# How to Build Your Entire AI Workforce in One Afternoon (Live Demo)

Channel: Greg Isenberg
Video: https://www.youtube.com/watch?v=oulVKbk0umo
Duration: 30 min
Language: English (United States)
Words: 5435
Transcript page: https://viewrankai.com/tools/youtube-transcript/oulVKbk0umo

---

[0:00] I just saw the future of AI agents and I'm not even overselling it. I had the founder of Lindy AI show us new stuff around how to use plain English to do vibe marketing, to build out AI agents that make you money and make you more productive. A lot of people have promised this. A lot of people have promised AI agent platforms that do this, but it is finally here and this opens the door for fully autonomous companies that are run by agents. And that was science fiction, you know, literally a few weeks ago. It's not anymore. I can't wait for you to see how the new Lindy AI

[0:45] works and how it uses agent swarms, something called computer use. And it's going to change the game. So tune in. When it comes to AI agents, Flo, you're probably one of my first calls that I make. And today's an extremely special episode because, Flo, what are we going to learn today? We are going to talk about the two major things we're announcing today, which are the agent builder and computer use. So the agent builder makes it super easy to build the agents and computer

[1:24] use makes the agents basically able to do pretty much anything that you can do on the computer. So I think, we've been talking to you and I, Greg, about the AI agents for a while now, and they've been sort of immature. These two things are a big, big, big step forward with AI agents. Okay, because there's a lot of episodes on AI agents on YouTube and X, including some of mine, where I'm , or me and people are , this is insane, this is insane. Flo, is this actually insane? If people stick to the end of this episode, will they be , my mind is blown? Look, I don't mean to sound too

[1:58] grandiloquent. the tech is new, right? That's the reality of it. I think we're , you know, we are driving as an industry towards what we're talking, what we're calling the Macintosh of AI agents, right? So we've been stuck in the terminal days of personal computers. It's , yes, it's insane. Yes, it's really cool. But , it's really hard to use. And I think that's one of the main concerns people

[2:20] have about AI agents. It's just super hard to use and to create and leverage this new technology. So yes, this is a really, really, really big step forward because now you're going to be able to create really complex AI agents really simply. It's you just talk to it. It's , hey, I want you to do X moving forward and this and that, and you just have a conversation. And it's effectively an AI agent creating an AI agent for you. The two concerns people have about AI agents

[2:45] are number one, too hard, can't figure it out. I'm not an engineer, I don't have the time for that. Number two, it doesn't integrate with my tools. we're basically solving 90% of both problems today. So look, we don't have AGI yet, but yes, it's a really, really big step forward. Okay, you have my attention. I'm intrigued. All right, let me see my screen then. So I'll talk about marketing, I can talk about sales, and I can talk about customer support. Those are three really big use cases. Support is a trade for world. It's just yesterday, I was catching up with a friend And he's spending $12,000 a month on customer support. And this is after optimization. He's got a bunch

[3:30] of VAs in the Philippines. And he's in the process of replacing all of that by Lindy. And until now, he was holding back because he was , yeah, but they do so much. It's more than just answering a ticket. , yes, we've been doing the RAG thing, the retrieval thing. , yes, if you ask a question about a refund policy or a return policy, we can answer that, no problem.

[3:51] But very often, they check your order status. on Shopify or they issue a refund on Stripe. There's all sorts of things that it does. So now with computer use, you can't do it. It's , okay, you receive a support ticket or they're asking for a refund. You insert a human into the loop, probably at that point. You're going to receive notifications every so often. It's , boss, I'm about to issue this refund. Yeah or nay? And then say yay. And it just goes on Stripe or , hey, check the order status on Shopify. It goes on Shopify. It can do all of that stuff for you. So

[4:21] support is just such an easy one. Especially for e-commerce businesses, if you have a little bit of seasonality, it's not just about the money saving. It's just easier to operate a business that's backed by AI agents than a business that's backed by humans. Because yes, AI agents are not perfect, but at least they're consistent. If you've got a need to work once, it is going to work a thousand times. Whereas humans, you never know. And they don't scale. They quit on you. They start slacking off. They start sometimes stealing from you. So it's, you know, AI agents that you can just put it on the tablet and forget about it and then move on to another part of your business to focus

[4:56] on. So support is just a total no-brainer. Sales, we have so many sales agents. So I'll start with a huge caveat. I want to be real. , and I think that's one thing I really love with your podcast, Greg, is , I feel there's so much, again, grandiloquent promises out there that are empty promises. And I feel the AI SDR, frankly, is one of them. if you think you can just spin up an AI SDR agent and click a button and then infinite pipeline is going to come through your door you're out of your mind this is just not how it works it is still a lot of work if you

[5:32] don't have an adbound motion even if you hire the human SDR and you don't have an adbound motion you should expect I think 40% chance that they never succeed and even if they do succeed you should expect a couple of months of heavy iteration and this is not because the human is dumb It's just, it's hard. There's a lot of iteration. But once you have figured it out, yes, AI SDRs can be very effective. My recommendation is that I think an AI SDR is basically supplementing a human SDR. So that's the whole ad bound thing. I think the way of the future is you should hire a person

[6:08] full-time who is basically 10x more effective than they would be if they were alone, but you still need a person full-time to manage your AI SDR agents. and so here with computer use the things that we are unlocking with our AISDRs, man I feel I'm going to get banned from LinkedIn but LinkedIn, LinkedIn DMs so you've got to have a lot of touches so you know we send an email to people, we send them a text message, we can do that if they have opt-in, you've got to collect opt-in, that's a legal thing, but if they have opt-in we make them a phone call, the agents can do that,

[6:42] and then we hit them on LinkedIn, so it's just all of these touches and it all happens inside a single workflow. And LinkedIn actually turns out it's the most effective, at least for now. So outreach is a really big one. Nurturing is another really big one. So we jump on these calls and a good win rate, if you've got a sales team, is 25, 30%. So it means at best you're going to lose only 70% of your deals. But those are actually really good leads. They jumped on the call with you. that have the beginning of a relationship and so forth. And so what we do is when a deal is lost we log it

[7:23] in the spreadsheet And then we have an AI agent that nurtures the lost deals And so what it does is in that spreadsheet, we log the use case that this person had. And then this AI agent observes two categories of events. It observes, did we just win a deal in a similar use case or in a similar industry? So if we're in touch with , I don't know, a B2B SaaS finance company of 50 people. And then we close another company that's also B2B SaaS finance 50 people. the AI agent is going to observe that win. So , , fuck, we won one. We're going to go back to that guy. So it

[7:59] looks at the spreadsheet. It's , , there's one guy here. I'm going to go back to him. And , Bob, how have you been? I was just thinking about you. We just closed another business. It's , most likely they know about them because they might even be competitors, right? So now they'll be the FOMO, right? So that's one. And the other kind of event it observes is the releases. So again, we basically talk to the agent or it observes our change logs and so forth. And it's , hey, okay, we're releasing agent builders and we're releasing computer use. And so we also have that last reason in the thing. So it's , , we couldn't integrate with, we've got a lot of customers in

[8:32] healthcare. That's one reason I'm so excited about computer use because these guys, they have this a healthcare record management systems. And these systems do not offer APIs, Epic and stuff. And if they do, it costs $100,000. you have to go through a year-long review process. It's a nightmare. With computer use, you're basically going through the back door. And so right here, this nurturing agent

[8:53] is going to look , aha, it's , Bob, last time we spoke, you wanted to integrate with your Epic EMR. Computer use just came out. We can do it now. Do you want to book a call? Can we build from scratch a full-on agent showing computer use just from scratch? Yeah. let me think of a use case that we could go after. Let's just do a simple use case that's going to DM people on LinkedIn. I'm going to go here. Before you even do this, 99.9% of people are not DMing people on LinkedIn. 99.9%, including myself. But I know that if I probably reach out to people on LinkedIn, DM, I would

[9:47] probably close more business. I'm just not doing it because in my mind it's cringe and I don't want to think about it. But if I could spin this up and it works, that's a huge unlock for me. So I just want to preface it with that. 100%. And I think that's one thing we hear very often. People are always , agents are coming for our jobs. Sure, maybe over the long term, but in practice right now, what we're seeing is a lot of people who just do stuff they wouldn't do before. , we know we should post on LinkedIn. We don't do it because it's really hard to automate and it's kind of

[10:23] cringe and we don't want to think about it. Okay, so for now, I'm just going to make it manual, but I could be , hey, every time I ping you on Slack or every time someone submits a form or every time someone gets added to this spreadsheet, hit them up on LinkedIn. For now, I'm just going to be , when I send you a message with a LinkedIn profile, I want you to send them a DM on LinkedIn

[10:54] asking if they would be interested in AI agent services. Tailor your DM to the person's profile. and I'm going to be , I want you to use computer use to send a DM on LinkedIn asking if they would be interested in the agent sources. And the reason I'm doing that is because we also have LinkedIn integrations that, this is the ironic part, by the way, people think, and I thought that API integrations would just be the bomb. That's just universally better. Actually, computer use

[11:30] very often works better. So it's , hey, okay. So when you send me a message, it's confirming. It's , yeah, you want me to do that and start a computerization and do all of that stuff. , yep. this is crazy dude you've been building this for a while so you probably are numb to it but the fact that you can just literally talk to it and it's building it live this is the dream oh yeah I agree I mean we have been thank you Greg you're right I've become a little numb to it but yeah I mean this has been what we've been working two wheels for the last three years straight up

[12:04] this has always been the vision you tell it what you want to do and it just does it for you. You shouldn't have to be an engineer. It's just doing it. Dude, you're going to put all the N8N bros out of business. Look, we do have some competitors that I think are unnecessarily complex. All right, it's done. I've created your LinkedIn DM at your agent. When you send a message to the LinkedIn profile URL, and if I inspect the agent here, it's , all right, you're going to send me a message. If the message contains a LinkedIn profile URL, I'm going to start a computer session. And then you

[12:42] are helping to send a personalized LinkedIn DM to a prospect about AI agent services. We are going to redact here because what I need to do is I need to enter, and this is not going to be here in production, but I have a computer ID here that I need to use for it to have my logins. You can see it even named the agent. It's LinkedIn DM outreach agent. I'm going to save this and let's do it. Greg I'm gonna spam you man oh I liked it when I was DMing not getting the DM wait a minute you know the tweet is I didn't realize the leopard eating people's faces would eat my faces

[13:22] exactly okay so the LinkedIn the condition passed it's starting a computer we should probably do something in the meanwhile because again the big problem is it's not a big problem but it's just slow this is gonna be a pain in the butt Yeah, I mean, while it's doing that, we can, I mean, brainstorm other ideas to just give people a sense of what sort of things that they could be creating. Yeah. Oh, actually, it's not that slow. Let me just open the browser and then we can tab away once it starts being boring. All right. We did it. Open phone is my shared phone

[14:01] number. We is my assistant. I still have a human assistant. All right. And so you can take control. And when you've taken control, you can then release control and let the agent know, , hey, I just did something. So here I'm going to be , I just loved it. I feel I'm watching the future, dude. , this is crazy. Yes. It is. And again, it's early. I always want to be careful to, , be real with people. But, , it's working. Yeah. It's interesting because I was so excited about ChatGPT. when it came out, because to me that was a glimpse of the future, but what it was missing was the agent builder the workflow There was only so far it could

[14:46] go Yeah 100 I think being able to plug it to your applications also via APIs and being able to give it to different triggers is really powerful. And then being able to have this state transition. In the case of my ex to LinkedIn, it's the same agent, but it moves from stage to stage and that makes it a lot more reliable. because otherwise, , operator is just, it's surprising. , it's not good enough to do a lot of this stuff, even simple tasks this. But here, you're, , holding its hand ever so slightly. , you can see I'm not telling it click-by-click what to do, but I'm still, , all right, step one, you're going to check my tutorial. Step two, you're

[15:25] going to check on LinkedIn. And if any of these steps is screwing up, I can really pinpoint where it's doing that, , what step of this workflow, and I can update the workflow. And from a prompt perspective, , do you have any advice on how to create optimized prompts for Lindy? None. The agent builder takes care of all of that for you. I mean, you've seen me create this agent. It's not rocket science. I just talk to it I talk to an intern and it just figures it out. And again, that's the beauty of it. It's , oh, okay, it's found your profile. And now it's going to... ,

[16:03] fuck, you don't have DMs. You're in a photo. Let's see if you can figure this out. But this is a good example of what I mean by, , that's the advantage is, , you have this double mode of, , this double loop of, , you can edit the agent and the agent's instructions and you can see the agent operate. And it's sort of ping pong between these two modes. Because if the agent screws up when it operates, you can just go in these instructions and you're , okay, , keep this in mind moving forward. Okay, so now it's looking at your, , profile to, , know how to pitch you. Yeah,

[16:35] I disabled messages. Okay. From random people. I think it's going to become smarter and smarter. Yeah, it's looking for the message button. I think it's just going to give up. But it's interesting to see. , I want to see what happens. And then we can always bug someone else after this. , it's not dumb. Message Greg is right here. Oh, you know why? Because I have LinkedIn recruiter. It's kind of messed up how LinkedIn does that. , you've disabled DMs, but with LinkedIn recruiter, it's fine because I'm paying for it. So he's probably going to ask me to log into a LinkedIn recruiter account. Oh, my God. All right, Greg. Are you interested in the AI services? I might

[17:19] be after this DM, man. And so, again, you saw me create this agent. It took me literally, I mean, we can look at the recording, but two minutes, if that, two to three minutes or something that, it's not going to be perfect first shot. It's working. So that's always something. But you're going to want to iterate on the copy that it writes the DMs for and so forth. So I think for, and this is what you said earlier, I introduced the human in the loop, not just because it's imperfect, but also because there's a dose of subjectivity. And so I think that's where we are with AI agents

[17:52] now. It's we're sort of graduating from the phase of , hey, it screws up, and we're moving towards the phase of , no, it's just subjective. And sometimes it does lack common sense. Wait, dude, this is crazy. Hi, Greg. Just listen to your Startup Ideas podcast. Love how you're always spotting opportunities to, quote unquote, get the juices flowing in the entrepreneurial space. As someone managing a portfolio of Internet companies at Late Checkout, I imagine you're consistently looking for ways to scale operations efficiently across your ventures. As a matter of fact, I am. I've been working with portfolio CEOs who are using AI agents to automate repetitive

[18:28] tasks lead qualification, customer support, and content creation, freeing up their teams to focus on the high-impact work that actually moves the needles. Given your background scaling companies from islands to advising TikTok, you probably see the potential for AI to handle operational heavy lifting while founders focus on strategy and growth. It's true. Would you be interested in a quick chat about how other portfolio companies are leveraging AI agents? Ooh, I'd love that. I'd love to share some specific use cases that might be relevant for your portfolio. That's actually really smart. Dude, are you kidding me? Still shot. Are you kidding me? This is when it's all wealthy, Greg. , you know, being a founder sucks. This is the highlight of , it's you build Kudas

[19:20] products and you're grinding on it for three years. And at some point it clicks. And , look, it's more 1% of the way there. It's in your inbox. Dude, did you hear that? That was, someone just messaged me on LinkedIn. I wonder who it is. And so again, now I can chain it with , I could find your phone number. And I could be , hey, send him a text message to be , Greg, I just hit you up on LinkedIn. And if you opted in, I could also make you a phone call. Right? And now it could, I'm in, there you go. So I can, I could now iterate on this agent to go , hey, if they say they're in,

[20:00] look at my Google Calendar. Offer them times. Send them an invite and so forth. So if you wanted to do that, you would go back to Flow Editor and just prompt it? Is that what you'd do? Basically, yeah. I would be , okay, after that, you're going to wait for an hour or something that. And then you're going to wait for 12 hours and then 48 hours and something that. And every time you wake up after waiting, you check if there was a reply if there was a reply and if they said that in you check my availabilities on my calendar and you send the availabilities and you go back and

[20:37] forth until you found an availability that works so this is why you start really small and then you iterate and that's why I'm saying it's not in five minutes one click you're going to have infinite pipeline but you know you sit down for half a day or so and it's not complicated anymore. It's just a matter of iterating. And in the end, I can show you, in the end you have, I'll show you this example of an agent or this example of an agent. What else do I have? This example is really good. As you iterate, and again, you build your way to wield it, but your agents

[21:13] end up becoming really complex and really, really, really powerful. So you start small. It's really simple to get started. And , this is an agent that they use for recording because that's how I spend half my time. and that's what it does. And this is actually not even that complex of an agent, but little by little, you just keep iterating and iterating and iterating, and you land here. This is one of my favorite agents, my chief of staff agent that just live in this thing all day. So this is one of my more complex ones. This is my CRM manager agent. So this one I just talk to all

[21:44] day, and it logs the stuff in my CRM, it retrieves the stuff from my CRM. So that's where you end up after you've really iterated on your agents. Incredible. Dude, you've done it. You've done it again. I don think I ever had someone on actually no I never had anyone on the podcast where I this is the future I need to invest somehow and I need to get involved So that I think the reason I say that is I've been playing with a lot of these tools, and they've been close, but this is the closest thing I've seen to what I had in my mind, which is prompt it to workflows and agent flows,

[22:32] and then having a computer go and actually go and fulfill that task. So for me, what's going through my mind flow, just to be frank, is how do I take stock of, we have a bunch of companies that we're running, all the tasks that we're doing, how we can productize it, and where is there an unfair advantage and where there's our opportunity to actually be more productive and make more money? Yeah, I always use the Factorio image. I don't know if people, your listeners are familiar with Factorio, but it's this factory simulation video game. But I really do think, and I've been using this image for years, I really do think the business of the future looks Factorio. So

[23:16] you end up thinking of your business as , open a Figma, a FigJam, just whiteboard it out. , okay, my business is basically this pipeline. Every business is basically a pipeline. And then you just take stock of , where am I spending most money? Where am I spending most time? What is the current bottleneck of the business? And let me just overwhelm the bottleneck with the agents. I have a top of funnel problem. I don't struggle to close people once they're in the funnel, but they struggle to get top of funnel. Okay. Marketing agents. I'm going to spin up a bunch of agents that create a ton of content. We'll, we have a VO3 integration that's coming. So you could

[23:52] totally have an agent that's , Hey, I want you to post a hundred videos on TikTok and Instagram every day. And then I'm going to give you feedback about the videos, and then you should learn. You can very much prompt your agent to be , hey, also every day, you could have another agent do that. I want you to look back at your videos, views history, and see what clicks. And then I want you to maintain a Google Doc with your learnings so far that you just keep iterating on. And then the other agent that creates the videos consults this Google Doc every day. So you could totally do

[24:22] that. I was , okay, I've solved my top of funnel problem. Now I have a bottom of funnel problem. People are a pain in the ass to support. I'm just going to overwhelm the bottleneck with AI agents, and I'm going to solve that, and I'm never going to touch a support ticket again. And just keep doing that. People don't understand. And I have been careful during this entire conversation to not be overly grandiloquent. I hope people recognize that. I have been saying for years, the fully autonomous company is coming. And I don't mean that as a grandiloquent, this is coming in 20 years. I'm , no, no, no, no, no. This is coming now. in the next two years, perhaps. I think

[25:00] in the next, as early as 12 months, I do think we're going to see people create fully autonomous agentic companies. I mean, when I see something this, it makes sense, right? So of course there's going to be kinks in the product as you're rolling it out. this is the first day that it's live. But, you know, I think that if you have an idea and you have distribution and you have a plan to get more distribution, you create these vibe marketing agents and LinkedIn DMs and you , what is, what is a business? If you think about it, it's customer acquisition and fulfillment of the product or

[25:44] service. Right. So if you, of course it's going to be autonomous at some point. Because if you could figure out how to create AI agents that are automating the distribution, we've just seen right now, and you can figure out how you can create AI agents and automations to do the fulfillment, then why wouldn't there be an autonomous company? Yeah, the building blocks are here. This is no longer a sci-fi thing. It's , yeah, it's here. So my Lindy recruiter, for example, okay, this is a Lindy recruiter that I use. And so I can show you, I'm , hey, find me, I feel bad for the guys at

[26:31] Zapier. I the guys at Zapier. But I'm , hey, find me 20 software engineers working at Zapier in the US. And it's , all right, I'm going to find the people. And it's , perfect, I found 20 software engineers. Here's the list. And if I look, I can look at their LinkedIn. and moving forward, I'm going to be able to, I created this one before we had computer use. I should update it to now DM

[26:51] them on LinkedIn. But it's , okay, this is a bunch of software engineers at Zapier. Okay, found them. And if you look at the timestamps, okay, this was at 6.30 p.m. Okay. And so it's , all right, I found these guys. Who do you want me to reach out to? Which of these candidates would you me to reach out to? I'm , yeah, all of them. This is a correct list. This is awesome. Do it. And it's , okay, I'm going to do it. It keeps track of its campaigns because sometimes I get banned on Gmail because I just sent too many emails. And so they're , okay. And then we have this feature

[27:18] called agent swarms. So it's , it just pins up a swarm of 20 sub-agents. And if you look here, so this was at 6.34. Okay, so I kicked off the search at 6.31. Three minutes later, I had emails going out. Okay, 20 of them. And the first thing this sub-agent is doing is it's searching if I've reached out to this guy before. So it's , no, you've not reached out. And then it sends an email. and then it wakes up, sends follow-up emails, sends follow-up emails. It just doesn't let it go. So I have the fulfillment part of the equation going. I could totally just also spin up another Lindy that goes on LinkedIn jobs or whatever and looks for companies that are

[27:56] recruiting and sends them a message to be , yo, I saw you were recruiting. I'm an AI agent. Maybe I actually think that would be part of it. It's , I'm an AI agent. I can help you recruit for a tenth of the price. Just tell me, yes. And you don't even have to sign up. Here's your Stripe link. You don't even have to look at the LinkedIn interface. You'll just try Plank. Let me know when you've done it, and then I'll start reaching out to people for you. Dude, thanks for coming on and sharing this with us and giving us the exclusive on this. I appreciate that. I appreciate you being

[28:26] real, honestly. It's just not over-promising. and how do people get started and start building using Lindy? Yeah, you go to lindy.ai slash Greg that entitles you to a couple thousand free credits and yeah, you can just get started immediately. Okay, so if they go to that link, they get a few thousand free credits. So I'll include that in the show notes then so people can go there, check it out. and I'm going to be sharing this with my team and I hope to build some cool stuff yeah love it let me know what you think and by the way people I love having Flo on the show so

[29:16] people in the comment section let me know if you want to have Flo come back on the show to go deeper on some other workflows that we could be building together so let me know in the comment section would really appreciate that and flow. I love Prashan. Thanks, Greg.

---

About this transcript
Read from YouTube's own caption track and laid out by ViewRank AI (https://viewrankai.com).
ViewRank AI finds the videos already beating a creator's own average on Instagram, TikTok and YouTube Shorts, transcribes them from the audio itself in more than 60 languages, and turns what worked into new ideas and scripts.
Free transcript tools, no account needed: https://viewrankai.com/tools
How to read any video this way: https://viewrankai.com/llms.txt