitable for flexible workflows needing context-aware decisions.
- n8n: More technical, customisable, and suited for complex workflows with enhanced security.
- Practical use cases in marketing:
- Complex content repurposing, maintaining context and tone.
- Automating in-depth lead research, reporting, and prep.
- Key considerations before diving in:
- Steep learning curve—requires skills in automations, prompt engineering, and workflow planning.
- Strong recommendation to start simple and focus deeply on mastering one platform first.
- Reliability & testing:
- Thorough testing and refinement are critical to ensure agent reliability and accuracy.
- Always include a human-review step for sensitive or public-facing outputs.
Overall, start small, focus your learning, and test rigorously to ensure reliable AI agent performance.
Transcript
hello lovely humans it’s so lovely to see you here today I’m laughing at
0:07Helen’s message in the chat which says Hey from Nottingham just got off the phone with Sky TV and now need this to
0:13calm down um but is just fabulous so happy to provide that service for you uh
0:19there Helen and thank you everyone already who has popped a message in the chat it’s so lovely to see you here
0:25today uh if you haven’t already uh do fire up that chat feature uh so everyone can know that you’re here and do say
0:32hello uh let us know where you’re watching in from I can see bunches and bunches of messages coming in already
0:38quite a few of them uh coming into to host and panelists only so there’s some some instructions on your screen right
0:43now that if you’re uh one of those people has that chat feature switch to hosts and panelists uh then uh please uh
0:52do uh change that to to everyone by T ticking the little toggle in your chat feature so everyone uh can see your mage
1:00messages uh just like Daria has in Poland uh Maggie has in London uh Sam
1:06has in Ipswich uh we’ve got folks in York Norwich Newcastle Sheffield I’ve seen people in Croatia Dublin Edinburgh
1:13uh it goes on so thank you all so so much uh for joining us today today we
1:19have the fabulous Ash Stern who’s who’s um a new friend who was actually
1:24introduced by a previous TMM speaker Heather Murray uh who delivered I think
1:30the most popular talk of 2025 Heather hit um and I’ve had one conversation
1:35with Ash and in that one conversation she just absolutely blew my socks off with just how kind and present and human
1:43uh Ash is to speak about the topic of AI which uh is often steeped in mystery and
1:50um sort of a lot of Bluster so thrilled to bits to have Ash uh with us today to speak all about AI agents uh which is
1:57something which um I I think is shrouded in another level of mystery and so uh to
2:05to be able to explore this today uh is fabulous for the marketing Meetup before we get started I just want to say a big
2:11big thank you to our sponsors so this week’s featured sponsor is a company called frontify who we’ve been working
2:17with for maybe the last 18 months or so now frontify are fabulous because what they do is bring all of your digital
2:24assets into a single place and make it super coherent for your team to be able to roll out your assets and that’s
2:31important when it comes to Brand because I think we all know the feeling of when Dave from accounts is using the wrong
2:37logo and you got to go and have that awkward conversation uh frontify help you with that and right now they’ve got
2:43a fabulous resource uh which enable you to uh scale out your assets from a
2:49single place into Global campaigns something which I know that a lot of folks watching in will feel the headache
2:54of um so do take the time to check out frontify lovely people and a useful Serv
3:00service a big thank you also to story block Cambridge marting College plannable Red Gate score rap we’ll speak
3:06about each of these sponsors in Greater depth uh over the weeks as we go through the rest of our brand season um Now’s
3:13the Time to also say a big thank you to everyone who’s been saying hello so hello to Julie Daniel Tom Daniel Hannah
3:20Janice Victoria Chelsea kathan Gail Lauren Daria uh so many more is what a
3:26pleasure it is uh with all that said uh that’s my ruction done so Ash um thank
3:32you for showing up as you do uh thank you for being here and uh it’s over to
3:37you yeah no thanks Joe and thanks everyone for for turning up today so
3:43what I’m going to do is just share my screen
3:49um okay and just we’re going to get straight into it [Music]
3:55um cana’s hard for this so like you’re you’re you’re doing
4:01it’s um yeah it’s the first time I’ve actually used it and just realized I’m
4:06not on the right slide there we go yeah it’s the first time I’ve used it to present anything so let’s hope it all
4:13goes all right I think fabulous over to enjoy enjoy Ash thank you thanks all
4:20right so my um presentation today and really what my aim and goal is to do um
4:27for you guys is to really break down what agents or AI agents are so this is
4:33really going to be an introduction into what they are um very key and specific
4:38things you need to know um in order to either build your own agents or whether
4:43you’re going to hire someone and train someone in house um so it’s really just going to bring it down to what are the
4:51the fundamentals of of a an AI agent okay right so the just an over you
4:59I’m just going to go over who I Who I Am uh who am I my background a little bit about me um what are agents uh do you
5:07really need AI agents um so we’re going to be looking at some existing tools as
5:13well um such as make and zapia which you will need just a a little bit of an
5:18introduction into these tools just to understand I guess the bigger picture of AI agents as well we will then look at
5:26um other agent building platforms which is relevance Ai and N they are currently
5:32one two of the biggest platforms out there and most popular um then we’ll look at when to use what so when is it
5:41going from a simple automation to then jumping into something like an AI agent
5:47and then of course uh I have to admit my favorite is Reality Checking and just
5:52setting some expectations there for you guys um and what it really does take to
5:58build agents um and everything that kind of goes into that just based off my own
6:03personal experience um and my own Journey so who am I um yeah you’re
6:10probably wondering who am I um so I come from a very very non-technical
6:15background actually I have been a creative my entire life so writing was something I’ve been doing since I was a
6:22kid um I was very artistic runs in my dad’s side of the family definitely not
6:28my mom’s um so it’s writing for me has always been a I guess a bit of a a gift
6:35um it wasn’t until I really started traveling quite a bit and then became a a bit of a an expert uh living in Italy
6:43uh and traveling back and forth quite often between Italy and Australia so as
6:49um transitioning away from uh I was actually previously a disability support
6:55worker to now working online it made sense to really start up Skilling in um
7:00in terms of copywriting digital marketing um all of those things social media management um and so in two years
7:08I did that and I managed to land my first um online job which was uh writing
7:15Google ad copy uh for new clients so I really went into this copywriting job
7:23the agency was a wonderful wonderful agency to be working for they were also
7:28very open-minded to us using AI um slowly my role did develop and
7:34transition more into then social media management and um writing blogs uh as
7:39well as writing ad copy for both Facebook and Google so it really um yeah
7:47there it it was a lot uh so I was doing multiple things um at once I think the biggest change for me when I actually
7:54started using a lot of AI was um especially chat GPT was when our social
8:00media management uh sorry social media strategy uh changed to a very very aggressive repurposing content um
8:08strategy which I wasn’t completely uh I did not agree with uh because I think it
8:13would came across too spammy but it gave me the opportunity then to leverage something like chat GPT now originally I
8:20had not use chat GPT I was afraid of it um as I think a lot of people were when
8:26it was released um but it really gave me the opport Unity to um leverage and
8:32streamline my my workflow so just to give you a bit of a picture I was producing about 120 pieces of content
8:39spread over six platforms both video uh content and written content so obviously
8:47cat gbt was a wonderful gift um and then my it it just all took off from there so
8:54I was introduced to AI agents about 13 months ago and ever since then I’ve been
8:59obsessed um one of the co-founders of relevance AI was introduced into a community and that was when I was just
9:07booked he gave a demonstration of what they’ve built and their platform and
9:13yeah it was just the rest is history so here I am just over 13 months later um I
9:18self I was self-taught so I had to teach myself how to build agents specifically
9:23with relevance AI simply because there was the documentation was very um poor
9:30um coming from a non-technical background as well was really difficult the learning curve uh without any kind
9:36of resources it was a lot of trial and error um but they do say that sometimes
9:42uh failing um actually helps you you know develop the skills quicker and better um and that seems to have
9:49happened and worked with me thank goodness um so I’ve done a lot of screaming and kicking around with the
9:55the computer didn’t break anything but I I finally um managed to to pretty much
10:00nail it and so that’s where I am now um all right so we’re going to get now
10:08into what are agents so agents I love this whole analogy that agents are like
10:13digital employees um when you look about when you think about you know when you go hiring a new employee into your
10:19marketing department you know there’s the hiring process you have to go out you have to find them do their skills
10:24match your uh job description um you know so essentially in AI agent is like
10:31that it’s a digital employee that you’re looking to hire but instead of going through a hiring process you’re going
10:36through a building process so you’re actually creating these digital employees for your department um to
10:43assist other employees when it comes to anything like Mar marketing related especially content production is quite
10:49you know content stuff is is really quite good with AI agents um what
10:55defines an AI agent they can execute tasks on their own um have the ability to complex problem solve uh they can
11:02interact with various external tools and they can work independently towards goals without constant human input so
11:11that’s where the whole um autonomy comes into it um which is just one of the
11:17those bigger defining things of AI agents so this is just them in a
11:24nutshell uh you got to think of them as upgrading from an AI that just answers questions like chat GPT where it’s very
11:30much a a back and forth exchange between you and and and chat GPT to one that can
11:36actually take action and get things done for you without constant
11:42supervision so now we’re going to try and shift away um shift more into what
11:47are the differences between an AI agent custom gbt which you might or might not
11:53have heard of and a traditional chatbot so if you haven’t heard of a custom gbt
11:58it is almost like it’s it’s chat gbt essentially that’s the the model it’s running of but you’re allowed to upload
12:06extra data and Specific Instructions so that’s it’s able to perform better in a particular area uh so it’s trained on
12:13targeting uh targeted content so it understands the nuances of a specific topic industry or task and it follows
12:20defined guidelines on style and tone so this is where you can upload things like uh you know various PDF files CSV files
12:27that contain brand guidelines um extra data points uh about your um
12:33you know ICP or your client client information for example and then so that
12:39output that you get from your custom GPT is really going to be tailored based on that I guess knowledge that you’ve
12:46uploaded into it so that’s what a custom gbt is now they do have access to
12:53external things so you can do like a Google uh like a search a web search um and other things like that but it’s
12:59quite Limited in how many external tools uh it can actually leverage uh and
13:05that’s where AI agents really excel in that area where you can actually Implement a lot of tools within your AI
13:13agent so you can do a Google search um you can do all sorts of stuff with external um apps as well like zenes for
13:21example um now moving into a traditional chat bot and I’m sure all of you if
13:26you’ve done any kind of shopping online I sure you would have seen the box that annoying box that sometimes I can’t find
13:32the X for for some reason um that pops up in the corner uh that’s that’s
13:37essentially a a chat bot okay so it has a lot of you know it will might give you some do you have any questions for me
13:43what did you know and give you some suggestions of questions um that is like a traditional chatbot um that you will
13:50most often see on on websites where you’re buying things oh not necessarily buying things
13:56but that’s the most I guess common place where you might see them all right so this is an analogy and
14:03like I mentioned I am an expert and I do love traveling so this is kind of just uh just an analogy that I obviously had
14:10chat gbt helped me out with as well so when you look at an AI agent it’s like a tour guide okay so it’s knowledgeable um
14:19doesn’t only just show you around but it also adapts your preferences it will answer unexpected questions it makes
14:24decisions on the Fly uh changing the route due to weather or for example if someone’s um well and they need medical
14:30assistance um they can even book tickets arrange transportation or solve problems that arise during the trip so the key
14:38traits that they’re autonomous they’re proactive and they’re capable of handling complex multi-step
14:45tasks all right now moving on to our custom gbt a custom gbt based system is
14:51like a detailed travel book tailor to your destination provides Rich information and insights about places to
14:57visit but it won’t act on your your behalf so don’t expect it to go booking a plane ticket for you cuz it won’t do
15:02that um you have to do that yourself so the key traits is that it’s context aware and informative but it lacks
15:09autonomy and it lacks action-taking abilities our next one is an information
15:16stand or an information kiosk okay and this is a traditional um chatbot I actually use mid journey I haven’t used
15:22it in ages to generate these graphics and I think this one is absolutely so cute um so traditional
15:29chat bot is like a fixed information kyos get a t spot it will provide answers to common questions um but only
15:35with it within a programmed scope if you ask anything outside of the questions
15:41it’s got in its um I guess brain or knowledge it’s not going to be able to help you and that’s when you’re going to
15:47get the annoying message of I’m sorry I can’t answer that for you I will pass you on to a human right and that’s
15:54generally how a chatbot will work the key traits is that they based and they’re very limited in
16:02scope so this um so you guys are going to get access to this presentation so I won’t go into detail but this is just
16:08kind of a um a bit of a comparison here between an AI agent traditional chatbot
16:15and a custom GPT so you’ve got all of these things on the left like purpose the task complexity and then just some
16:22very very brief examples at the bottom there um I guess one thing to highlight
16:27here is the autonomy level so an AI agent is quite High it can work independently toward towards goals
16:33traditional chatbot is low it requires human input for each step and then our
16:39custom gbt is kind of in the medium it works within defined parameters but needs user guidance okay so that’s a
16:46custom gbt a lot of people will get confused with that actually being an agent and it’s not simply because it
16:54really does require a lot from you still so you still have to give it prompts you still have to give instructions yes it
17:00has um all that extra information which is great which means the output is going to be a lot more tailored to how you
17:06want it but there is still that constant back and forth exchange um between you
17:12and the custom gbt so the big question is do you really
17:18need them um this is something where people will sometimes get a little bit
17:24um stuck they’re not entirely sure if they’re over complicating it if what they already have is enough um and if it
17:33gets the job done and so I guess the piece of I guess the recommendation I would make is try not to over complicate
17:41a workflow that you’ve got um if it’s already working so I recently had a call
17:46with someone who she just wanted to figure out is AI agents the way to go for her it turns out she had about 15
17:54custom gpts already set up and running each have their pre like their spefic specific tasks uploaded with the
18:00specific information um and that was it that was enough for her she didn’t need
18:06to transition into AI agents because her custom gpts yes it was a lot but they
18:13were fine um and she didn’t want to then make that further investment of time and
18:19money to be doing agents or building her own AI agents when the custom gbts were
18:25doing beautifully well so that just um is is a question uh to really think
18:31about um when you come across AI agents because and as you’ll see I won’t get
18:37ahead of myself but you’ll see that there’s a lot more to it than than a lot of people think all right so to kind of understand
18:46AI agents we need to briefly go over two tools um which are make and Saia zapia
18:53is a very wellknown uh they both do the same thing they’re essentially uh just cre create automations that’s all they
19:00are you set them up they have triggers um right so here is actually a good way to put it so it uses a list based
19:07interface which consists of triggers events that start a workflow and actions
19:13which are tasks performed as a result it’s very linear so you have you know it
19:18starts and it ends it gets from A to B and that’s it okay so zapia has about
19:247,000 app Integrations it’s great for users who prioritize ease of setup over Advanced customization make is one that
19:31is very popular um I see a lot of people using it I personally have used make
19:36I’ve never used zapia um but make uh for very very basic and light things um like
19:43I’m talking super simple automations where I’ve seen people build out very
19:49complex things and and it scares me if I’m really honest it does scare me with
19:54how complex somebody can can build it but for the sake of this it’s just um it
20:01allows you to to build more complex workflows it provides a visual flowchart
20:06style interface but make is actually quite more it’s more flexible and it’s
20:11lower pricing so the thing to get out of here is that automations um automations
20:18make up uh your AI agents so this is the reason why I need to introduce these two
20:24two tools that perhaps you have heard of before because they essentially go into
20:30automations which make up a big part of your AI agents all right these are just some of
20:36the the workflows but as you can see here the difference here with these now not to overwhelm you ignore this one cuz
20:42that one’s freaky um but it starts from here it ends here it ends there and it
20:48ends there you’ve got all these things things like this things like this and
20:54this all right they are conditions they’re rules they’re defined rules and and parameters that somebody who’s built
21:00this automation has set up so that the work so that the automation works and it
21:05functions but again it’s very rigid it’s very structured there’s no um
21:11flexibility in there there’s no autonomy so it’s not able to just if it fails it
21:16fails if it works it works um and and that’s it so it gets from A to B and then you’re done
21:23right so now we’re going to introduce some of the um uh the AI agent building
21:30platforms just to give you an idea of what they’re capable of so relevance AI in a nutshell it’s a platform that
21:36specializes in making your workflows so your automations smarter by adding decision making through AI instead of
21:43just following a set of predefined steps right these are predefined steps
21:49essentially okay um it can make it can take action based on context
21:55understanding data and the input for example relevant AI could scan thousands of product reviews and social mentions
22:02identifying reoccurring themes all right using natural language processing
22:07generates data backed content ideas such as blog posts addressing common concerns
22:12or Instagram reals highlighting popular product features essentially does more thinking and decision- making in your
22:18automations okay that’s that’s really when you get down to the you know the core of it that’s essentially what it
22:24does um and what these AI agents with relevance AI
22:29do okay this is just an example of what you might see within relevance AI so it
22:35really does go into the definition of you know uh you’ve got a Content production manager up here right and
22:41then you’ve got another three I guess what we’d call sub agents so um other
22:46employees of your digital agency um right so we’ve got our content organizer
22:52we’ve got linkoln um that’s actually quite smart with how I name that uh which if you’ve guessed is for LinkedIn
22:58um and then we’ve got our guide writer as well so as you can see it’s delegating um the tasks to the sub
23:06agents right and then in that you’ve got conversations happening so information
23:11is being passed from One agent to another agent the task is done that output or the answer and results from
23:18that task is then being passed on to our LinkedIn agent and then so on and so
23:24forth until then everything’s done and then it comes back up and our content
23:29production manager here is going to provide an update for the user it’s going to be me or it’s going to be somebody else with the content that’s
23:36been generated so the key features of relevance AI it’s intelligent
23:43decision-making doesn’t just pass data from point A to point B like our the make um automations which I showed you
23:50just before but it can interpret classify and decide the next best action AI agents you can set up agent likee
23:57workflows that make branching decisions automatically depending on context
24:03context awareness uh it can adapt when the situation changes uh so if a user’s message is negative internal indicates
24:09urgency this is really fantastic and I’ve seen this happen with what other people have built where that an agent
24:16has been um has been set up to handle inbound uh
24:22so emails that are coming and they’ve actually opened up an entire email thread that’s happened between agent and
24:29a potential lead and this lead has just gone back and forth um between this agent and the agent did make it aware to
24:35the person as well that they are an AI agent this isn’t you know um a human but
24:41it was just so interesting to see this back and forth uh exchange between a real person and the AI agent via email
24:48was incredible um natural language processing and sentiment analysis so it
24:53can read text okay so like I mentioned emails or chat messages and figure out sentiment and tag them accordingly uh it
25:00also has the ability to get human in the loop so for complex or sensitive issues you can set it up to involve a human for
25:07final approval now this is really great for when you have things like um content as well since obviously I’m coming from
25:13a Content marketing background um there always needs to be a human in the loop involved uh so that for the reviewing
25:20and editing process of content nothing’s going to be straightaway scheduled uh via an AI agent it’s always going to
25:27have that reviewing process involved okay and the scalability if
25:32you’re dealing with massive data sets or complex tasks relevance AI can scale to handle
25:39it all right so NN um all I have to say about n8n and I
25:46will go through with this it’s a lot more technical than relevance AI um it’s a lot more advanced it’s highly
25:52customizable um you you kind of think of it as a power user version of of make it
25:57offers a similar drag and drop experience but it gives you a far deeper control over how your automations run
26:03and the biggest defining thing with n8n is that it gives you the freedom to self
26:09host so this just allows you to take to take the data oh sorry to take the the
26:16um the product itself and I guess the code um and it allows you um to it’s
26:23open source so it’s a technical way of just going you can go under the hood uh your technical team can really look at
26:28how it’s built and they can change it um so that’s a massive defining difference
26:34between also NN and relevance AI which relevance AI doesn’t offer this um but
26:41it’s great for for big businesses uh that have stricter security or compliance requirements so that’s a a
26:46huge bonus of Ann um but also the level of complexity uh as well but it is more
26:53technical um and it freaks me out because I also tried to use n8n and I’m
26:59just starting to to learn it um and it’s if you’re a non-technical person like me
27:06um it’s uh definitely a little bit freaky to look at so this is something
27:12uh that somebody has done um an automation with n8n um and as you can see it’s actually quite similar to make
27:18so it’s very much you’ve got your little modules here and then you’ve got the branching connectors to each um module
27:26you could call them nodes as well and that’s essentially what it would look like
27:32um yeah so the key features uh it’s powerful branching and merging it’s cre
27:39complex branching workflows that handle multiple PS at once like make and an
27:45integrates with a lot of apps so it has currently a lot more app Integrations
27:50than relevance AI does um it’s Community Driven which I think is actually quite
27:55brilliant um because it’s open source means there is a large community of developers that are creating new
28:01Integrations with NN um which is great because it’s just always growing and
28:07there’s always things being built with an an um that just adds that extra level
28:14of advanced complexity and and flexibility I guess with it as well so it’s Advanced AI agent you can build
28:21sophisticated AI agents with full control over implementation and integration open source you can run n on
28:27your own ser service for full control compliance and security and custom code yes you both with relevance Ai and N you
28:35have the ability to also add in code but we don’t need that uh because these are
28:40low SL no code platforms so you do have the ability to add code in there but you
28:47can also accomplish a lot without any of any coding nothing none whatsoever
28:52pretty much so um but it’s just good to know that you’ve got that there especially if you’ve got a technical
28:58person who who wants to add something um that is currently not offered within the platform and they can do it with code
29:06instead so when to use what so how do you know when to use make relevance AI
29:12or NN so here’s a quick guy um so choose make when you need an automation tool
29:18that’s easy for anyone to use um so the documentation that make provides is
29:23really great they have their own Academy if you need a simple automation setup
29:28um make us make us great for that um if your tasks follow the same steps every
29:33time okay so like I was saying it’s quite rigid uh you’ve got rules set in there so as long as the the workflow is
29:40or the task sorry is the same thing every single time then it’s great um if you’re connecting to a bunch of apps in
29:47a very straightforward way if you need something affordable and you just want something uh to get something simple up
29:54and running fast when to use relevance AI your tasks
30:00needs some decision-making power your workflows might change based on the situation you need a human to review
30:06things you need more flexibility in how things are handled you’re serious about leveraging AI Beyond Simple triggers and
30:13actions which is M something like make n n you need a more robust
30:21production ready solution you want to consolidate your Automation and AI agent implementation in one platform you’re
30:28dealing with high volume tasks that need to run reliably and advanced error handling you want the flexibility to
30:33self-host a better security control and compliance reasons which is one of the
30:39top features of n8n in my opinion um if you have a technical team that’s comfortable with working with a bit more
30:46complexity um if you are building things your own like you you’re doing
30:52that on your own um I would actually highly recommend starting off with relevance AI um you saw between the
30:59difference in let me just go back in between this how it’s set up within n8n
31:05and how it’s set up within relevance AI the interface uh the user experience is
31:11very very different between each platform relevance AI is has been built
31:17to to be easily navigated and I guess a lot more user friendly in comparison to
31:23NN um yeah it just it it really does depend depend on what you want and what you
31:29want to get out of your agents and how much learning you’re willing to actually do as
31:35well all right so this is just um another table a comparison here between um some workflow types so let’s look at
31:42the marketing automation so when you use make um so automatically add webinar attendees to a MailChimp list send them
31:49a followup email with resources and so the agent is create personalized marketing campaigns adjust messaging
31:55based on user behavior and predict engagement rate uh let’s check out the
32:01content creation automate the publishing of pre-written blog posts to Wordpress and share them on social media
32:07researching um trending topics write blog posts using AI format them for WordPress and schedule on social media
32:14um there’s also a really good um add here to this one here as well is that
32:19obviously you’ve got the ability to uh for the human who’s reviewing the process to be able to send a message to
32:25your AI agent say look this actually not what I want can you go back and make
32:30this and this and this edit um so it’s going to take the users’s comments into consideration and it’s going to go ahead
32:36and it’s going to edit the post for you or the content for
32:42you so my personal and professional opinion as a non-key on AI agents and
32:47what platform to use um so I’m just going to give you a general um if you’re
32:53starting off building your own AI agents I would recommend starting off with relevant and say I now I will admit I am
33:00a little bit biased because that’s where I’ve started it took me a lot of months to learn it um but it’s in terms of how
33:08easy it is to navigate um it is is a lot more user friendly uh whereas just based
33:14on my personal experience when I signed up to n8n and I tried to figure it out I got scared and I ran in the opposite direction I just I just it just freaked
33:23me out so not to say that I won’t learn it because I will be learning it and I’ve started um but just to kind of I
33:30guess relevance a is that first step and then n an might be that second step up
33:35once you’ve kind of nailed it they are very different between each other how they’ve they’ve set up um is is very
33:43different so this is what I mean when building AI agents um and something that
33:48I will now get into is just setting some expectations here um the learning curve
33:57uh I’m going to jump down to this one here the learning curve is quite steep um you have to learn multiple things
34:03okay so you’re developing multiple skills at once now the reason why I wanted to introduce you to automations
34:10is because within either platform you decide to use you are going to learn
34:16automations it’s a skill that you have to learn um to get these agents up and running the the the foundations of them
34:24are automations they are those very linear um go to A to B um automations
34:30that you’re going to be equipping your agents with when you’re looking at a platform like relevance AI it is
34:37structured in almost like separate components your tools essentially are these automations that I keep referring
34:44to right A to B they’re very linear um and these tools act as the skills to
34:51your digital employees um so you’re going to be equipping your agents with
34:56these tools so that it’s able to perform the tasks you give it um so automations
35:03is one thing um you also need to have already a basic knowledge of prompt
35:10engineering yeah um you do uh because these agents especially if you’re
35:17starting off with relevance AI each agent is programmed based off a
35:22prompt um and now there are different ways to prompt your agents and it’s very much an agent-based prompting technique
35:31it’s very much a Chain of Thought prompt but again you have to understand how it
35:36works for your agent to be able to I guess function properly um because
35:41you’re going to be specifically mentioning things like what tools to use how to use them how to then delegate
35:47tasks to your sub agents um you know so it really does uh prompt engineering is
35:54really really important to at least have the basic knowledge and understanding of of how to construct um a Chain of
36:02Thought prompt um for example okay um the learning curve is is
36:08is quite steep now everyone each to their own with how they how they learn I’m someone who um can learn really
36:15quickly uh but likes to learn the hard way by failing multiple times when I probably didn’t need to go down that
36:22path but I do um so it like it kind of all ties into the multiple skills I mean
36:29you’re you know you’re you’re learning automations if you’re not coming from a non-technical background you’ve got to learn that okay then after that you’re
36:36learning how um agents work um prompting how to get your agents to actually
36:41follow instructions um then you’re actually looking at well how does an AI
36:48agent how are they structured what is when you look at a human Department you know you’re always going to have like a
36:54manager then you’re going to have employees under that manager um and colleagues right it’s the it’s a very
37:01very similar thing here with your AI agents you need to also learn how I guess agent hierarchy Works um and this
37:10is what I might mentioned here is before you actually go into building anything
37:16you need to understand okay well what is it you’re automating um what process now
37:22you can go some of these things can be quite complex because AI agents are extreme ly autonomous in what they can
37:29get done so just think of um the first process you got to plan it you got to
37:34strategize it then you got to plan your hierarchy um how many agents are you going to have then you look at okay what
37:41are the skills I want my agents to have and these are going to be your tools um then that’s when you get into building
37:48um but even so just to have that overall plan is fantastic and Brilliant just so you’ve got an idea of okay where one
37:56thing finishes another thing ends and then the results of that previous task
38:02how is that then being taken to my next agent who then needs to perform their
38:07tasks um so it is a lot of planning it is a lot of thinking about how it’s all
38:14going to work before then you build um and it isn’t a plug-and-play situation
38:20um using templates is really quite difficult especially if you don’t understand the nuances between how how
38:27they set a set up um I’ve had people I’ve seen people set them up and it works then if they want to customize it
38:33that’s when they’re running into issues because they’re making changes and then the tools not working or the agents not
38:39working and they’re not quite understanding why it’s because they’ve made this little change and this change
38:46has actually just almost um stuffed everything up uh because they don’t
38:51understand how that one change has affected everything else so it is really nuanced um
38:58I would highly suggest just going into one of either NN or relevance a just learn it and just learn one um to begin
39:05with um obviously I’d recommend relevance AI to begin with and then transition to n8n but just really
39:10dedicate your time to learning one Platform One platform only just so you’re not getting overwhelmed um
39:17because that did happen to me multiple times uh more times than I can count so but I’ve I managed to crack it did take
39:23me ages but I guess um you guys are going in it with just hopefully I’ve set
39:29these expectations for you as well um just expect that it will take a bit of
39:34time to to learn and to actually then Implement and then obviously you’ve got
39:40the the iteration phase and then the um error handling as well CU sooner or
39:46later something is probably not going to work out or your agent hasn’t responded in a way that you wanted to and you’ve
39:51got to be able to debug that um okay uh thank you and um I hope this
39:59has kind of given you guys just a bit of a snapshot into what to expect when when building your agents just kind of how to
40:06equip you with with a bit of knowledge um before going into that as well it it’s a fun process it’s a great process
40:13I’ve learned a lot um but yeah it’s it can be challenging at the best of times
40:19but the it’s it’s worth it it really is worth it especially where um technolog is heading these days so um yeah so
40:28thank you thank you so much Ash thank you for walking us through that because I think there’s so
40:33much confusion in all of this you know and so the fact you’re able to there was that
40:39moment where you presented that table even halfway through which was just the difference between an agent a custom GPT
40:47and uh just a plain old plain old AI uh you know plain old which you know is
40:53it’s one of those things that when you’re presented with it um uh go you go okay I get it but like I think if you
40:59ask nine people on the street uh what’s the difference then uh you know the
41:05people wouldn’t have had it presented to them that way so thank you so much uh that’s endlessly appreciated and thank you there’s
41:12there’s um there’s some open questions from the community so folks if you’ve got a question that you like in the Q&A
41:19give it a thumbs up and we’ll make sure to answer that um I’ve also got some questions that I’d love to uh to go I’m
41:26just smiling at Victoria’s chat comment here saying mind blown um and we’ve got uh folks saying
41:34thank you so much as well so I hope you can see that because there’s a there’s a lot of gratitude coming your way right
41:40now Ash awesome thanks everyone so um first things first i’ I’d
41:46love to know um because there was one one of the comments in the chat here was like
41:52uh what AI agents have you built because I think it’ll be interesting to unpack
41:58you know your process going into this and why you chose to do an AI agent rather than a custom GTP Etc on this
42:04learning process that that that you’ve been through yeah so
42:10um let me start with let me start with with content
42:15stuff so I’ve actually currently built something for someone which again is um
42:20on the basis of repurposing content so what I’ve built is something that is
42:25able to take very very long um video transcriptions um and is able to really
42:32dissect uh that amount of content what happens with llms and when you’re going
42:38into chat GPT or a custom GPT is that they they they kind of fizzle out the more you give it um that the response
42:46can sometimes by the end of it and you can see the difference and I I tried it I saw an insane difference between the
42:52top six paragraphs and the last five paragraphs and the difference it got more vague um it wasn’t able to just
42:59retain that context so what happens with the ability of an AI agent that I’ve built for that specific purpose is that
43:07it’s able to split the text in a way that it’s able to retain the context from the previous um paragraphs I would
43:15say um while then still maintaining that level of detail and the tone of voice
43:20and stuff like that so that is for an AI agent it’s just that next level
43:27um I guess of of customization and flexibility of what you can actually do
43:32with the content you’re dealing with whereas a custom GPT you don’t have that ability to to
43:39summarize it text to do it in chunks to do all sorts of this data handling um so
43:45that it maintains that context awareness um and then of course I’ve got it doing multiple things like generating graphics
43:53and all this other other stuff as well and that’s actually been a lot of fun to to build another one that I’ve I’ve done
44:00is for my sales process um so it is a lot of um ex I’ve use a couple of
44:06Integrations with external tools as well so um but essentially what it does is as soon as someone books a call with me um
44:13I’ve got an agent that will research that lead um it will go do Google research you do LinkedIn research it
44:19collects all of that information um and then have an agent who does some more
44:25data things I can’t remember now but it’s does a whole bunch of other stuff and it generates me a report um as well
44:31based on that lead so that I’ve used that um for my own process because I noticed just how long it was taking me
44:38to research people and that was starting to get annoying so I was like you know what I just build an agent for that so
44:45that’s they’re kind of the two main things yeah love that I love that so much and I think um an obvious follow
44:51followup question in those instances is why did you choose to do an AI agent rather than custom G GPT but the fact
44:57that you described it there as well is really really helpful because I think you pointed to elements of knowing the
45:03technical limitations of the other platforms so like you know not being a to handle and like goodness I’ve seen
45:09that too uh you know sometimes you catch out your your AI platform he said you didn’t actually Analyze That bit did you
45:15and so you know I’ve I’ve experienced that and so thank you for explaining that I think knowing not only how you’re
45:22using it but also the thought process for deciding why is is really really important um yeah yeah there was there
45:30was one more question that I have before we go into the community questions um which is what part of the AI agent hype
45:40are you are you just not having or are you not buying into because it feels like there’s a lot of chat about AI
45:46agents and um the the interesting thing for me in
45:51in uh watching your presentation today is is not just the information but also how there is a big learning curve and I
45:59think Folks at the moment are speaking about it as like the future and and like
46:04how it’s it’s here and it’s going to take but actually it seems like quite a technical product or a technical
46:10platform service and so for me my observation is that some of the
46:16hype is perhaps a little bit unwarranted because the learning curve is going to take several years for folks rather than
46:22sort of immediately but that’s my lay person uh perspective what’s
46:28yours um yeah so I I think um there are
46:33a lot of people that will we’ll talk about agents and how good they are and how wonderful they are and they do this
46:39and they do that um but when it actually comes to yeah to to building them like I
46:45said just setting that expect it it takes a lot it takes a lot of time and
46:51this is why when you go to someone who wants and you want them to build it for you this is why they are charging a lot
46:57because the amount of time and effort it goes to actually building something is a is a lot um and then people who might be
47:06talking about it are generally just talking about it and that’s what they’re doing they’re not actually building it
47:13um so I guess they they’re just adding to the hype but to really kind of cut
47:19through that noise and just set it straight just don’t expect it’s going to be an easy breezy thing to do it because
47:25it’s it’s not and then depending on how you learn as well um you know so yeah
47:32love that I love it’s been a while since it’s almost say easy breezy but um I I missed that and I am picking that up
47:39into my vernacular so thank you very much uh let’s let’s head into the the
47:46community questions um and so what I’m gonna do I’m gonna I was gonna say I’m
47:52gonna prompt which has a whole new context in this conversation um yeah but I’m going to mention this first question
47:59but then I’m going to head into the second question and the reason is the the first one asked for an example and
48:05when people ask for examples uh I don’t know about you but my mind goes blank so we’re gonna come back to this one in a
48:11second so this is the first question that I’m gonna I’m just gonna prompt which is what’s the best AI agent you’ve
48:18seen or built within the marketing sector uh who is nailing this so we can see what’s possible and replicate um so
48:25that’s that’s the question so an example we’ll come back down to this in a moment okay um let’s go to the next one though
48:33and we can have a chat about this uh for the moment so the next one comes from Ry and uh Ray says I’m so excited about
48:41this um what should be my dream big goal be for B2B marketing agent usage so um
48:48what’s you know if someone’s excited about this and they’re starting to dream big uh what would be uh their what
48:57what would you encourage them to do
49:02um I would take uh take Little Steps take
49:08baby steps when it comes to if some if something like this an AI agent or whatever it might be that you
49:15want to or AI in general even just um I think you’ve got
49:20to try to cut through and filter out all um the noise um that are some wonderful
49:28people out there who really do um provide a lot of value and finding that
49:33and and just a sticking to it and learning um as well but without getting
49:39distracted with with everything else that comes into um marketing and Ai and
49:44and everything like that so I think you really got to just um pick a couple of
49:49things whether they’re tools whether they’re people same thing applies to people that you’re following and just
49:55stick to it um and try to filter out everything else I think the one reason why I’ve done so
50:01well as I have is because I stuck to relevance Ai and I didn’t go shopping around for anything else I was like you
50:06know what I’m going to dedicate my time can see that this is going to be really steep to learn um and and I did it yeah
50:15so it’s just having that that Focus I think is really important nice I love
50:21that it’s um there there’s nothing quite like um creativ within within boundaries and and
50:28I think the same applies here right you know if you got that Mastery of something uh then then you know you
50:35enable yourself to dream big which I guess would also hopefully make it more fun as well frankly because um yeah you
50:42see progress rather than sort of feeling like you have to build the Dream from day one um yeah exactly there’s there’s
50:49a question that’s com in um from theal uh which is also echoed in the Q&A so
50:56it’s the same version of the question um which is how do you make an AI agent reliable or make it actually do
51:04what it’s supposed to do which is one of the biggest concerns because I guess you’re sending this thing off to go and
51:11do things right yeah so how are you how do you tend to build in these sort of
51:16checks and balances to make sure that it’s not out there causing damage yeah right so I think
51:23um when you finish building the is a phase where you’re testing um and that
51:30can go for a while depending on the consistency of the output so you need to
51:36ensure that when you’re before you’re deploying these agents you’re testing it and you’re testing it and you’re testing it and you’re refining it so if the
51:43output isn’t exactly what you want or you don’t want it to respond in a specific way then it will mean then
51:50going back in and refining that process and how you do that specifically within relevance AI a lot of it is to do with
51:56your prompting um your your agents are equipped with a very detailed prompt
52:03with how it’s supposed to work how it’s supposed to instruct if there’s any kind of ambiguity around a specific thing
52:10that it’s doing and you notice that it’s doing it wrong or it’s not just quite nailing it then perhaps you need to
52:16refine that prompt to make it clearer so that it becomes more reliable and you’ve got two ways to be able to do this in
52:22relevance AI you’ve got the orang uh so I think I think it’s custom gpts I get
52:28my terms mixed up custom gpts or even GPT allows the or instructions or it’s
52:35custom instructions I think they call it where you’re able to tell it what um how
52:40you want it to respond in what kind of way you want it to respond it’s similar with an AI agent and that you do with a
52:46a very long large prompt or you can also do it in their flow Builder and their
52:52flow Builder just allows to add I guess another layer of of um reliability and
53:00consistency as well with with what your agent is is doing and the flow Builder
53:05is very is it’s structured very differently but it’s taking it in chunks
53:10right so you’ve got the instructions then you’ve got any conditions if this then do this instead um but it just yeah
53:17it adds to it as well so nice it helps it reminds me of one of the points
53:24that you made in the presentation towards the end there which is like having an awareness of what the
53:30hierarchy is going to be before you go and build it and it feels like that for
53:36someone like me who rightly or wrongly will go into these things and just start building
53:41probably a little bit mindlessly that doesn’t feel like that’s the right approach here I mean how do you actually
53:48manage that process of building that hierarchy and and starting that process because that
53:55thought process up front feels really really important but it’s perhaps not a natural skill for for some people yeah
54:03yeah that was if I’m honest that was something that actually took me the longest to learn was to actually have
54:09this uh I guess this higher level thought of thinking of how everything is
54:15structured um so yeah uh how I normally do it now is
54:23well I actually I write I actually don’t use any kind of app or anything like that I write it down um and I’ll draw
54:30diagrams in my my journal that is just full of scribble and all sorts of stuff in it so that’s how I just so I can
54:37visually see at least mapping out what agent is going to be connecting to another agent and and what have you um
54:44but I am also creating a um an app that
54:50allows you to be able to program that visual hierarchy so it’s very much like a mind map you always going to start
54:57with your your top agent and then it gives you the ability to be able to add all these other different agents with
55:03the specific things you need for an agent to work so it’s going to have the
55:08description the inputs and the outputs so that I can visually see it um
55:14relevance a is also releasing a new feature that allows that as well n8n has that
55:19visual thing where you can actually visually see it uh which helps a lot but
55:25I guess when ites comes to planning yeah it’s just you’ve always got to start with one point and then Branch out as
55:31much as you can and get into as much detail as you can um don’t worry about the building the building is a next
55:39thing at least having a brief idea and overview of how it’s going to be looking
55:44how it’s going to be structured helps it definitely helps before building yeah love that so much
55:51um thank you there are so many questions here and and I want to make sure that we priortize uh the community ones but so
56:00I’m going to get to those I want to know about the limitations of AI agents but we’ll probably have to do this for another day so uh let’s let’s take one
56:09that’s been echoed in a couple of questions in the Q&A so we’re making sure we’re ticking those off and so um
56:17Trevor and Rona have both asked about how data gets passed between the
56:22different platforms and um so Rona has is data protected for example do we need
56:29to disclose that the information will be shared with the parties Etc um
56:37presumably the answer is yes um but I’d love if there’s any Nuance uh here that
56:43you’re aware of um so in is this in terms of like how data is is passed
56:50through different system yeah yeah to yeah so so I guess with
56:58um yeah well yes um but I’m not entirely
57:04well-rounded or quite sure of just how the data so I mean how data is passed
57:09and how it’s protected between between passing of information and stuff like that it is a bit of a um Ray um area I
57:18think as well uh but if you’re using chat GPT then there’s always a risk uh
57:24there’s always a risk for for for data and protection and what’s being passed and who’s taking what and all of that
57:31kind of stuff so it is something to be aware of it is something to be careful of um I would always uh you know any
57:38kind of sensitive information especially if you’re building these agents for your clients and sensitive information is
57:44going into those AI agents you do have the ability to anonymize is that how you
57:50say it anonymize where you right okay that’s it yeah um certain information
57:55being passed from one app to another app or you know if you’ve got other external
58:01Integrations with your agents and passing all of that kind of of stuff as well so I don’t know if that’s really
58:08answered the question it might have been a bit vague but yeah it’s one of those that um I think a general takeaway is
58:16like be wary and be aware and and build according you know which is is a great
58:22takeaway and solidify that um yeah let’s let’s come back to the uh the final so
58:29this will be the final question but it was also the first one I prompted up front so uh do you have any uh favorite
58:37examples in marketing of how folks are using AI agents currently um just so
58:44folks can uh here here’s some case studies I guess yeah so I guess uh one
58:51of the the biggest ones um which was actually quite fantastic was one that I
58:58saw being built for I think it was for an e-commerce uh so an e-commerce client
59:03and it was all to do with with content and all of this kind of stuff so they’d set up a very very complex system I
59:09think it was about 10 or 15 different agents um but essentially what it allowed them to do was to take um
59:17various uh uh pages from their products that they have listed on their website and generate um uh descriptions and
59:26pages of of content for that so products descriptions and stuff like that and what have you uh that was all SEO
59:32optimized um it had U links to uh databases where all of the information
59:38of those products were um kept um but I don’t think it was just content I think
59:44it was doing a whole bunch of different things as well but like I said it was about a 15 agent um setup so it was uh
59:53quite complex and there are did involve a lot of different different external Integrations as well with it um so it
59:59did a whole bunch of research as well uh competitor analysis it was generating
1:00:04reports as well based on that competitor analysis um as well as doing the content
1:00:09side for for their individual products and then it would give um suggestions of
1:00:15internal links so it would link to to other products um which was quite quite
1:00:21something um and I’ve then seen other people build about you know upwards of
1:00:2720 agents you know um don’t ask me the ins and outs of that one cuz I I’ve
1:00:32totally forgotten how it how it all works but it was 20 agents it was quite large um I think um the main things as
1:00:42well is people are doing a lot of stuff with with LinkedIn LinkedIn is a big
1:00:48platform um it seems like in this day and age to to be on as a business owner
1:00:53um and I think someone has also set up a way uh to do the whole Outreach system
1:01:00on LinkedIn um so that means uh collecting people um researching them
1:01:07doing all sorts of stuff on profiling um analyzing and then it will actually
1:01:12reach out to them via LinkedIn or um email and then having agents that then
1:01:18do the back and forth exchange via email as well as through Linkedin uh which is
1:01:23quite quite something so that’s just just another little thing there as well I say little but it’s not little they
1:01:31they can be profound can’t they they really can be profound for business um yeah with all that said Ash you’ve
1:01:37nailed it so thank you so much I mean I hope in the side of side of your eyes you’ve been speaking just now you’ve
1:01:44also been able to see so many people saying thank you uh as well for for
1:01:49sharing your knowledge today so like it’s uh it’s endlessly appreciated and I I think in this like I was going to say
1:01:57in this rapidly changing World which makes me sound like an AI an AI right you know there’s something to say about
1:02:03that it seems like some some of the language that AI is saying we’re now replicating that with our own language
1:02:08and our speak so it can get very philosophical very quickly just
1:02:14say let’s not do that let’s not that’s a rabbit hole for another day yeah yeah
1:02:19absolutely uh but thank you so much for taking us through that today um for introducing so many of us uh to uh AI
1:02:28agents and and I hope I’ve seen already in the in the chat feature that you’ve really cleared a lot for a lot of folks
1:02:34so that is endlessly appreciated uh and thank you to everyone for your fabulous questions as well today um I really hope
1:02:41it’s helped you um you are the absolute best every week um so with all that said
1:02:48uh we’ll see you all hopefully next week um but in the meantime a big big thank you Ash a big thank you to all of our
1:02:54sponsors and uh yeah we’ll see you next week cheers thanks everyone thanks jry
1:03:00thanks everyone have a good one bye