Mistakes Marketers Keep Making with AI and How to Fix Them (APAC/EU)

Hello. Hello. I see you guys are slowly trickling in as we have started today's uh live workshop, live session. Uh I need to check a few things before we begin. If you're here saying hi, you can uh say hi to each other in the chat where you guys are dialing in from. And hopefully you can hear me. You should be able to hear me. Hey guys, don't be shy. Yep. Yep. Yay. Awesome. Zoom has done really interesting things to change up the live stream capability. So I have to make sure that things are working just right before we begin. Okay. So just just hang on for a minute. Huh you guys. Uh hi, we can hear you. Hi, we can hear you. Yeah, if you cannot hear me, uh just make sure you have the right uh speaker settings on in your in your Zoom. It looks like uh it's it's not an issue for everyone else. So, okay. Meeting meeting. They really moved around the live streaming set because I have no idea where it is now. So strange. All right. Also, if you get out here, ensure you click the presentation. Yeah, it's a good tip. Hello everyone. Thanks for joining us today. Glad to see how everyone is uh filing in. Thank you so much for joining. Uh today's webinar on mistakes marketers keep making with AI and how to fix them. I'm your host Constance, product marketer here at HS. Today I've put together examples of how we use AI the wrong way and tips on how to fix them. A number of these are my mistakes in particular, but I've spoken with like team as well as shared with fellow colleagues. So I know that this is not unique to the experience that I've had when using AI, especially as a person who doesn't have a strong technical background or like a coding background. So I pulled together best practices also from across the HS marketing team. So it's not just like learnings from me you're going to learn but also like tips and tricks that you'll hear from a number of our uh like HS marketing members today and you will um see them uh throughout the presentation. So today's uh session is 30 minutes but I have more than 30 minutes of material. So but here's here how here's how it's going to split up. I'm first going to talk about like the mistakes and then like suggestions and like tips on ways you can help to fix them and then I have a section at the end where I talk about just how we should go about using AI in general. Uh so those are like the bulk of like the the mistakes and the fixes themselves should be within the 30 minutes. Okay, but we'll see how the time goes. And a bonus if you stay until the 30 minutes. At the end of the 30 minutes, I will be giving away uh to three people a one month free trial of agent A. Okay? And this is specific to if you are not a HF existing HF subscriber. I'll tell you by the end why that is the case. Uh but yeah, so that's the bonus. So please stay until the end so that you can be one of the lucky people to win a one month free trial of a day. Okay. Oh yeah. I hope you guys like my webinar webinar thumbnail. All right. Awesome. So without further ado, uh let's do some quick housekeeping. Um we will first do the 30 minutes presentation. Then we'll give out the the giveaway of the three one month free trials for agent A and then we'll go into Q&A. And as usual, if you have questions, please use the Q&A function that Zoom has. It should be under your more in the toolbar. You should see like a Q&A thing. This is so that as chat goes up and um people like keep talking, I will not miss your question because I am uh hosting this today's live session by myself today. So um if you ask it the question only in the chat, I may miss it. Okay? So please use the Q&A function so I can help to answer them after the first 30 minutes and the giveaway. And this question gets asked every time with every webinar. Yes, this session will be recorded. It is uh I mean I was trying to set up a live stream to YouTube. I don't think it happened. So it's not being live streamed, but of course I will upload the replay later on uh immediately after the session ends and so like um and the link to that recording will be sent to everyone who has registered. Okay. So don't worry, you will get to watch it again at the end if you like to rewatch a certain part or you like to share it with uh someone else who um was not able to attend or something like that. And this question about recording keeps happening. People keep asking it. So uh audience audience members please do me a favor when the next person asks whether or not this uh session will be recorded, please help to answer yes because I keep having to answer this question every time. Okay. So without further ado, let's begin with the first mistake and it's one of the most obvious ones. Mediocre output from AI. Okay, we are all um we are all guilty of this from time to time. Everyone be writing their AI slot, but no one wants to read it. Everyone is creating more and more content than ever, maybe posting on socials more than ever. But have you started to notice like everyone for at least on LinkedIn it started to sound the same. They keep saying things like not just this but also this. It's just this but not this. We have evolved beyond the m dashes but not beyond the solless way to talk about things where it's just relevant enough to like make a point but not actually you know being actually saying the getting the point over in like a concise way. In fact, I feel that this um it's better to just speak your own words rather than try to optimize it so that people read it on LinkedIn more because you just sound like an AI bot. So, and people are going to tune out if they think that you're using AI to write it. At least I do. But how do you make it not obvious? Uh I I have referenced this uh image from uh Ryan's blog on how AI short-term arbitrage with long-term strategy aka you should not be using it to like create mass content because everyone's like you're just going to make the same like bunch of stuff that everyone else is making and I'm not the first person who have told you this. I'm very sure. So the way that we create better content using AI is to split up the like creation uh process into different steps and put like human reviews gated at each step. An example of course the very um strong example we have showcased many times across our blogs but also like even our podcast and even previous uh webinars is uh Ryan's content engineering workflow which pretty much the everyone in our content use uh content team currently uses and or has modified to suit their personal needs. And what it really does is it splits up the content creation process from the beginning where you start with researching about what content you want to create to creating a brief that um shows the approach of how the the content like what is the purpose of this content and how it's going to like solve like a specific uh audience uh intent. uh to things like an outline and then a first draft and then subsequent drafts and then steps to make sure you reference your company or your product and and add in screenshots. These are all stepby-step processes where you you just ask AI to do the thing and then you stop and then you use the a AI to do the thing and then you stop. And you can apply this logic to any kind of task out there like creating uh a real workshop where you have to create maybe digital assets for it. You have to write down a copy about what this workshop is going to be about, who's going to present, and how you're going to post it in social media. These are all steps that you should just split up once at a time to make sure that what AI is generating is good enough quality before you proceed to the next one. And then AI sounds um will will create content that will look a lot less like everyone else and something that actually is high quality that you would be creating. All right. Uh yeah. And I think uh like Ryan will be posting some variation of this uh into as one of the pre-built skills in agent A very soon. Uh but yeah and and so AI doesn't replace expertise. In fact, experts are the best at maximizing what AI can do because they have the years of experience of knowing just the steps that a person should take to be able to create something of high quality. And that is true for like any sort of like specialist and stuff. So um even if you think like even if AI can and spit out this like content in one shot, don't do it. Um there there are reasons why these steps are there and it's really a a way of guiding thinking to make sure that like the the good checks of like um what you create and how to improve them at every step of the way helps to make sure your final product looks great. Just ask AI to rep like like um copy the process. Okay. Uh next and a lot of good marketing is good content. So interesting stuff, unique stuff. Great copywriting is only one part of the story. Just like like creating your video or creating your interesting images. In the end, you can't engage your audience if you don't have something original, unique to share. So use AI to give you the human more time to find and make cool things. And then let AI do all the boring stuff, right? We'll cover this again at the end. Second mistake, spending too much time and tokens. This is definitely true for anyone who has already spent quite a bit of time in AI, you will know what I am talking about. uh what if you like for example when if you just started using AI this would have been a very common uh experience where you start off by giving some instructions to AI just like it could be even like an innocent prompt of like hey AI please help me create um this uh application where I want to automate these three four five steps for creating a dashboard you know you just make this one prompt okay and it goes off to like make the dashboard for you and say here you go this is what you asked for and Then 80% of the time afterwards, the so-called like time savings that AI is supposed to give you is spent trying to fix the bugs and tell AI, no, you made this thing wrong. Uh, no, you've you've like certain errors are showing up or not showing up. They'll just say, yeah, the button's working. You click it, nothing happens. So all of your time gets spent in this fixing bugs mode and it's very frustrating because as a non-programmer you have no idea like why something is wrong when something goes wrong right um so I give like a personal example for this uh I helped to uh update and create uh the help articles for for hrefs like our main platform and I once had an idea for a project to uh create a system where it would take in all the information um that currently sits on AI on the on our help center and try to combine it with like our product updates and try to use AI to find an like articles that need updating like articles that out ofdate like suggestions and create like a system to also create new articles and review them and stuff like that. It was such like a unnecessarily huge project to start with. I thought like this whole thing could just be one prompt. So I gave it that prompt and it wrote eight implement sta implementation stages for plans of like oh here's how we should go about it. It took me a whole week to like um it started off by the making the plan part took maybe like two hours or something and then or and then afterwards like to actually implement like stage one, stage two, stage three took a whole week you guys. I have I did not even see like a working app until the end of the week. And of course, it's not working because you should go in and check to see if the different things that it said it built is actually working. And usually it doesn't get it right on its first try. So don't be like me. Start start with like one piece at a time. I should have just like started by automating that one step and then make sure that one step works and then already start using it in my workflow rather than like waiting for the whole app to be built, right? Um so after like um building things with AI for a while, this is now how I give uh AI context of something I want to build. I give it a plan. I uh would basically have a section that talks about uh how the flow should be like like you know it should have a homepage. It should have buttons on the homepage to go to different workflows on a specific workflow one. It should go through these steps one, two, three and four. It should give me options to A, B, and C. It should be able to show this information and then I have a different like either document I have a separate document or a separate section that talks about the calculations what is actually supposed to be calculated at this step and at that step. You just split up the plan and really try to go into detail as much as possible of what exactly you want to happen. And yeah, you're not going to be good at this at the first like time that you use AI because you don't know what you don't know you need to tell AI. You just sort of have an idea in your mind. Um this really does come with practice. But like if you want to avoid the mistakes I made, start small. Start trying to you know note down like the different details of how it should look, how it should function, what should be calculated, things like that, right? Um yeah, so I mentioned this earlier, build one thing at a time and you so like I you should build one thing, make sure it works, you can use that that like automated thing. So, for example, now if I want to create a new help article, it's one skill um where I ask it to take like a text dump and then like optimize it to a specific format that I I usually use to uh publish the help articles. And then I use a separate skill to help with creating the the screenshots. I use I also use another skill to manually paste in like sp specific articles and check for improvements. So that's just one thing at a time instead of creating like this whole uh help article optimiz optimizing like masterpiece software right uh and then if I want to create this big like orchestration like application later on I have the working pieces already right and then it becomes much easier to string along it will become much easier to test what is working or not working uh prototyping you'll notice if you have like if some of you guys have a a programming programing background, you notice that a lot of things are just concepts that are already known to people who are programmers. Uh but I mean as a person without a programming background, you don't realize why they're not uh like why programmers do it and they just turn out to be useful for everyone. Make a prototype create create like a initial thing. It could be nothing could be working but it could show the idea. It could you know put the things out. it could use maybe just a limited data set that you update upload as a CSV rather than use live API endpoints. Right? So this tip came from Matos where like he has some ideas that you he wants to share with with our team or even both the engineering team and the marketing team of things that you can do, right? He just prototypes it. He starts with something that doesn't work fully just to see how the idea flows and then gets feedback from other humans, not necessarily AI. So it's it's this becomes so much and you can use a cheaper like um LLM to do it um to use this first uh to create this and it becomes like cheaper for you faster for you and and like you'll fix all the mistakes at the beginning rather than like struggle with it at the end and after you spent like hundreds of dollars in tokens right uh use planning skills so if you're using claude or I think now a lot of like I think chat GPT has it as as well as many of the the main platforms will provide some kind of planning skill. So anytime that you have like your first prompt where you want to build something, AI is prompted in that skill to ask clarifying questions about when you meant to build this, do you need to where should it be? How should it work? How where should the data be restructured, should we use a live database or just files on your computer? Um should we uh take in data in like a CSV or will you copy and paste things in or is it an image? Right? um you know different questions like that you can ask AI to ask you the questions clarifying questions ahead of time um in claude I just um there's like a GitHub directory of superpowers as a skills thing you should definitely if you guys haven't heard about it check it out because this is the main skill I use in cloud specifically to start by like brainstorming a plan or like thinking about what are the details I actually need to care about to make something work and then plan it and then build you'll be saving so much like stress and effort that way. Uh and there's some platforms of course uh that have it built in of course like our platform agent A has it uh built in so you don't even need to download anything ahead of time. It already starts asking you questions. Uh yeah uh upload files and writing mega prompts. So uh your initial prompt yes uh like a lot of the AI platforms would tell you wow you can with every chat that you can give AI you can give up to like a whole novel description of what exactly you want to build in in real practice that's not very uh you don't get very reliable results that way. What you should be doing instead is splitting up like your reference documents and materials into separate files and uploading that and then referring to that. They say like um for for the you know planning the layout workflow you should use this markdown file for referencing the UI you should use the colors and styling of this image you know image a uh file name a or whatever so so this becomes more structured and then AI doesn't have to take an entire context in a single like uh digestion uh when it's processing a prompt and uh it becomes it also becomes cleaner for you to like reference what you gave AI later on and update it if you'd like to make uh improvements on what you're building. Uh so I you know like as I've built I spent less time and I spent uh less tokens. There was a uh so whenever I do these uh sessions live sessions with you guys um Zoom gives me like this CSV file that's a dump that tells me okay uh how many attendees attended the whole session or like they they only joined like halfway in and which country they come from uh what were the comments that they had the feedback that they have. It's like a CSV dump. So I used AI to create like a analyzer to to give me stats about like each webinar and then across multiple webinars how they how they performed and how they changed over time. The original project wasn't lovable. It took a really long time to build it. Um that's probably on my end not necessarily about how lovable was. I didn't really spend too much time on that platform but when I started off it was my early days of using AI. It took a really long time to build it and it was not great. it didn't like it it turned out to not do like uh 100% of all the functionality that I was hoping to do uh with my application and then uh I made it again in cloud code uh later on after like as a company we decided like not to use level as much um so I built it again in cloud code it took me a lot less time because like I knew now how to give instructions to AI uh about what I what I wanted and and then later on as as a company we decided to use our own tools. So like we all started to use agent A. So then I now like built the same thing in agent A in like just a couple hours, less than a couple hours. And it was just so much better. Like things were working um so much quicker like in in the at the very beginning. And it's just like a matter of like trying to do things over and over again until you figure out what are the best practices for it. Maybe the best thing to do is write it out. Pen and paper, guys. I like for for some today if I want to plan out something uh interesting I just um try to write down things without looking at a screen uh and and just actually think about what exactly I want to build and ask me questions like ask myself questions about like what part of it is it reasonably automatable that should be something that I I should build with AI and what is what is not really something that AI is going to do well just just start with that the more planning you do up front the better your outcome will be afterwards. Uh AI makes mistakes and struggles to fix them. And I I this is the next uh mistake that I'm going to refer to. And what I mean by that is when is specifically when we uh look at things like numbers not matching across an API call versus your vibecoded app, right? You know, like you can't figure out why these numbers are not exactly the same. You then copy and paste this to AI to try and fix it because that's the main way we do it and AI goes yeah uh you know I found the error and I solved it and then you load the app again and the numbers still don't match. So a lot of these um a lot of debugging and like problems and mistakes that AI does when it like takes in like API from like thirdparty connectors or and tries to implement something that you have planned. Um and then when you try to fix it by asking AI to to give like uh you know giving examples of like what is not matching and what's not working it will if you sort of not know the the tips and tricks in the beginning you will keep struggling at this point forever. Uh it's it's just like because like you don't know how the code works and then AI doesn't know how the code works and it's just trying to figure things out, right? I mean, it's cool that you're able to like copy and paste errors to AI and it can try to fix fix it out, but like um you know, there's limitations to that, especially if you don't know how the API works. Um, there was a uh live workshop I did last month, last month where I was featuring a dashboard that you can build with API endpoints in Hrefs that spend zero like uh API units in in in your in your subscription. So you don't have to worry about how many units is spending on your on your subscription because u the endpoints themselves don't spend anything. Uh, I thought if I would just say, "Hey, um, AI, here's my plan. Uh, use HF's MCP to like create the the platform." It hallucinated a bunch of different endpoints. like it just I think because the the database that I asked it to build was in like MySQL, it just assumed a bunch of um it just hallucinated a bunch of uh parameters and like so basically an example would be like uh fields that s share a voice. For example, if this is rank tracker and I want to monitor like my share of voice of my keywords, uh, and it was it was just trying to put in all kinds of random words instead of the actual fields that are supported in in that API endpoint and it just made it up. So, so like this kept happening across like different endpoints uh even though I asked it to uh reference HF's uh MCP. So the only way I got it to consistently work in the end is to actually build the API endpoints of what I wanted to calculate and give the logic like reference the logic of what exactly I want the calculation to be and give like the specific fields that I wanted to uh like use for the data to AI and say just take the data this way you know take and calculate it this way be very specific that will require you to learn how the AI API works. Um I mean like I don't think it's very uh like realistic to be able to get very good results with what you want to calculate, how you want to calculate things if you don't know like how the data coming out of the platform they're using API for like how it comes out. So um but the good thing is that actually you should be able to use either AI to to teach you how the API works. uh or for example on our platform we have like helper buttons like an API button where you can click and you can just copy the API URL already with all the fields like like filled in right so there there um there are ways in which you can like go about creating the API even if you don't understand everything fully as long as you check the documentation or you get help or just come forward to recognize that you need to get used to learning how the API works um so that you can create like the functioning uh application that you want to make um solution. So there are some tips that um that I got from Matos and Glenn. So one lesson is to save what AI learned for the next time. So like as you go into your session and you copy and paste your errors in and it fixes those errors, you say AI please take the lessons from what we learned in this uh session here and save it in a file like memory. markdown and and like things that you should avoid uh when trying to vibe code this kind of application. So you can ask AI to write it. You should also personally write it so that you have those notes in the future. uh and uh Glenn who is helping to uh like build out the different skills that we have uh in our agent a platform now is able to very consistently and in a short amount of time get like functionally working applications out just because you just iterate on this process of learning from your mistakes uh make a way to save versions of your work. This is uh something that I think if you're not familiar with coding, you should you probably haven't thought about doing. Uh but any versioning um software platform that supports git is a very easy fix. And like for example, GitHub is free and many other platforms offer some kind of free option as well to upload your work and it can be saved privately somewhere in your laptop as well. It's just a function of being able to save versions of your work over time. So that because like AI is prone to like creating something that works and then the next time that you you work on it, it will break the originally functioning thing or it may like completely miss out a page or you may make a decision like in the beginning that you want to bring back later on. So just having a way to save versions of work like saves you a lot of headache and time. Uh a readme file that describes what you're working on. So you may be working and coding on your thing and then like over time you forget what you've been working on and you need to go back and refer to what exactly you've built just having uh that at the inside of every project and you can ask the AI to write this for you. You can also add your own notes inside and this gives a reference point for both AI and you to uh reference uh what's happening especially if like you were talking to AI and you had to close the session and you want to start working on this project again maybe weeks later or months later. A file like this helps you to keep the context of what you're trying to do and like what was currently uh happening in this project so that it you don't have to start from scratch to try and figure out how to fix this project or uh this application that you're building later on. Uh okay so we are all we have three minutes left about the personal opinions time. I'm going to just cover a few of them and then go straight to uh our giveaway uh later on. So my main point I want to do talk about with like AI is that like it should be used to help you do like free you up to do creative work. Uh there is this uh book that is um was very popular. I think now it's become less popular because of AI but it's called automate the boring stuff and it was a book about like how um it's really for people that are not engineer backgrounds. uh that was that was showing them that here here are the things that you can actually use code to like automate and fix your life uh rather than you know creating like a cool application to get a job or something and and like this these were one of the books that helped me me personally like and this was like way back learn about what exactly you can do with um like programming and like how it can actually solve problems for you like as holistically and then it dived into like practical um applications but that's the approach you should take I think is very useful to take when you use AI because like um like you should not be looking at how you can magically put things into AI and then automatically create amazing things to come out. Um it should be helping you do the stuff that you don't want to spend time doing. Uh so busy work like creating tasks uh short codes translating transcribing content uh you know creating like notes from recordings like doing doing the stuff like like send repetitive emails that um don't require you to do any sort of creative thinking at all. AI is great for this kind of like like repeatable tasks to help you build them. uh don't automate things like uh developing a marketing strategy. Okay, I know there are companies out there that offer like a marketer CMO like AI CMO and stuff like that, but like um and I have like that it could be great. Uh but fundamentally as a person, as a human being, I feel like this stuff is something that it should be prioritized on like improving on your own personal basis. like you should be trying to be better at creating um like a strategy based on the knowledge you have about customers. Uh and I and I don't mean like I I mean specifically thinking about asking questions about like who's your audience and how you should reach them and what are the angles that you can get their attention and resonate with them. That's the stuff that humans are really good at like learning things. Okay, I know you have automated um creating notes out of that, you know, 4,000word book that you didn't you don't have time to read. Uh but like at least you should be trying to, you know, digest the notes in some way or at least or even read the the the the book manually yourself just to really see how like um what knowledge you can absorb and and turn into insights later on. Uh yeah. and and like there other like managerial decisions that like you know I mean really humans should should be using be doing this instead of robots. Uh okay so we are at the 30 minute mark. I can come back to this material later but let's skip all the way to the end uh which is to talk about agent A. So we at HFS use AI so much that we built our own platform to incorporate all the best practices for marketers into one platform. Uh the main things that are great about this is that um the first one is that instead of using like the HS API, it actually has a full like full connectors to all the functionality that we have on the HS platform and not just what our API endpoints support. Uh and that also means that there's a lot less of a chance to fabricate some of these API end uh endpoints and and hallucinate the parameters like it's just built into the platform. So you're never going to make this mistake. And the the plus is that using these connectors will not spend any of your accounts uh API limit. So that's really like a great bonus for you. Uh we built a lot of pre-built skills. Remember how I talked about how much time and tokens it's spent to like build these like like functionalities from the beginning? Well, if you have them pre-built by um you know, our dedicated teams that are focused on building skills that are useful for marketers like us, that's money that you're saving and that's time that you're um saving to already create like your your working dashboards or your uh analysis reports out of the box. So, that's another reason why it is, you know, I think it's unique and it's useful to use a platform like ours. Uh there's also like we don't talk about this very much but a number of quality of life skills. Just being able to export something to PDF was like if you you have to either find someone else that has built this skill uh or this application on GitHub somewhere and install it into your favorite AI platform. Uh or else it's AI will do a pretty bad job. So, um there is a number of like quality of life stuff that helps you create um you know there's surrounding stuff that helps you create a functional working application or dashboard from the get-go just because those are already there uh to help with the little things like this. It just saves you so much more time and and like money to to get something working. Uh easy sharing built-in by default. This is not unique to our platform like many other AI coding platforms um provide this as well but it's just I just want to mention it's like a like a like a two threeclick way to share what you've built with your teammates or to the public so you can create like demos which I did uh in last week at HS evolve or like share certain like um working applications with like my teammates who also wanted to build case studies or help articles and things like that. Uh you can choose different LLM models. So like as I mentioned earlier about how you can use the cheaper ones to do prototyping, you can choose out of the platform out of the box like the cheaper ones here um to create like the the early stuff planning prototyping and then use the more expensive uh like LLMs to actually implement or do the actual uh deeper work that um the bigger more expensive models are good at doing. Uh yes. Okay, we are here um at the one month trial. So why did I mention earlier that like the the three uh free one month trials is that's because all existing HF subscribers should get your one month already. Just go to uh hrefs.comaga and you will see uh this message where you've been selected for a free month of agent A worth $99 and you start your free trial. That's if you are currently a paying customer. Uh it should it's I think it's rolling out to pretty much all our um subscribers. So, but it may not be everyone, but I think by the end of this month for sure it'll be all of our uh subscribers. So, if you are an existing HR subscriber, of course, please check this out to to get access to your free trial uh as soon as possible. So, not an existing HR user, first three to say agent A please in the chat gets a free month of Agent A and you have to be I said agent A please. You guys got so excited. All right. Um all right. Okay. Um I can easily check who is not an HF subscriber. So please be honest like you now for this is the first three people. Uh Nicole uh Shara Lampos and uh Mick Payne. Can you please DM me your email addresses so that I can follow up with your free trial? Um there's a setting at uh at where you can send chat to and click host and panelists here at the at the bottom. Uh but I can't see. Yeah. So Jason um if you can't see this like message of you've been selected to for a free month of agent A uh that means it's not rolled out to you yet. Uh but we are like progressively rolling out because we have a lot of subscribers and a lot of users not just subscribers. Okay. Uh, and last person to send me is there you go. All right. Thank you for the three people that have reached out. Um, I'll get back to you via email with your free trials. Fastest fingers first and fastest fingers with the ride instructions first as well. Okay. So now we're at Q&A and uh I will do a Q&A first because I bet that I have slides to answer some of your questions based on what I prepared earlier. Uh let's start with the questions. Uh for those who can um would like to open up the Q&A, if you see a question that someone asked uh and and you really like their question, please upvote it. So I'll prioritize those questions first. Uh, John John Ames asked, "AI will often default to American English. They will reverse the logic of what you ask for if coding if asking for coding advice. AI can sometimes just show you code it found from internet searches before I implement any AI suggestions. Check it first. Also, send output from one AI to other AI." I think this is not a question uh but I think this is a tip from someone. I think um the main way to check if this approach works for building your AI coded thing is to test your AI coded thing and see how that works. Um so I guess it's a good tip. Thanks Amos. Uh I should go by who asked the question first. So let's go to Jeff. Jeff asked how long is the free trial subscriber for agent A available for? Um, like I think we don't we're just trying to get as many users on the platform as possible. So that offer will be there if it's your first time using agent A uh for one month. So I don't think there's a deadline to claim it. Uh although yeah, I don't think there's a deadline to claim it. think we're interested in getting as many people on the platform as possible to check it out, give us feedback on how it can be improved and if it's good enough, obviously stay with us uh and continue your subscription and yes, it is a separate subscription to your HS account. It is not included. Um Mick Payne asked, would you advocate providing an LLM.txt file for your website? Uh so I remember seeing like something that uh I think there was some discussion Lily Ray had with John Müller recently where he asked about LLM's.txt. It is it is like it probably does not hurt but it really is not necessarily going to move the needle per se to have this text file. There's so many other things that you should be doing first to optimize your website before you bother with this. I'm not sure if we even bother with LLMs.txt ourselves actually. So, we focus more on the other stuff like making sure the content of your like that that you have that you're willing to share about your brand is answering your your uh your audience and and like whether or not it's well, you know, structured SEO like optimized content, things like that. It still matters and it would move the needle more. Sorry. Uh let me pick Jenny's first question. How can I be sure that the skills in agent A is good for my case? Uh you there's a preview. So you can preview the skill before you use it. Uh are there no possibilities in agent A something to customize? So you can start with the skill. So that's what I've done uh is that I took the skill of what my colleagues have built to begin with and then I just asked um AI to optimize it. So you can create a version of what was already there and like optimize it for your needs. Um yeah, so you can preview the scale and see how it works and then and then you can see the output and then see uh whether it's it at least contains components of like what you want to build uh as well as like you know give additional instructions if you need to customize it for your case. But the skills are very useful for things like you know already figuring out the ways to calculate like existing HFS data or or creates um applications in a certain way. So like the main benefit to it is like it already builds parts of it that you don't have to worry about uh from the very from scratch. Uh Jenny also had another question which is how can I monitor the limit of the API budget HS API by using MCP? It will be in your account settings. Let me see if I can showcase it to you guys. maybe this page. Okay. Account settings. I have to log in. Uh but basically, yeah, I let me let me go back. I have to I have to log in and stuff. It's going to take time. Um basically uh in your account settings there will be a tab that shows like limits and usage uh for your account and that's for everything. So like uh API units will be one of the one of the liners. Uh that's how you monitor your usage there and it refreshes every month. Jeremy asked uh would you say that agent A is more accurate than claude uh plus HSN NCP to retrieve uh HF's data from rank tracker uh because it's using the direct connectors to our platform and not the API endpoints. Yes, it's going to be more reliable in bringing the right data directly from the HS platform into your agent A environment, your whatever your your uh vibe coded app whether will agent A hallucinate less than Claude. Um well, the thing is is that agent A is using depending on what you pick Claude's LLM anyway. So, uh I can't say that it will hallucinate less, but like at least for um pulling out like data from HFS, it would be you're going to make a lot less mistakes just because we have it built in already in the in the platform. But it is using like LLMs from cloud and open AI and deep sync depending on what you choose. Uh does this is the most well voted question. Cloud code versus cloud code versus co-work. When to use what? Do they have different use cases? I haven't been able to figure it out. Uh full disclosure, I have never used co-work. I'm sorry. I'm not the right person to ask about that. I believe co-work sounds to be more for like agents that run in the background and do things on a regular basis. Um, for those in the chat who are more familiar with cloud co-work, I would actually invite you to maybe answer uh GE's question here about cloud code versus co-work. But my impression is that it's like for building uh like agents that help you do stuff regularly uh or like without you typing to it. And they give some examples and stuff like that. I don't I personally don't really have use cases for this. Uh and now I'm on agent A like our company's platform. So uh I mainly see it like agents as like if if I were to use an agent it'll be things like pulling data regularly for my reports or dashboards. Um like I don't really use it for anything else. So answer Nicole's question. Are there any basic must have notes to have as lessons learned to give AI like in a markdown document? something the AI tends to do across usage is make sure it doesn't do uh so like if you if you you can give that so um a good example of how to give instructions to especially for crafting a tone for like social media or something I do gift uh give a list of these are phrasings that are bad these are phrasings that are good these are I also um have uh I asked like it's a it's a skill to help me generate like social media copy. It's a skill that also references uh specific uh examples that I've saved in the different folder. So I save like good examples of social media posts that I personally like uh into a specific folder and I say okay and I tag it. So I say like this is good for promoting live events. This is good for uh talking about uh specific tools on our platform or uh talking about new products uh releases that we done with that we've done. So I give examples for that and then in the skill I say if I am trying to make a post about X reference these examples. So that's just a good way of being able to create something that's more closer to what you want from the get-go. uh it still makes mistakes. None of this uh like I think uh bar Ryan's like ninestep process where he almost does not touch any of what AI has created at all which is really impressive. I still you know do some final touches with my copy at the end and uh I'm very happy doing that just because I don't want to be completely brain dead when posting something. I should know what I'm posting. Uh and and like I and and like sometimes I want to try new ways in in typing things. So like you're how you want to create content will also change over time. Like there's no like um flawless way of posting on social media or flawless way of creating content like I think expect that to evolve as you evolve. Yeah. Uh for people that are I've already given the giveaway so so you don't have to say agent A please anymore. Uh can agent A integrate with make.com animation? I think currently it does not but I can give that uh feedback into our team on how you can um have this automation for the future. uh like connect to make.com as one of the connectors and agent A. I can give that feedback. So I think it currently can't do that. Uh agent A is basically a harness with built-in skills and HF's MCP. Can it connect to other data sources, other MCP servers? Yes, it can. Uh but we are slowly adding the connectors uh where you can do that. So I I hope I can share it with you. Can I share it with you? Sorry. There you go. Let me enter my workspace here and show you guys. So these are the ones we currently support. Air Table, Apify, ChargeB, Cloudflare, Fathom, Fire Host, GitHub, Gong, HubSpot, Linear, Mailchimp. So we have a list. Uh we are going to keep adding it, but we add it like you know slowly. We don't allow just any MCP server to connect. Yes. So if you have a certain SCP service that you want to connect um please share it as a feedback to our support team and we will prioritize accordingly. But we're adding like every week we add like new new connectors that that we we've checked that we can approve and support. Well, okay. Uh that's all for the questions. Uh yeah, most welcome. I'm glad you found the the webinar thumbnail. I thought I was like, what a what a great opportunity to feature an actual AI fail. Um I I I do want to I think this is relevant to um the question that I believe was asked earlier. Uh let me quickly go back to my slides here to show you what I mean. AI can be a pretty good teacher and reviewer. So like um I when I ask for let let's say for uh HR's Evolve Singapore last week we had um a few of our team like uh sponsors come and do workshops for us like the day before and uh we were helping them craft the title and description and it generated some examples and based on examples I also then created an example and sent it to AI and I asked it to review it and say how Can it be improved? You know, here's a live description where I wrote, "How can it be improved to be more readable? Is there any information I was missing based on the original content that the sponsor gave me about what he wanted the uh workshop to be about?" And it gave its feedback. So, it gave some examples like it like readability can be improved like you're it's it's very front-loaded of information that you know doesn't really talk about the business um example of like why I should care about this workshop. uh things that were missing. So, um the advice it gave was pretty good uh in like for uh analyzing my original prompt to like my example title and description. Uh but the suggested rewrite it created was terrible. So, but because I gave I asked AI to give me critique, I used um the feedback that they give to create the final copy which like I believe fixed all the problems but also doesn't have the problems of this final copy. Uh so that was what ended up on the website for HF evolve. So like it this is also really useful because like you can ask AI to like be take the persona of a specific reviewer like that has information about so- and so brand. So I've done many years of like copywriting that follows so and so frameworks which you can research and say I want to follow like certain like uh frameworks or or uh techniques that are that are important for my writing. please be that persona to crit um critique my work and see how it can be improved rather than just ask AI to create like a copy because you don't know what they you know what rules or how they set up the the copy and you're and if it only gives you the rewrite without the feedback then it's not uh you're just going to think well you know AI just created a bunch of slop and actually it can also help you figure out new ways that to improve what you have created so I thought that's like something that is a useful to a way that not many people are actually using AFR, especially when you're writing like copy, especially if you don't have uh years of like copywriting experience or years of blogging experience or whatever, which not every marketer does have. But yeah, uh one tip I want to share for the rest of the people who are still in the session, stop worrying about if you're the 2x, 10x, 100x marketer. everyone is putting it in their profiles these days and like making posts about how they're using AI to basically, you know, fire their whole team or whatever. Um, I mean, social media is a bubble in my opinion. It it like you can just say whatever about like how you're 10x or 100x and just make everyone worried that you're falling behind with AI. Um, that's not going to help you actually use AI and and like practice using AI to create like actually useful things for your job. uh and it's more useful to think of it as AI can help you thing do things that previously you needed expertise for. So like previously you needed to be a programmer, you needed to be a data analyst, you needed to be a graphic designer, you needed to be an illustrator, you needed to be a person that can play instruments. I mean today AI can help you bring something end to end in like pushing a marketing campaign because it can do all the specialist specialist tasks for you rather than thinking about oh you have to be you know 10 times more productive like push out 10 times more marketing campaigns. It doesn't work like this. That's not really what I in my practice and like when like the best people using AI that that's not what they're doing. And many of the best users of AI aren't posting all day about LinkedIn anyways. there like this is Samo if you guys many many of you guys will know about Samo from our YouTube channel here's like his clustering internal linking tool that he has built recently to help find um pages and the the keywords that they link uh that they rank for and find uh ways in which they should link to each other. you can zoom into a specific cluster and see how the internal link should be built and what the uh like which parts of the application uh or or like the article should should be linked, you know. Um, our teammate Eric uh is uh head of international uh marketing here at HS and he helps he's built a performance tracker for all the different languages that we localize it for and um you know what percentage it goes to our like paid ads and what percentage is like you know finds its organic traffic and you know different different things that you can analyze. These are actually useful tools that you can um use for your day-to-day work and it's not like about something that is impressive and flashy as a demo. Yeah, I say LinkedIn a lot. It could be X, it could be, I don't know, Tik Tok, Instagram, uh whatever your platform choice. People are doing this and it's not the most productive way to learn about or or get, you know, really uh effective with AI usage. Hey, those are the main points that I want to cover that didn't get into the whole 30 minute uh front part. I see that there are no other questions. This is your last minute uh for everyone who may have some final questions that they'd like to ask. Um please let is there anyone that has any other questions or like topics that you like to that we should cover in today's session before we end? You can react with thumbs up if you believe we're all good to go or you don't have any questions. Hey, hey, I think we are good to go then. Uh thank you so much for staying back and uh thank you. I hope you found today's session very useful. Um when we when you close the page you can get uh you will get a feedback form to like share information about what you liked about today's session, what you feel should have been improved. Um of course I've always used that to help improve future sessions and it also helps me improve as well. So thank you so much. Uh so please share your feedback if you can. Um, look out for the free trial uh that should be rolling out to you as a paid subscriber of HRFS and please give agent A try so that you can tell us how it can be much better for you. Okay, so we really really value that feedback because we want to make something super awesome like how we've created what I think is a very fantastic platform with the main HFS platform. Okay, so thank you everyone for staying back and thank you everyone for joining today. Uh, I'll see you guys in the next one. Take care. Have a good day. Bye-bye.