AI Expert: Learn AI Agents Before 2025 Ends🔥[Build 5 AI Agents in 50 Minutes]

0:00 start saving like five or 10 hours a week. I save like 50 hours a week. It's like a double my output. And today we are with Flo in San Francisco and we'll be learning how to build AI agents from scratch as an absolute beginner. It's trouble to find a reason why humans should keep working. You can talk to an AI agent to help you and ask JPT can do anything like that. And you will ask me like what kind of skill is good to acquire. I would say number one being able to think about your processes in a step-by-step fashion. Number two, prompt tech. And I dislike prompt engineering because it makes it sound more complex than it is. It's just speaking in English. You're going to have to adopt these systems very fast otherwise someone else is going to do it and they're going to pick you out of the job. Hi Flo, thanks so much for taking out the time. Yeah, thanks for having me.

0:45 Flo runs this company called Lindy which makes it really easy for you to build AI agents from scratch. We'll talk more about that in some time. But my audience are people who are getting to their career, people who are already thinking of starting a business, people who are curious about learning about AI. So to them, if you had to explain from scratch, what is an AI agent? Why is it useful and what are the different applications of it? How do you describe it? Yeah, an AI agent, you can think of it as like Chad GPT except it does stuff. So Teddy today, you can just chat with it and it responds. It's awesome. is really useful but it doesn't really do anything right so an agent imagine if you could talk to and you could ask it like hey I want you to help me with my marketing and I want you to post

1:31 campaigns on Facebook ads for example or I want you to help me create a website or I want you to help me answer my emails and draft my emails and record my meetings and take notes for my meetings that's what an AI agent is so you can almost think of it as like an AI employee that actually does things for you I think today it's sort of like all like an AI intern that does very simple things, but little by little it's gaining more and more in capabilities and eventually it's going to be just like a full AI employee and I think we're pretty close to having like fully autonomous AI organizations that'll just completely run by AI employees. That's a lot for people to take in. But for people who have just used GPD, how is this different from chat GPD? What's the difference between an AI agent and charge GPD? Yeah, an AI agent connects to your apps. It connects to your

2:16 systems and it can use your apps and your systems in order to do stuff. So for example, one of the biggest disease that people had for AI agents like one personal use case is like uh pre-drafting responses to their email. So me every morning I have an agent and it just every time I receive an email it drafts a response for me and so I just open Gmail and I have a response here for me ready for me to review and if it's good I send it. If it's not good I change it. I send it and then the agent notices that I changed the draft that it wrote and it learns from it and so little by little it's getting better and better and better or for example I have an AI agent that sits with me in all my meetings and it takes notes for the meetings. If I mention that there is an action item for someone I'm like hey action item for is to send this email to this customer. It's then going to create

3:02 a task for in like oh task tracker in here. It's going to send reminders. It's going to send the meeting summary to the people who were in the meeting. is going to coach me about the meetings to be like, "Hey, I saw that you said this and this. I thought you could have done that better." Um, so AI agents can use your apps. Yeah, very interesting. That's the main difference between charge and agents. Now, if I want to understand the applications of AI agents, I know you mentioned some of them. Why are AI agents important to businesses and does it mean that people who are watching are going to lose their jobs? Oh, wow. Starting with the tough questions, I think I'll start with the easy one about why do you want it in your business? There's two big reasons. There's the the

3:47 cost savings, but I think most importantly there's like the speed and the scalability. So yes, there's the cost savings which is like look it is cheaper than a human and it's like way faster and way and and most importantly and that's the second thing way more scalable. So in touch with customers one big use case for AI agents is customer support, right? So you you got your customers got your AI agent and your AI agents watch your Zenesk or Freshesk intercom whatever customer support inbox you use. It monitors this inbox and it responds to support tickets and in order to do that it leverages your own knowledge base your own documentation whether that's on Google Docs or notion or whatever and it perhaps issues refunds when the customers ask for a refund. It like tells them about their a status if you're managing like an e-commerce business and so on and so

4:33 forth. Again, it is cheaper than a human. The main reason why our customers use our Lindy AI agents for uh the customer support is because otherwise it's a nightmare. You need to hire you need to continuously hire customer support people as you scared. So it's just a lot of complexity to like you've got to find good customer support people. You've got to interview them. You've got to hire them. You've got to train them and on board them. You've got to manage them. Very often they don't do a good job. You've got to coach them if they don't do a good job. You don't fire them if they don't do a good job. And then we end up with like some big companies that have hundreds and hundreds of customer people, customer support people. So now you need like all sorts of like systems and processes have like layers of management and it's just a headache, right? And so AI agents let you just build your one AI agent for

5:21 customer support. You build it once and every so often you paid it, but you know that it's going to scale infinitively. It's like it's like the cloud. It's like a just cloud for knowledge work and it's going to scale elastically with your demand. If you get a million support tickets tomorrow because you're blowing up is going to be able to handle all million support tickets. And if you get zero, it's not going to cost you anything. Yeah. Very interesting. Now, for people who are watching this and want to start building AI agents and using them, what are the most important skill sets that they would need to learn like in the AI later world? What are the most important skill sets that they need to have to thrive and do well, you know? I would say it's really hard to isolate

6:06 it to a single skill set. I would just call it creating and managing AI agents. And it's it's kind of like if you said like, oh, what's the one skill set I've got to acquire to be a good tennis player? And it's just like there's like a thousand things. It's just like a lot of things. And so what I would tell people is just like play tennis. Just play tennis. If you want to be a good it's like you, you know, like you you're like a great content reader. You've done thousands of videos. There's no secret. Like if you ask people what what's the secret of success to make a video, it's like look, you can give me five points or 10 or 20, but really you've just got to do it. if there do it. And so what I would tell people and this is not to be trained to like sell my stuff by the way like whether it's Lindy or any other platform. I think we're the best one. Uh you you you you got to do it. You got to just like spend your life uh using using these systems. And so to go back to your

6:51 question about like all people are going to lose their jobs long term. Yes. Like longterm like look AGI is coming as trouble to find a reason why humans should keep working. To me it's the same thing as like look we got tractors now so we don't have really farmers well except in like poor countries but like in developed countries we don't have like farmers like picking things with our bare hands or like working the years like machines to do that like why would humans do this I think that the same thing is going to happen to knowledge work physical work got away mostly knowledge work is the next one and like it's just going to be so much better so much faster so much cheaper smarter just like better in every way that why would humans work over the long term. I I simply don't see a compelling reason why they should work. And I I think that's desirable. I

7:38 think that's the outcome we want. Like like jobs are a mean to an end. You know, if we can have what we want is the products and the food and the services and the housing. If we can have that without requiring people to work, why would we work? You know, so very long term, yes. Uh it is coming for everyone's job, including mine, including yours. Um uh short and medium term, um I do think it's going to be distracted. I don't think it's going to cost people's jobs. And but just to characterize what I mean by short and medium deal, I mean five to 10 years. Um and and so yes, it's going to be deeply disruptive and you all going to have to adopt these systems very fast otherwise someone else is going to do it and they're going to pick you out of a job. So uh you've got a short medium term your job is going to

8:23 be transformed from doing the work to managing the agents that do the work. M yeah we'll we'll do a demo of how to build agents as well. But if you are to talk about the basics of what it requires to build an AI agent what would it be if if someone wants to build the AI agent today? What would they need to get started and what is happening behind the scenes inside a AI agent? Yeah, I think step one is to be really clear about what you want to achieve and if you can think of the work you want to done and if you if and if you can articulate it as a sequence of steps and branches it's almost it's it's actually sort of similar to programming right it's just a programming in natural language right it's like step one for example uh we have an AI agent that

9:11 receives support tickets so step one you receive a support ticket on top Step two, you check the knowledge base on notion. Step three, did you find the answer to the question in the knowledge base? If so, respond. If not, perhaps ping me on Slack. Tell me, hey, I I can't find I can't find the answer. Someone's reaching out to me. I don't know what to do. You know, if that's the case, now you're that second branch. If I give you the answer, send that insult to the customer and then update your own noted phase on notion. So that way, you don't ask me the same question twice. So like that's how you make these agents continue. I forgot to up the second question. Uh it was what's happening behind the scenes. Oh yeah, behind the scenes. What's happening behind the scenes is it's an LLM

9:58 on rails. So basically what you do is that you define the rails and the boundaries of the LLM. So for example like the the the the flow I just mentioned, I just said step one intercom message received. Step two, check the knowledge base on notion. That's just always going to happen in this sequence. You're like, I'm not asking you AI agent whether to check the knowledge base. I'm telling you every time this is happening, right? Step three, did you find the answer in the knowledge base? Yes or no? That's all you get to say as an agent. That's your rail. Yes or no? Yes. Answer to the support ticket. And here your rail is like answer to the support ticket. All you have to decide is what you answer based on the knowledge base that you defined. Do you do you have more examples of of EI

10:44 agents which are like highly popular? Yeah, sales is a really big one. So, for example, uh lead outreach, lead qualification. So, hey, a lead a prospective customer is submitting a form on on my website. They want to hear more about my product or my service. Um what you can do is you can have an AI agent perform research to this person online. So literally go on Google and look up their website, look up their LinkedIn, look look up all sorts of information about them and then um what you have found is that information decide whether to engage with them. So that's another branch and if you do decide to engage based on what your criteria criteria you uh engage with them in a personalized fashion. So here you give the agent an email template like hey this is the sort of email I like to send

11:30 but you're going to have to personalize the email based on the information that you found about this person online. Uh that's that's another pretty popular agent. Amazing. Now there are two types of people that are watching this video. One are people who are developers and other are who are not really tech oriented people. How can someone build an AI agent for both types of people? ones who can code and ones who cannot. There's there's different platforms out there. Some of them are more catered to developers and some of them like ours really catering to like everyone like a no code platform that helps you build your AI agent. I would say people get intimidated. They're like ah AI agent it sounds so complicated. It's actually gotten quite simple. You know if you can use air tableable or notion or bubble or

12:15 even like Google sheets you can figure out how to build an AI agent. It's like everything. It's like at first you're going to struggle. You're going to be like how does this work? But after literally just a couple of hours, you're going to get the hang of it and you're going to be able to create really good agents. How do you uh get unstuck like if if you are stuck with a problem, what do you do to move fast? Because most people give up. Totally. Uh well, I think step one, you just try a bunch of crap. You just like ideally you have a mental model for what's going on, but like worst days are just stumbling in the dark and you try a bunch of things until you can unlock that that is surprisingly effective honestly. Um, another thing you can do is you can talk to an AI agent to help you. Uh, we have an AI agent in our product. Like there's like a little bubble in the corner. So you can ask for help to an agent. You can ask JTP, you

13:00 can do anything like that. And then there's like communities. So like we have like a Slack community where people are like asking questions and sending like helping each other and so forth. Oh, you can also like watch tutorials. There's like a lot of tutorials online or how to create a there's a lot of ways to to teach yourself and get unstuck. Amazing. Let's get on to Lindy and try to make our simple AI agent from scratch. That's right. We are on Lindy AI. Yeah. Okay. Take us through the interface. Tell us how to how do we get started with building a first AI agent? Yeah. First of all, you click uh try for free here. There is uh the sign up form here. Do you want to sign up with your Google account? Yep. Let's do it. It always starts by setting you up with a meeting note taker because it's just like a universally useful agent that like everybody can benefit from. Everybody's got meetings. It just sucks to be able to to have to uh take notes

13:47 during your meetings. So, it's asking for your calendar permissions and your email permissions in order to be able to uh take notes during your meetings, join your meetings, and then sends the summaries by email after the meeting. So, it already knows what agents I want to build. It always starts with this one default agent, but then on top of that, you can build as many agents as you want. Mhm. Okay, let's go for it. This is the first agent that I'm building without writing a single line of code. That's right. Lindy can schedule your meetings too. So this is another meeting. This is the meeting scheduleuler. So you can just like CC Lindy to your emails and get her to schedule your meeting just like a human assistant would. Oh yeah. So basically if someone sends me a meeting saying that let's meet at 9 900 p.m. So I can

14:35 add this email. That's correct. CC. So you basically just treat Lindy like a human assistant. So like Lindy can help us find time. So if the person offered the time, she can just like send the calendar invite. If they didn't, Lindy can like look at your calendar. She can offer times. You can go back and forth and so forth. Very cool. That's it. You're in the dashboard. Yes. Uh what do we have here? So here you've got the Lindesies that we installed for you. So, we start with a meeting noteaker, a meeting prep assistant that every day is going to look at your calendar and prepare you for your meetings by researching people who are on your calendar. And a meeting scheduleuler that's like the one we just discussed. I'm going to go ahead and and dismiss this. Um um do you want to go ahead and create an agent? Yeah, let's do it. Awesome. Um what agent can we create? Do you have

15:22 any any thing you're looking to automate in your business? So one thing we do all the time is sales lead generation then sending out the emails cold outreach stuff like that. I think lead outreacher would be a good one. Yeah. Uh let's let's do that then. Let's create a lead generation plus outreach allin-one. So basically and again this is this thing I just mentioned about if you can think about your system and your process in a step-by-step way then you you're like 80% of the way there. And so here the step-by-step process is going to be like, I'm going to tell this sales agent what kind of leads I'm looking for. And then it's going to look for these leads. It's going to ask me which ones look good, which ones I wanted to reach out to. And then I'm going to tell it and then it's going to reach out to all of these leads. So that's the that's the

16:08 step-by-step process. So I'm going to start from scratch and the trigger. So the trigger is the thing that makes the Lindy start to work. So in this case, it's just me going to this agent, we call them lindies, and asking it to generate some leads for me. So I'm going to be like chat with this Lindy message received. Um, and here like the greeting message is the what the lady is asking me when when we start a conversation. So here it's like, hi, I'm a sales agent. Tell me what kind of leads you are looking for. You can tell me job titles, previous companies, company sizes, localities, etc. And this is the message that the agent

16:53 will send me when I open the agent. That's correct. Perfect. And once I've I've given it uh this information, we're going to perform an action. And we have an action that's called uh search for people. So search for people right here. And how does it work exactly? So, it's using a prospecting API. It's like a sales prospecting API to find people. Um, it's a bit expensive. We're working on making it cheaper. For the record, we are realizing no margin on this. This is actually the cost of the provider. So, this is like 40 credits per lead that it finds. Oh, so it is a bit pricey. There are ways to make it cheaper, but for now, this is like by far the simplest way. And have you tried the Apollo API? Yeah, that's that's one of the things we're looking into. Yeah. Um, in our

17:39 experience, this gives you better results than the Apollo API, but it is more expensive. So, and and this is the the magic of it is so anytime you create an action in an AI agent, you've got these fields that the action is expecting and the fields can be in one of three settings. There's either you can set the field manually, okay? Or you can prompt the AI. So, you give it a prompt for the AI to know what to put in this field, or you can just let the AI figure it out completely autonomously. No prompt. In this case, I'm just going to like the search query. I'm gonna be auto because it saw the message I sent it, right? And it's like I I don't need to tell you what to do. You know what to do, right? And here there's a limit and this is the number of of leads that I want. I'm going to set this manually to five for now. Oh, let let's just go with a

18:24 prompt. So, I'm going to be like um default to five if I didn't ask if I didn't specify a number. Right. Right. But otherwise, I can specify a number. Okay. So now it's going to find this and now now I'm going to enter another step in the agent and I'm going to be like show me the list of people you found and ask me which ones I want to reach out to. Okay. And again the beauty of it is and this is what I mean by it's so easy like I'm literally talking to it in English. So an AI agent here is a node that keeps looping. This is why there's that blue arrow like that's rounding around it until what we call an exit condition is true. So it's almost like you're on the highway and there's like an off-ramp, right? So you just keep circling the

19:10 highway and then at some point you take the off ramp. Okay, in this case the offramp is I'm going to and and and it can execute any skills that it wants while it's on the highway. So here one of the skill I'm going to give it is I'm going to give it the search for people skill again. And the reason I'm doing this is because suppose that it didn't find the right people. So I'm going to be like no this is not this is not what I want. Keep finding people until you find the right people. That's correct. Until the until is the key word. And now the until this exit condition here is like exit once. The default says until the task is complete. Right. But actually it should just say until exit once. I have told you uh to go ahead and reach out to people. Okay. Got it. And now what what this agent is going to do is it's going to loop through the people I've asked it to

19:55 reach out to. So, I'm going to do a prompt here and I'm going to see it's like items to look through. So, I'm going to go like the people I have asked you to reach out to and I'm going to go ahead and um you know perhaps just do like a perplexity search. Yeah. About this person. So now I'm inside the loop. This is this like purple box here. And this loop basically is going to execute once for each lead that we found. H. So for each lead is going to search perplexity and I'm going to be like search this search information about this lead on perplexity. And now what I'm going to do is I'm going to make it send an email to this lead. So uh we have a Gmail send action send email. Do you have an

20:42 example of like a template or something you want to reach out about? Yeah. We can get that from my mail itself. I was reaching out to people for for this trip of mine. This is what we normally send. Hey Jeff. Oh, I love this. I love this. Um, okay. So, I'm going to put that in a prompt here. Uhhuh. And and that name could be a placeholder. That is correct. So, here. And because it's a prompt, people ask me what I do is I use like curly braces. Mhm. You know, but you you're basically at this point you're talking to cloud, right? So, it's a really advanced AI like it's going to understand what you mean here. You can just like you can do anything you want. was the definition of what you said like basically it's a LM which has guardrias. That's exactly right. That's exactly right. So okay so I'm visiting SF um I

21:31 create content around AI uh tech and business and I love to collaborate with MB robotics I came across. So see here I can see like company name uh and found it to be really awesome. And here I can be like feel free to slightly edit this to make it a bit more personalized based on what you found. that this person again it's just it's just English it's literally and by the way supports every other language because if you try to talk to cloud in Hindi or whatever it's going to work um so I'd love to collaborate with uh company name to help it create content or promote it on my socials I would love to catch up in person to how we can collaborate boom boom here there's a signature we defaulted to send the L&D but let's let's just remove it um and what I'm

22:18 going to do so people often ask like oh my god it's so because it's going to send an email for me. How do I trust it and all of that stuff which I think is a good is a good uh question. So what we do is in one click you can make it you can make sure that it asks for confirmation before doing anything. So here boom it's just it won't be able this is one guardrail. You won't be able to send the email without your confirmation. So what is going to do is it will show you the email. You're going to have a chance to change it and then send it. Um, and then after the email is sent, it's going to exit the loop. And then in the end, I'm going to enter another. And I'm going to be like, at the end of the loop, once it's done all of this, I'm going to be like, tell me what you did. Got it. That's it. Summarize the thing that you did. Exactly. Uh, we created an agent just now. That's it. Uh, I'm going to I'm

23:04 going to name it, uh, sales agent. And we find out the email automatically of that person. I'm sorry. It will find out the email of that person automatically. That's correct. M um the sky is the limit. Like you can do so much more here. You could send it reminders. You could wait if the person replies. You could like then CC your scheduling automatically. Uh you can send follow up and all. That's right. You can do everything like this is just like a pretty simple AI agent like it took me two minutes to create. Um let's try it. So I'm going to turn it on and let's go to tasks. So here this is the greeting message I defined. Do you have an idea for this kind of person you're looking for? Yeah, let's let's reach out to some uh I want to reach out to the latest YC batch

23:49 companies who are into robotics. Let's say for example, I don't think you'll be able to find the latest YC batch companies because of the prospecting API we're using. There is a way to do this too. Mhm. Uh but let's say I want to reach out to maybe 2023 We can try we can try that, right? We can try that. We're into let's let's say something into marketing for example. Yeah, let's try it. So, usually it the way it works best is if you give it like a title. So, like a CEO of a company like a software company with less than 100 people in San Francisco, right? Yeah. Yeah, let's say I want to reach out to CMO uh

24:34 CMOS of companies with less than 50 employees and marketing. Yeah, you could be like CMO or head of marketing or director of growth. You could even like add in parenthesis like come up with other titles if you want, you know. H perfect. Let's do this. Let's do it. I think the YC 2022 may be what trips it up. So here, this is the rail. It's like the agent is not even deciding. It just knows it's got to search for people. It has found some people.

25:20 So it found these people and now it's asking you which I did not specify SF as the location. You did not specify SF. Did you specify there? I I did not. I think I All right. Well, this is why we gave it the skill. You can just do it again. Just be like, "Oh, no. I meant SF." Keep the location to SF. Yeah, let's try this. Let's go with that. You can dismiss this. One sec. Let me take a picture close. Oh, so this is the first node which is search for people. We've not yet started the next AI agent one which again no we are currently in the AI agent okay because it already found people and now it's like using its scale to do other searches

26:06 and now we have all the companies that are in SF who are into marketing from the YC23 badge that's right very cool so now you can be like yeah reach out to all of them perfect reach out to all of them and this is all run by Claude Sonnet. This is cloud set. Why why not uh Gemini or or photo? You can decide. So um when you go here and when you set up your agent any step you can customize which agent which model powers your agent. We support GI 03 GP4.1 GPT4 everything. I really like cloud. That's all default because it's just really powerful and really cheap. Mhm. But yeah, so you see here what it's doing is it's looping over the agent and basically there's

26:52 like three sub agents that were spun up and it searched perplexity for all of those and now it wrote this email for all of those. So it says you see um I came across next patient and I was really impressed by how your patient experience through solutions. You know, you're going to have to prompt it a little bit better. I don't know about you, like I don't really like how LLM's right sometimes. So like prompt it a little bit better or explain. Yeah. You know, it's like oh let's let's delve, you know. Um, delve. Um, delve is the word. So, you know, you can customize it. And this is the part where it's asking you for confirmation and then once you like it, you can just send the email. Very cool. I think maybe we can just let's just go with it. Go with it. Let's do it. Yep. Let's say send this. And it found the email. Now it is sending it out the

27:39 email itself. And basically, you can do this for each of the three people that it found right now. Oh, but if I would have turned that off, it would automatically send all the emails. I'm sorry. If I would have turned that off, the the thing which says ask me before SS. That's correct. I actually have another agent that's almost identical to this one that I use for recruiting. And so I've turned it off because what I do is I reach out to like 200 people at a time because people recruiting a lot. And so I just be like, "Hey, I want 200 software engineers in San Francisco or Palo Alto or whatever. Send." and then just does everything and it finds them, sends 200 emails, send reminders, if they say yes, reaches out and like schedules the interviews. It's basically like a fully autonomous AI recruiter. Damn. Yeah. Okay, there you

28:25 go. We've now just sent two emails. And so if you go back to the parent task here, try reloading the page. You can go back to the parent task and just reload the page. I think it's just it's still loading. Once it's done, basically it's going to tell you what it did. Yeah, I think it's taking this is the update. Yeah. So, it's like I it's basically this is part of the loop itself. Yeah. [Music] I can open this up as well. That's right. So, now you're inside the loop. M. So that's what I mean like by it's pretty easy to create the agents like look I'm an expert obviously because I

29:11 built the platform but like once you're good and you can become this good in like half a day you can create AI agents like this in like it took us what five minutes something like that right and then the sky is the limit and just keep building and building and building on your agents. Yeah. Show me some more examples. Yeah. Templates. Um well the the meeting prep assistant template uh is is interesting. Um, so here it's it's actually right here. Um, when a calendar event is starting, it is checking if it's an external meeting. Again, it's just it's just English. It's like if the meeting includes external attendees, then you go into flexity, you research the attendees, you find the profile of the LinkedIn URL and then you check your email history with this person and then you send me an email with the meeting

29:57 prep. Got it. Yeah. And and basically people can just save this and start using it in their workflow. Yeah. Exactly. This is amazing. Yeah. Yeah. Um other templates, I really like the meeting not ticker. It's just like a template that I use all the time. Mh. Uh it's in all my meetings all day. And at a high level, it's really simple. It's like, hey, when a calendar event is be is starting. Uh if it's a virtual meeting with a meeting link, you're going to record the meeting. You're going to join it. And then you're going to email me your notes after the meeting. This is basic logic that you're writing. Let's go. It's so simple. Yeah. Right. And this is what I mean by people let themselves be intimidated like oh my god AI agents there's like this kind of platform out there now that makes it super super super simple. This is actually very cool. I've tried other automation platforms but all of them

30:43 require you to like you know enter some JSON or like know some background of like REST APIs and things like that for it to actually work. But this is just plain English 100%. How did you uh decide that you want to go this route? Lots of iteration. lots of iteration and as you can tell it's still like not perfect like it is actually going to be a lot better. One thing I'm particularly excited about is we are building an agent that builds agents for you. So it's like cursor for agents. So like you won't even have to do this work. You're just going to have like a little copilot on the side and you're like hey I want a sales agent that does XY and Z and it's going to be like okay boom boom boom boom boom boom boom sets up the agent for you and then runs the test. Super cool. Yeah. What else can we try out with Lindy? Let's see. Uh do you have

31:29 any other thing that you're eager to automate? Do you have any supporting or do you have any like in I assume you've got a lot of like prospective sponsors that reach out to you? Yeah. Yeah. Yeah. So that's that's one big use case like people who reach out to people who use Mindy to triage their email inbox. So uh we are going to create so again it always starts with the trigger. You ask yourself how does the Mindy know when to wake up and in this case it's just I'm going to receive an email right? So, email received um and filter by only emails with external sending addresses. Right. Right. So, when your team emails you, it shouldn't it shouldn't uh do that. And um what you can do is we're going to have a condition here. So, this is one of these branching paths. If this looks like a brand deal, a what? Oh, yeah. Exactly. If this looks like

32:16 a brand deal, partnership or or partnership offering, right? someone wanting to partner for them to go on live broadcast or something like that, right? Um then you could even add another condition which is like if they look serious. If they look serious. Yeah. If they seem serious agents like they seem serious and look I mean this is where you basically and you were asking me like what kind of skill is good to acquire. I would say number one being able to think about your processes in a step-by-step fashion. Number two, prompt. And I dislike prompt engineering because it makes it sound more complex than it is. It's just speaking in English, right? It's like you just try to explain it very clearly. So like how do you know when someone is serious? Is it a question for you? Like how do you know? I assume like 90% of

33:03 the people who reach out to you are like maybe they would be more in-depth in the company that they're talking about in the work they want to do right in the deliverables that they're asking a media kit something like that. some details about their company like the kind of partnership they have in mind etc. And so if they seem serious then you can uh perhaps respond to the email or perhaps just ping you on Slack. What do you want? Like just Slack. Sheets is great. I think Yeah, let's do sheets. Let's do Sheets. So what we're going to do is we're going to make it log its uh what it what what to do in a spreadsheet here. And so I'm going to go like um partnership opportunities sheet. And here I'm going

33:49 to go like date in format y-m dd. And so basically the agent is going to see the the header here. So if I want the agent to log any more information I can literally just add columns here. So I can be like date I can be like name of person who reached out company details. That's it right details notes. And now I'm going to go back to Lindy and I'm going to be like append a row to a spreadsheet. It's it's requiring your Google Sheets permissions. Yep. And right here it found the spreadsheet we created. I'm going to select the full sheet and I'm literally going to leave this alone. It's just auto the AI. Oh,

34:37 this is the stuff that they make you to write like the other automation platform. I'm literally just leaving this alone. That's it. And then if they don't seem serious, maybe we just archive the email or maybe just label it with something else. I'm gonna go here. I'm gonna go like I don't think there is a label right here. Add label to email. Oh, it's going to it's going to ask for your authorizations. The problem is we always require the strict minimum of authorizations at any given moment. Yeah. So here it's going to need your label permissions. Um, and so here it's like if this doesn't seem like a serious offer, then I'm just going to label it like unserious. That's it. Makes sense. Um, well done. We've got an email triagger. So, email uh partnership

35:23 triasure. Um, I can run a test now. I could email you too if you want. Let's see. Let's see. Or let's just So this is running in real time. This is running in real time. Got it. So what is your email address again? It is this one, right? Uh partnership opportunity. Hi Isan. Uh I'm a big fan of the podcast. I would love to explore partnership opportunities. I am very serious. I'm working for a company called Lindy. or uh willing to explore different avenues to partner. Wow,

36:09 that's speed. This is not Lindy. This is with flow, you know, not affiliated. I just boom, my email just showed up here. Click save. Oh, wow. Wow. And now it's going to run the test. So, you're going to see it go through the steps here. So, it's starting with the email rece. Do you think it'll look serious? Uh, we'll see. Boom. Oh, this doesn't seem like a serious offer. How serious. Oh my god, I'm so insulted. Let's ask what do you think is a serious offer? Uh, well, in order to do that, you would need to enter an AI agent here. Um, I I don't even want to ask it, but this is this is the part where you have to work a little bit on the on the flow, right? It's like it doesn't seem like it's very software. For example, um

36:55 if the person is sending this, what was your Yeah, the problem is the non-serious ones say just partnership or something like that, right? The better ones say something more. Oh, so I'll give you an example. Let's say I'm working with uh Duelingo, my Dualingo partnership. Maybe this is what the dualingo thing looks like. Something like that. Okay. Yeah. Uh but but I think I uh the idea is that this will help me sort out all of my partnership emails. Yeah. Through the through the agent itself. Yep. Very cool. So, and this is the part where

37:43 like it's like an intern. Just imagine you on board someone in your team and you're like, "Hey, if it seems serious, you've got to you've got to tell it a little bit. Maybe like what what is a serious email?" So this is the part where you tell it, right? Um we can also just ask it to reply to it. You could ask it to do that. So instead of append row, we can just say that reply to this email. We can do both. Uh so we can do send a reply. I'm going to make it ask for confirmation. Uh and here you can be like uh say that you're interested. Uh and ask for more information. Yeah. Ask for more deliverables and timeline maybe. Yeah. for maybe with that. That's it. And so, uh, save it. Well, save it. Test it.

38:34 Um, I'm going to reuse the same email. Aha, I was this time. Hey, say it's now serious. This time it thinks I'm serious. Thank you for reaching out. I'm interested in exploring partnership opportunities. Could you share more details about the specific type of partnerships? Confirm. Boom. Let's go. So, this was a reply. So, I can go on my Gmail right now and I should have this as a reply. That's correct. Wow. This is insane. Took two minutes. This is uh literally going to like double my revenue. Well, I I only need 10%. Yeah. Um, yeah, we've actually got a lot of influencers. I can't say who, but there's some extremely big influencers

39:20 that are basically building their business on India. This makes a lot of sense cuz like look at look at this. This is all like deals and deals. Like look at this. Are you doing this yourself or do you have a team doing this for you? I have a team who handles after what I see this. But I have to see this like look at this. This is a brand deal and it's just on and on like that. [Laughter] Um, you know, the crazy thing is, so we have a lot of influencers who use us for this exact Tuesdays and then we have a lot of brands that use Lindy to reach out to these influenc. It is. It is. That's happened to multiple times people. Absolutely. Yeah. Yeah. And I actually think it's good, right? Because it just makes the market more efficient. Yeah. Right. It just means like, hey, influencers just receive a lot of offers and like just

40:06 the right offers go through and it's all agents doing all the middle work. Yeah. Very cool. Now how do you avoid something like that? What just happened? It thought it was not serious then it made it look serious. Uh so is that some hallucination which is happening by by by Claude or or how can you prevent something like that from happening? Yeah, I think it's just models models are smart but they're not like human level smart yet, right? Like I think right here this condition. So that's the beauty of of of this platform is like you can select which model to power each step individually. M so here for example maybe I'm like hey this step always fails I'm going to I'm going to change it and I'm going to make it 01 okay right now I'm going to make it 03 or I'm going to make it GI 2.5 pro that's one

40:52 way you can make it perform better or I actually think cloud is going to succeed here you just need to make it you just need to like tweak the condition a little bit and we can also change this from email receive to just check emails for today or for this week just check all the emails for this week you could and that you could loop through the emails that it Yeah. Yeah, you could do that, too. So, we can replace this. Um, right. So, like here, basically, the trigger would be the timer, right? You want it to wake up every week or something. Yeah. Like, let's say every morning. Every morning. And just see the last 24-hour emails. Totally. So, um, here the trigger is going to be every morning. Yes. So, there is, uh, what we call there is a a timer trigger. So, on recurring schedule, and we can just say

41:39 every day at 9:00 a.m. And I'm going to make it search through emails. Uh, we're going to run into a platform limitation. Uh, but we're going to make it, which is we're going to need to take all of this here and make it inside the loop. Oh, today there is no way to move nodes to inside the loop which makes sense. It's really annoying. But then we can we can get get rid of the not serious one. Yeah, right. I actually think I think there's a hack. I think if you select a node and then I click here. No, you can just drag all of this and drop it inside. No, that's that's the platform. Look at it. So, oh wait, I think there's a workar around. I think if I do this Oh my god, look how

42:25 buggy it is. Um, basically what I got to do is this. And I'm going to I'm going to ask you to Oh my god, it worked. Oh wow. Yeah. No, no human is going to put on this. So this is Yeah, ideally you would want to be able to just drag and drop and we're going to have this soon. Can't believe it worked. Um, okay. So, uh, um, I am going to, you know, we don't even need this first condition because what we can do, I'm just going to delete this condition. And here I can just add a prompt and I can be like um uh the emails for the last 24 hours that look like they are offers. They're like proposals for proposals for partnership partnership or brand deal.

43:10 And so now it's going to search for email. And so this is the search for emails action takes a lot of parameter but I'm just going to like leave all of this empty. I'm going to set this manually. It's going to be empty. It's going to be empty. It's going to be empty. And before I just like today, like right now, and after 24 hours ago, right? So like this this is like the last 24 hours. Yeah. And again, this is what I mean by like if you can think in this way, it's almost it's almost thinking like a programmer. You're not coding, but you're really like building this logic. Pseudo codeord. That's right. It's pseudo code. That's that's the right way to put it. Um okay, I think we are in a good spot. And now it's going to do all of this. Let's make it run right right now. 11:30.

43:56 Okay. And change the time. Well, if you click test, it just manually runs the schedule. So, wow. My audience might not understand this, but basically every day I get like tens of dozens of emails for partnerships. So, you see it's found this today. Mhm. And now it's looping. So if I click here, it's going to open the loop. And this doesn't seem like a serious offer. This doesn't seem like a serious offer. This Oh, wow. It just seems all of them are not serious. Well, yeah, because this one is just like, let's connect during my SF visit. Yeah, this one I don't know what it is. Partnership. Oh, wait. This one is mine again. It just keeps thinking I'm not

44:42 serious. Let's You know what? You know what? Let's what what you can do? Let's make this 01, right? Like, this. Um, let's retry. So, it's going to take it's going to take a lot longer. Yeah. O1 is super slow. But the cool part is you don't have to enter your own API keys and all that stuff handled by Lindy itself. Totally. Um, it's going to take a bit longer. This doesn't seem like a serious offer. This is the same one. This is the one that we did right now. I think I think to its defense like I don't think it

45:28 does seem like a serious software. Like it's kind of No one is going to say I'm very serious. Yeah. Right. In a serious offer. So, um, yeah, if this doesn't seem like a serious offer, for example, the person is sending this from a Gmail account instead of a corporate account, uh, or they just seem to want free coverage or something like that. Like, this is the part where you're just going to have to like um, yeah. All right, let's do a little test. Oh, but the thing is that it's now 12 a.m. So, I don't know if there are any emails. Well, it's just using the last

46:16 24 hours. Okay. Yeah. So, let's see what it's doing. So, that's the part about 01 again. It's just very slow. [Music] This seems serious right here. So you see like the prompting worked here. Ah this is me. My email seems serious. I won. It tried to linked but the agent arrived at this. Oh yes it can't send reply because it used to be able to send a reply because it was just below a received email thing. So now what we have to do is we have to replace this by a Gmail. I think there is a send email. Yeah, I think we just need to like draft

47:01 email. No, I think send email is good. And then we just ask for confirmation. And right here, this is when this is where you sell it. You tell it. Yeah. Um I'm just Yeah, ask for confirmation. Done. And so this is the part where you're like uh tell them you received their email and ask for more details. Oh, what? What else did you say? You said details of deliverables and timelines, deliverables, timelines, etc. All right, let's do one last test. And so this is what I mean by like it's just working through it. You just keep iterating figuring it out by just writing. It's the same as coding. You know, when you code, you're never going to get it right first. You just you just compile, you run it, you refresh your page, you see what happens, and you just rinse and repeat.

47:51 Excuse me. Guess I'm serious. And now it's sending me the email. Perfect. Hi, Flo. That's it. Incredible. I love this. Great. So amazing. So we learned how to make a simple sales agent, email negotiate, email reply agent. Yeah. And we also looked at meeting prep assistant and meeting note taker. Yeah. What else can people do for them to try it out? Like if people are still watching this till the end, what would you want them to try out within the AI? So if you own a business, customer support, sales, like those are easy. Otherwise, like personal assistance is like the very big one. And so here exactly what we cover like the meeting recording, the meeting

48:36 preparation, uh the email triaging is a really big one. I have an agent. I have two other agents I really like. I have one that helps me make better decisions. So what it does is it wakes up every week on Friday and at the end of the week it just pings me and it asks me for decisions I made this week and then it remembers all the decisions I made and it it asks me for like it it it pings me again in six months and 12 months after each decision and it's like flow made this decision a year ago how exactly how did that work and then it confused me ah what did we learn here this didn't go well this went well what can we how can we improve our thinking but you're informing it about every decision taking I I I it sets up a quick message. I send

49:22 it and I don't even need to think about it. It just sends me these messages. I've also connected this agent to my meeting note taker agent. So this agent knows of every meeting I had. So say, hey, I noticed in that meeting you made a decision here to like hire or not hire someone, for example. You know, um so this one I really like. I have an agent to help you manage a personal network. We for the record, I think it's considered a bit cringe to have like the personal CRM. It's like very transactional but it's not for friends you know it's for it's for professional connections how does it work so it it has a Google fees and I can submit a message and I'm like hey adden to my to my uh CRM and I'm like Isen is like an influencer he's based in India okay and then every so often I I go and I send it a message not to ask someone to my CRM

50:08 but to help me find people in my CRM. So I'm like who do I know? For example, if I'm traveling to India, I'm like who do I know in India? And it's going to be like oh is it actually goes even a step further because it also observes my inbox and it watches those condition that's like if I receive a flight confirmation and then look in my CRM for people who live in the place I'm visiting. Wow. And then send me an email. So if I go to New York, I just like book a flight to New York. Soon will do that for me. I'm going to receive a flight transformation and then it's going to be like you're going to New York and I I don't need to do anything. It just sends me an email and then flow I saw you're going to New York. Here's like eight people that you know in New York that you should meet. Oh wow. Do you want me to send him an email to arrive for affiliate? That could be incredible. And I can just respond to the email. I don't really need to open it. I just respond to the emails he sends me and I'm like oh yeah like totally can you help different time

50:53 with like B and John and and Johnny. Yeah. Wow. This is going to save so much time. It is. Yeah. You receive a you will start receiving an email at the end of the week that tells you Lydia has saved you so many hours this week. And it's it's pretty it's an estimate, you know, but like most people when you just get started, you start saving like five or 10 hours a week. I save like 50 hours a week. It's like I double my output. There's a myth. But then the question then becomes if you're saving all this time, what are you doing in the time that you save? Oh, are you working even more or I am working harder? I am working harder than ever. Which is why I was answering your question. You were like, is it going to replace jobs? I actually think people are going to become more busy. Um because then they become the bottleneck of a very productive asset. Uh but over the very

51:39 long term, yes, it is coming for all jobs. Amazing. No, thank you so much. Thank you so much for having me. It was incredible. In this entire video, we started with understanding the basics of AI agents. How do AI agents work? What are the applications of it? We then went hands-on in Lindi looking at exactly how can you build AI agents from scratch without knowing anything about coding by just typing it out. And the most important skill that you can learn today is the skill to write and think clearly and converting that into the flows that we saw today. The flow editor flow. Amazing. Thank you so much for watching us till the end. The link to Lindy will be in the description of the video. Go have a look at it. Try it out yourself and we'll be back again with some another video in the case. We'll then book a flight. Sounds good. Lindy.

52:26 Amazing. Thank you so much. Thank you.