This AI Finds the Idea, Customers, and Does the Work — Transcript

Full transcript

1. 0:00Anything that you do on your computer, your agent  could be doing and in many cases should be doing.
2. 0:05This is someone on my quote unquote payroll that  never stops working and cost 99% less than an
3. 0:10actual human. It's going to an actual browser and  browsing like a human. My hands are off the mouse
4. 0:15keyboard here. You can hook up Telegram, WhatsApp.  Anytime anything's on your mind while you're just
5. 0:20on your phone, you can put your agent to work. We  have probably 20 agents running autonomously, like
6. 0:26fully endto-end. I have an agent that is like my  visual designer and motion designer. This is crazy
7. 0:32to me because it would take me hours. You set it  up, you go to sleep, you wake up, you have leads.
8. 0:37You could do this for any niche. Your agent can  go and find these candidates and actually build
9. 0:41them new digital properties to the point where  you could sell the implementation of that website
10. 0:47specific to them. That becomes your template  that you then stamp out across all your clients.
11. 0:52You could start this for almost nothing. The guy  that I talked to actually did. I knew I asked it
12. 0:55to do this, but I didn't think it actually would.  That's incredible. Boom. That was incredibly easy.
13. 1:04I bet you can't find me one person that wouldn't  want AI to just go out there, an AI agent,
14. 1:09and do stuff for it while they sleep. Well, today  we're talking about how to do that. Cloud's cool.
15. 1:14ChatGBT, awesome. Those are awesome. What if  it went a step further? What if Chat GBT was
16. 1:19in a robot body and could go a step further and do  stuff for you? times a h 100red AI agents. That's
17. 1:24what we're talking about today. How to make money  with them, how to build them, why they're cool,
18. 1:28why they're important, why they're dangerous,  how to guard rail them. And if you're the type
19. 1:31of person that agrees it's better to show than  tell, stick around because at the end of this
20. 1:35I'm actually going to build something with Hyper  Agent for myself. Super cool. Please enjoy. What
21. 1:40is a practical realistic use case for an AI agent  today that you're seeing people use on a daily
22. 1:47basis? Yeah, I'll jump right into one that's super  visual and easy to understand. This will start off
23. 1:53as kind of like a one-time task, like a project  for an agent, and then I want to show you how you
24. 1:58can start to change your thinking and and consider  these agents as persistent 247 employees. Okay,
25. 2:04so this is a run of an agent that has found  local businesses that have great reviews,
26. 2:13but no website or outdated websites. finds their  contact information, learns about their brand,
27. 2:19learns what customers love about them, and then  actually builds them an entirely new website. So,
28. 2:23let me walk you through what the agent actually  did here. So, what you're looking at basically
29. 2:26is all of the thread with the agent here is  on the left. All of the thinking process,
30. 2:31it spins up sub aents to delegate different uh  tasks to break things down, make them happen
31. 2:36in parallel. And then on the right here are all  of the artifacts. So, you can see the agent has
32. 2:41visited some websites, taken some screenshots. It  actually literally opens and runs its own browser
33. 2:47over here on the canvas. So while it's running,  you will see it actually navigating the web. And
34. 2:51then the final deliverable here is in the form  of an all-encompassing web page which is meant
35. 2:57to be for my consumption which contains all  of the business candidates that it found as
36. 3:02well as the websites that it built for them.  So, for example, we've got CS Automotive here,
37. 3:07300 plus Google reviews, fantastic reviews, but  uh no real digital presence, no website. And so,
38. 3:14it has actually based on its understanding of  what they do just from the customer reviews and
39. 3:19the things that people love about it, built them  an entire website here that has a booking system
40. 3:25as well. Oh, man. So, you can come through here,  even get an estimate. These are all just based on
41. 3:30best judgment, best guess of of pricing, right?  But if you were to actually get this in front
42. 3:34of this business, you could make this specific  to them. So the idea here is from one prompt,
43. 3:39if you're watching the video, pause it, screenshot  it. You can do this anywhere. You could do this
44. 3:42for any niche. Your agent can go and find these  candidates and actually build them new digital
45. 3:47properties to the point where you could sell  the implementation of that website specific
46. 3:53to them. So you could put this in an offer and  say, "Hey, we've learned about your business. We
47. 3:57think people probably can't find you the way that  they should be based on all your amazing reviews."
48. 4:01here is an example of the type of website we'd be  able to stand up for you within days. Or there's
49. 4:07even this notion that maybe that website is just  like a free giveaway because of just how easy it
50. 4:13is and how fast it is to do this at scale. Hey,  this is yours. This is yours for for zero cost. So
51. 4:18you see a few different things that I'll call out  there that agents are capable of. One is extensive
52. 4:23deep research all happening in parallel. This  whole run was about maybe 20 to 30 minutes from
53. 4:29the research finding the candidates to actually  building all these unique websites. Browser use,
54. 4:34like I mentioned, being able to actually judge  visual quality kind of qualitatively to say,
55. 4:38"Hey, that that to me." That's right. Yeah. That  looks like kind of an outdated website that could
56. 4:43be better. That's going to be a candidate for the  outreach. Actually managing drafting that outreach
57. 4:48per business. So tool use in there as well. So  it's going to be able to use your email and draft
58. 4:53those messages. is you can approve them or you  can just trust the agent and say go for it. So,
59. 4:58uh really good example of what pretty close to  a oneshot task that you can give an agent and
60. 5:03this skill is on the marketplace for free  as well. If anyone wants to try this for a
61. 5:08certain domain of businesses or in their local,  you should uh you should give it a try. So,
62. 5:13let's say three years ago, I have a an internet  agency that makes websites for local business
63. 5:18owners. Yeah. I'm going to spend my downtime, you  know, browsing through Google Business Profile,
64. 5:22finding businesses with no website, looking  into the reviews. Do they have pictures? Are
65. 5:26they optimized? Oh, here's one. Find their email,  scrape it, cross reference it, email them. I see
66. 5:32you don't have a website. Can I build that for  you? Your response rate is going to be 0.5%,
67. 5:371% maybe. And then you respond, you hop  on a call back and forth, days elapse,
68. 5:42weeks elapse. Then it's like, "All right, what  kind of a website do you want? Let me close it.
69. 5:46Let me invoice you." And now you're saying  like you set this up, which you can just get
70. 5:51the template from you. We don't need to start  from scratch. You set it up, you go to sleep,
71. 5:55you wake up, theoretically you could have leads.  Yeah. So there's a Seinfeld episode where Kramer
72. 6:00takes this car out and he tries to push it as far  as he can go that you know how far it can drive
73. 6:05on empty. Yeah. If you were to get really bold  with this and go further, like send the invoice,
74. 6:10accept the payment, fulfill the service, hop on  a call with them with your product, how like how
75. 6:16far could you push this agent if you really wanted  to be hands-off? And I'm not saying people should.
76. 6:20I'm just curious. Anything that you do on your  computer, your agent could be doing and in many
77. 6:26cases should be doing. If it's more of a chore,  then your agent absolutely should be doing that.
78. 6:31Like if you feel while you're doing some computer  task that your brain is like almost off, you
79. 6:36should never be doing that task again. So actually  running the outreach and watching your inbox 247.
80. 6:42This is another huge one, Chris, where I'll uh  if I go into one of my agents here, I'll just
81. 6:48show you where we can set up tools. So when you go  into tools and integrations, you know, if a lead
82. 6:53comes in overnight, that needs to get a response  immediately, right? So there needs to be an inbox
83. 7:00kind of night owl agent as well where anything  that comes in inbound or earns a reply from your
84. 7:06outbound is immediately getting that that response  and that acknowledgement. We have agents running
85. 7:11uh ads as well autonomously like fully endtoend  generating the creative strategy optimizing the
86. 7:18spend uh doubling down on the winners killing  the losers. So everything related to filling your
87. 7:24pipeline, outbound ads, inbound sales coaching  is is another one. So get your notetaker on your
88. 7:30call, have those notes automatically passed  back to your team of agents so they can give
89. 7:34you feedback on your performance with that client.  How can you improve your close rates? How can you
90. 7:40sharpen your own script and uh and your ability  to actually secure that business? Daily marketing
91. 7:46competitive intel and industry briefs. feed  that into your team of agents as well to to
92. 7:51tune up the outreach that they're doing. Are there  things that are more timely or more relevant they
93. 7:55could be including? I'm also just showing directly  through the web interacting with these agents. But
94. 8:00another big shift is being able to just message  your agents as if they're teammates. So Slack is a
95. 8:06big one for us. You can hook up Telegram, you can  hook up WhatsApp, just like you would communicate
96. 8:10with an employee anytime anything's on your mind  while you're just on your phone. You can put your
97. 8:15agent to work. So, if I click in here to my team  of agents, each of these cards here is a different
98. 8:19agent that has a different job. And that helps  me actually organize. Okay, I only want to give
99. 8:25my agent access to be able to post on X if it it  is the agent that knows my brand voice and that
100. 8:30I'm explicitly told and granted that permission.  I want a separate agent that's actually touching
101. 8:35customer data and financial data. Those two things  better not ever cross. And I can create hard guard
102. 8:41rails by assigning those permissions to different  agents. It looks like with our team internally,
103. 8:46these aren't our literal agents, but we've got  four of us on the growth and marketing team here,
104. 8:52and we have probably 20 agents, each with  fairly broad sets of responsibility. Like,
105. 8:57I'm not talking like, oh, like one little workflow  or approval. You know, I have an agent that is
106. 9:02like my visual designer and motion designer that  creates all the stuff that we post on website or
107. 9:08on X. We have like a CMO agent that orchestrates  to all the other ones, copywriter agent.
108. 9:13you really start to think of these in terms of  full-time jobs. Okay. So, what's an example of a
109. 9:18breadandbut agent like simple but helpful that you  see a ton of your users using and then could you
110. 9:25show me how to build that live? Yes, sir. Okay.  So, the most popular agent on our marketplace,
111. 9:32which is where we have all these different agent  and skill templates, is called Signal. This is
112. 9:36your always on 247 sales agent. So it works  end to end owning the full pipeline. I'll walk
113. 9:43you through some of its key capabilities and then  I'll I'll show everyone how to actually grab this
114. 9:47agent and start setting it up for themselves.  So it is responsible for growing pipeline,
115. 9:51growing our business, getting us more customers.  So that's the explainer on Signal and then I'll
116. 9:54jump into seeing it in action. Yeah, let's do  it. Okay, so this is our thread with Signal.
117. 10:00So the way this is structured is in what we call  live mode. So there's basically one thread that
118. 10:06keeps getting the same prompt on a set cadence.  In this case, I have this set to every hour. And
119. 10:13Signal wrote this prompt, this loop effectively  for itself based on everything that I loosely
120. 10:19instructed it how to do. So this is a good like  general point with our agents as well. I just
121. 10:23describe a job description the same way I would  give an employee a job description. And the agent
122. 10:28is the one that can actually scribe those detailed  instructions for itself to the point where, you
123. 10:34know, there's been a lot of, I think, excitement  over the last several months about, you know,
124. 10:37skills and we have our skills on our marketplace.  I'm excited when a skill does the job. You won't
125. 10:42really find me diving into like the exact wording  in that skill or in the pieces of it because it's
126. 10:47not really for me anymore. It's for the agents to  run themselves. So every hour in this same thread,
127. 10:53this live mode prompt gets kind of reinserted into  that thread and then signal goes to work. So it
128. 10:59does everything that was just in that explainer.  It goes and finds new prospects based on our ICP.
129. 11:04It will build personalized landing pages for  them. So let's take a look at this one. I've
130. 11:08switched my context here. This is a different  type of company that we're kind of representing
131. 11:12the sales outbound function for. This is a real  company uh that I've been working with on their
132. 11:17hyper agent setup. It's called Swiss Solar and  they actually sell self self-cleaning solar panel
133. 11:23technology. Very very cool tech. That's cool. And  so you can imagine the types of prospects they're
134. 11:27looking to sell into are like industrial-grade  solar projects. They have found that their agent
135. 11:33can go out and actually source real information  and run deep research for specific people that
136. 11:40are the project leads on specific solar farms and  industrial projects. And then the idea here is
137. 11:46that this is a personalized landing page just for  that person. So you have you know what in the the
138. 11:53CTA in their case is let's let's discuss a study  or a pilot program. We actually have videos here
139. 11:59that the agent has created here as well. It's  got screenshots of the software. It's got real
140. 12:04images of that project that's all incorporated  into this hero video. Some animated explainers
141. 12:10discussing how the technology works. more shots  of the product UI. So, you can imagine like you
142. 12:17are receiving this very very personal and unique  proposal of a technology you're not familiar with
143. 12:23but has clearly put in the work to understand your  business, the specifics of your project. You know,
144. 12:28we think we have a much better chance getting  uh someone on the on the phone after receiving
145. 12:32a proposal like that than just a plain uh text  kind of cold email. I I feel like this is very
146. 12:37egofriendly. I like to say like if you're if we  were Yeah, I like that. on a live sales call or a
147. 12:42demo and I'm actually interested, but I don't want  to really show it, then I'm probably not gonna ask
148. 12:47as many questions as I want to ask. I don't want  to lead you on. So, I'm like, I don't know. I'll
149. 12:51check it out and then I might or I might not. But  this is very async, right? He can do it on his
150. 12:56own time. He could ask it a bunch of questions,  go through, take his time without the fear of
151. 13:00anyone watching him or anyone pressuring him, you  know? 100%. Yeah. And think about I mean again the
152. 13:05old school way to do this the I mean the classic  copout response from any buyer is like oh send me
153. 13:10some materials. Well now it's like I can send you  some materials. It is something hyper specific and
154. 13:15personalized to you. And hopefully if we do that  actually earlier in the process we stay out of
155. 13:20that trap of just needing to you know send someone  some materials. And so we're we're actually really
156. 13:24earning that buyer's attention and respect by  putting in that deeper work for what is in this
157. 13:30case especially a very high consideration type of  purchase. We're really starting that relationship
158. 13:35out on the right note. So, as I'm just scrolling  through here, you can see that Signal, this agent,
159. 13:39as it's been running in live mode, just  working 247 has been building out these
160. 13:44uh these contacts for real projects, been building  out all these pages, and then it's doing all of
161. 13:49its other kind of routine chores in every run as  well. CRM discipline. It's going to post an update
162. 13:55over to Slack for me as well. So, if I switch  over to Slack, this is what that looks like.
163. 13:59We're big fans of wherever possible, we think  agents should be prompting their humans rather
164. 14:05than humans needing to prompt their agents.  Your agents should just be working. Like your
165. 14:08agent should be working 24/7. They should have  clear guard rails, tool access, and then when
166. 14:14they need your attention or just to update you,  they will let you know, hey, in my 3pm loop,
167. 14:19here's what I did. Added some new prospects, built  out their pages here in the command center. Here
168. 14:24are some things I learned. That's the other really  powerful part. Once you get this wheel in motion
169. 14:30is the constant learning. So for the outreach that  does earn a reply, you're going to see more of
170. 14:35that from Signal. For the tactics that don't work,  you're going to see less of that and is always
171. 14:39going to be passing those learnings back to you.  Hey, here's the type of prospect that actually
172. 14:43seems to be responding better to our offer and the  type of messaging that's earning that click and
173. 14:47that call book. Interesting. Do you have any like  anecdotal case studies of people using your agent
174. 14:53for outbound sales that have like actual wins that  they've reported? Yeah, there's been a couple. Um,
175. 15:00let me show visually what some of it looks like.  We had, for example, a landscaping company that
176. 15:07has been using agents to actually build out their  proposals based on inbound leads in their case.
177. 15:14So, imagine you are looking to get some work done  on your yard. You send in a very short description
178. 15:19of what you're looking done with a picture of your  current yard. Maybe you send that out to three or
179. 15:22four different local companies. One of them gets  back to you, you know, very quickly and actually
180. 15:28has a vision for you of what your backyard could  look like, your actual yard. Like this is where
181. 15:34you started and this is what it could look like.  Meanwhile, the other three companies are like,
182. 15:39"Uh, well, we'll see when we can come on site."  Probably took them one or two days to get back to
183. 15:42you anyways. like what a competitive advantage it  is to not only have an agent that's able to reply
184. 15:47right away and kind of respect that that lead as  soon as it comes in, but actually potentially sell
185. 15:53them a little bit of a vision and not have the  customer have to use their imagination for an
186. 15:56offering like that. That's destined to fail.  We're not creative beings, right? Like we we
187. 16:01shouldn't have to expend brain power into how we  can give you more money. We need to do all that
188. 16:05for you. Take away all that. We got to see Yeah.  People have to see it to believe it. even things
189. 16:10that are, you know, a little bit less visual but  still uh kind of warrant this approach. We have a
190. 16:15paving company that is using satellite imagery of  uh large lots like parking lots that are clearly,
191. 16:23you know, past their prime. Yeah. And showing with  that satellite imagery and AI image generation,
192. 16:29here's exactly where we would repave estimates on  the cost based on what we can see from satellite.
193. 16:34Interesting. and you know the the impact to  the value and kind of like street curb appeal
194. 16:39of your of your business or your commercial  property whatever that might be. So there's
195. 16:43some really cool use cases like that. I'll share  one more was uh a company that does branded like
196. 16:48company apparel essentially. So you know you might  get your your company polos or hats or whatever.
197. 16:54Uh again you can see where this is going. Let's  build that prospect list. Let's actually take
198. 16:59their branding put it on some of our products.  use AI image generation to show, hey, this is what
199. 17:04your team could be looking like when they show up  to the site or show up to the trade show. Here's
200. 17:08some different visual styles, you know, interested  in in understanding more about what we can do. So,
201. 17:13those are some of the the fun ones that are a  little more on the visual side that people have
202. 17:16been doing. What about um this is just a personal  idea I have. So, I I just bought a ranch here in
203. 17:21North Texas and it has this pond that needs to  be dredged. And so, I'm looking for like pond
204. 17:27dredging companies, pond maintenance companies.  And and I had the thought and I prompted Claude
205. 17:32like, are there any companies that's like the Zoom  info for ponds? Like, if I were a pond management
206. 17:38company, could I buy a customer list? Could I get  leads from somewhere? And then I asked Claude,
207. 17:43"How many ponds are there on farms and ranches  in Texas?" 800,000. Okay, 800,000. And I said,
208. 17:50is there a directory? Is there a place where I  could sort through that? It's like, no, that's not
209. 17:54a thing. It's like, well, that that's very niche,  but certainly people would pay money for that. So,
210. 17:59could I use hyper agent to build an agent that  could go take I mean, all this stuff exists,
211. 18:05just satellite data, GIS, mapping, spreadsheets,  and just take a measurement. All right, like this
212. 18:10lake is one acre. The address is this based on the  PIN, and just put it all in like a Google sheet. I
213. 18:15like where you're going. Yeah. Do you think any of  that data would come from like zoning applications
214. 18:19or anything? Or would would straight up satellite  image analysis be the best sour? It could come
215. 18:23from tax records, zoning applications, like right,  let's try. Well, I'll I'll let Hyper Agent run and
216. 18:29we can we can keep chatting and we'll we'll see  what it does in the background. Let me prompt it
217. 18:33here. I just got an idea from my friend Chris  for uh our pond dredging company. Uh we're in
218. 18:38North Texas. you pick the the the town where this  might make sense, but what we want to do is build
219. 18:43out a database of all of the ponds that might need  dredging in the area. Uh we're thinking you might
220. 18:51be using satellite imagery to do that or maybe  zoning applications or other sources of data. Uh
221. 18:57do a little proof of concept sprint. See what you  can find using different approaches to find leads
222. 19:02for our pond dredging business. Oh man, that's  so good. So, we'll let our agent run there and
223. 19:09see what it comes up with. Yeah, it's not the  best use of time, but it is a little bit fun
224. 19:13to just watch them run. Yeah, you get to see their  whole thought process. You get to see them lay out
225. 19:19different approaches, rule them out, whether they  work or don't work. All of its actual um artifacts
226. 19:24will pop up on the canvas here as we go. It's like  Claude in a human form, right? It's like a robot
227. 19:30whose brain is cla or chadypty or grock or all of  the above or whichever one it decides to use with
228. 19:35the ability to go out and do the thing that you're  asking questions about. Yeah, exactly. Yeah. At a
229. 19:40technical level, it's worth mentioning in every  single thread, every agent actually literally
230. 19:45does have its own computer. It has its own virtual  machine where it can do things like run commands
231. 19:50and install libraries. And you're going to  watch it right here open up its own browser. So,
232. 19:54my hands are off the the mouse and keyboard here.  It has spawned what we call sub aents. So, it's
233. 19:59got a couple sub agents here where it's it's sort  of they're running like in parallel then. Yeah,
234. 20:03exactly. So, it's kind of got like the, you  know, it's got the gang together and said,
235. 20:07"All right, here's the overall plan. Agent one, I  need you to check the satellite data. Agent two,
236. 20:12I need you to check the zoning applications."  And they're going to go off and run in parallel
237. 20:16and then report back to uh to the main agent  as they they complete their their work. Okay,
238. 20:21that's a good one. Yeah, we'll see how it does.  I'm curious, what else have you seen from your
239. 20:25your audience when it comes to this concept of  having agents that you really start to treat more
240. 20:31like uh like employees or co-workers? Are people  pretty early in that journey? Are you seeing some
241. 20:35some cool experiments that people have started to  to get real value out of? Very early. I know a lot
242. 20:41of people that have AI implementation businesses,  they'll go into small to mediumsiz businesses and
243. 20:45do an AI audit or just help them install AI  tools. So I am curious how agents could help
244. 20:52them find more customers or help them fulfill  customers they already have. Yeah, this is a
245. 20:57group that we're thinking a ton about and I would  speak to it on two levels. So one is for the AI
246. 21:04automation agency itself for your own internal  operations. use these agents, whether you use
247. 21:10the pre-built templates or come in here and build  AI employees for yourself to find more customers
248. 21:15and also to service your customers, run your own  operations, project management, things like that.
249. 21:19But you could use that exact same flow that I  was sharing where that that same single agent
250. 21:23is running outbound, inbound, nurture, finding  leads within your niche, whether that's kind of
251. 21:29uh physical service businesses or something more  upmarket, whatever that might be. The second thing
252. 21:34is using this as a platform to actually deploy  the agents that you sell. So any agent here that
253. 21:40you create like you might have an agent for your  niche that let's say we stick with this like pond
254. 21:45dredging example that's like the pond dredging  project manager agent and we're going to go sell
255. 21:50this to all the pond dredging companies that  becomes your template that you then stamp out
256. 21:56across all your clients. So we think that our  platform also has a really strong position as
257. 22:01being the way that AI automation agencies actually  choose to sell, package, and deploy those agents
258. 22:08to the end client. And I'll I'll come back to this  agent in a second, but essentially the way that's
259. 22:13structured is we have this concept of teams.  They're kind of like different workspaces. And
260. 22:17so I could have a workspace for just one client  like Apex Landscaping. I could have workspace
261. 22:23for another client. And then I my client can  be in there and actually be the one paying the
262. 22:27bills for the ongoing AI costs, but I'm the one  actually managing and editing those agents that
263. 22:32are inside this workspace just for them. So I  have that backend access. The client can just
264. 22:38use the agents. You could give them edit access  if you want, but for most of those AI agencies,
265. 22:42we hear that they and their clients just want  the client to be an end user of the agent and
266. 22:46any backend changes to that agent, any config  is going to happen by the agency. And so you
267. 22:50can set up that exact model right in here. We  hear that AI implementation agencies are doing
268. 22:55a lot of like very creative and I would say  often like hacky stuff to be able to set up,
269. 23:02standardize and package their offering, deploy  it for clients with like a lot of technical stuff
270. 23:07with different private servers, everything like  that and then actually charging for it is uh is a
271. 23:12bit of a hassle as well. We can simplify a lot of  that with uh with this setup. Yeah. Beautiful. How
272. 23:17does Hyper Agent know what a good website looks  like? It's going to be looking both visually,
273. 23:23and by visually, I just mean literally like it'll  take a screenshot and then process visually what
274. 23:27it's saying. It can also look at the code itself.  So things like the color choices, the layout
275. 23:34choices, the use of negative space. I'm looking  at the landscaping one just because it's what I
276. 23:37had handy, but this is a obviously a page that  that agent has built itself. Code has become the
277. 23:43language of agents now. They're so good at looking  at code and understanding what they're seeing and
278. 23:48understanding how it could be better and how that  code is going to visually present itself. So it's
279. 23:52twofold. It's one is straight up image recognition  like if you were just take a screenshot of this
280. 23:56and say is this a clear layout? Is this good use  of color? Is this a modern you know font family?
281. 24:01Things like that. The second is actually by  inspecting the code. Those are the two ways that
282. 24:05that age can conclusion because they've seen all  the examples and all the references in the world
283. 24:10of every website that's out there and how web  design trends have have changed over time as well.
284. 24:15I mean, put maybe overly simplistically, it could  just Google best looking websites of 2026 and just
285. 24:21find random blogs and train itself on how that  looks. Like, all right, we're seeing a lot of SIF
286. 24:25fonts. We're seeing a lot of pastel colors, choo  pattern matching, pattern matching. Cool. Yeah,
287. 24:30exactly. If someone were total beginner to AI  agents, they've used Cloud, they've used Chad GBT,
288. 24:35never agents, what do you suggest they do from day  one on on your platform? Yeah, I would say pick a
289. 24:41part of your work that you find very tedious that  you would like to not do anymore that is like a
290. 24:48computer job and start with asking hyper agent to  build a dedicated agent for you to do that job.
291. 24:53That should hopefully give you the first taste  of what these agents can do. You can put that
292. 24:57agent into live mode as well like we talked about  with signal. So just every hour it comes through,
293. 25:02it runs its checklist, it does that job. From  there, you can start to get more ambitious. And
294. 25:06the metaphor is less about, oh, what are like  chores and tedious things that I don't want to
295. 25:10do? And it's more like, if I had all the money  and time in the world to make the dream higher
296. 25:16and make that person really successful, who would  I hire right now? And then what would they do 24/7
297. 25:21if they never had to sleep? So, for example,  with Signal, the sales agent, it's like, well,
298. 25:26dream sales hire would be constantly sourcing new  outbound, constantly sharpening its messaging and
299. 25:31its approach to our ICP definitions. It would go  super super deep in its research for every single
300. 25:37prospect and build them something very personal  as a proposal or pilot or initial offer. It would
301. 25:43never let a warm lead go cold. And it would track  everything perfectly in CRM. it would find that
302. 25:48you would get creative with where it finds contact  information. List those things out and then push
303. 25:53your agent to do as many of them as possible and  hopefully we've given a taste today of how far you
304. 25:58can go with that. What's something that you've  seen Hyper Agent do that genuinely surprised
305. 26:03you? Ooh, that's a good question. A lot of the  browser use stuff is really impressive because
306. 26:08it's so humanlike. The way it will actually  just navigate through web interfaces that are
307. 26:13not meant for agents. Like the web at large is  still pretty hostile to agents. You think about
308. 26:18things like even like captas that are just in the  way. They're like trivial for agents to solve now,
309. 26:22but they're still in the way from the era where we  were so concerned about bots. Now the shift that's
310. 26:27happening that's underway is people actually want  to make their websites very agent friendly so that
311. 26:32agents can find them as opposed to only being for  humans. But seeing agents actually browse the web,
312. 26:37visually process what they're seeing. We had our  agent do like a a landing page design tearown.
313. 26:43So if I go back through here, yeah, this landing  page tear down. So it was doing like this would be
314. 26:48for like a web design analysis or something like  that using its browser look at web design trends
315. 26:54and then come back with annotated markups the way  that like a really expert designer would. That
316. 26:59was really impressive to me between the browser  use, the visual understanding and you use that
317. 27:04that word taste which is becoming a hot topic  for sure with any kind of visual work or even
318. 27:09copyrightiting. In my experience, given the right  context and instruction, the models are getting
319. 27:12pretty good. A lot of the video work as well,  Chris, has become really, really exciting. So,
320. 27:18if I quickly go over to our X feed here, let me  show you some of the stuff that I've been able
321. 27:22to do with these agents. So, so let me let this  play and then I'll tell you about why this was so
322. 27:28exciting to me. So, this video that you're looking  at was created by a motion design agent. So,
323. 27:34it's very meta, but I have a motion design agent  creating videos about agents. Historically, for me
324. 27:38to do these demo videos, I have to use the product  as if I'm the customer I'm trying to represent and
325. 27:45then screen record the process, show the agent  generating its work, show a really good outcome,
326. 27:51and then edit that video together. It's very like  handcrafted and artisal. You got to get the timing
327. 27:56right. You got to make sure everything that's  on the screen is right. What's happening here is
328. 28:00this agent is actually using its own browser.  It logs into hyper agent and it says, "Hey,
329. 28:05Alex has asked me to create a demo of an agent  monitoring its inbox. It will actually do that,
330. 28:11understand how that would look as a real run  and then through code actually just create
331. 28:16that whole visual experience and animate it and  get all the timing right. It can also then take
332. 28:20creative liberties where it has like these popup  messages from the agent, a popup message from
333. 28:25the human user that are there because it helps  clarify the product story. This is crazy to me
334. 28:30because it would take me hours to create a screen  recording of this quality and there's even things
335. 28:36I wouldn't be able to do here where I'd have to  like awkwardly punch in on the text. Now the agent
336. 28:40can just have the whole storyboard of this whole  product animation exactly the way that it wants.
337. 28:45The creative possibilities that this unlocked have  been pretty amazing for me as a marketer. Okay,
338. 28:49first of all, that's wild. Second of all, how how  much babysitting is required? like with other vibe
339. 28:55vibe coding tools or prompting like it's loading  it's loading you're checking in you're fixing
340. 28:59you're fixing the last mile is really really hard  what would you say to that I would call it less
341. 29:04about babysitting it's not so much that I have to  watch it during runtime but there are definitely
342. 29:09a lot of iteration cycles so to get this motion  design agent for each of its runs I expect to
343. 29:16have to probably give it three or four rounds of  feedback so the the rhythm of that is I start the
344. 29:20thread I let it run it usually takes about 30 to  40 minutes because again it's using its browser to
345. 29:25actually study the product. It's actually coding  up a recreation simulation of the product. It's
346. 29:30thinking through the whole storyboard and then  rendering the video. It then pings me when it's
347. 29:33ready. There's probably things related to the  timing or the visuals or the sequence that I
348. 29:37want to see improved. Maybe some copy that is put  in the video is off. It runs again. That run's
349. 29:42going to be much shorter. Maybe 5 10 minutes. And  then maybe we have to do that a couple times. So
350. 29:46it will get to a serviceable output every time. I  don't have to babysit it in order to get it to a
351. 29:51decent output. But to get it to my standard is  usually going to take a couple tries. But that
352. 29:56time keeps getting lower because during every  run it is learning as it goes. And so it says,
353. 30:02"Okay, Alex has given me that feedback.  I'm not going to forget that." There's
354. 30:06always that perr run feedback and learning that  you're going to have to be prepared for. Again,
355. 30:10just like if you were onboarding an employee,  the difference here is that those learnings will
356. 30:14stick in the form of literal updates to its skills  and its memories. Okay. Speaking of marketplace,
357. 30:19can this browse Facebook marketplace and message  people? It can do anything through the browser. It
358. 30:24can do things more easily and faster when there's  an API. I haven't tried the marketplace example,
359. 30:29but of course you could do that through through  the browser directly. So, it could actually
360. 30:32literally open marketplace, scroll through based  on what you're looking for, try different search
361. 30:37terms, and either buy or sell on your behalf. For  sure. So, another example, you know, I bought this
362. 30:42ranch, so there's a bunch of stuff I need to buy.  I need to buy a side by side, a Bobcat, a tractor,
363. 30:47ton of stuff. So, I'm on Facebook Marketplace all  the time and sorting and talking when can I come
364. 30:52see it? And I have my assistant helping me, but  it's still a pain. So, I'm wondering if I could
365. 30:56go to my Claude project that I already have for  this. It knows exactly the makes and models and
366. 31:01prices that I'm looking for. Just say, "Claude,  give me a markdown file of everything of all this.
367. 31:06Give that to Hyper Agent and say, "Hey, let me log  into my Facebook account. You know what I want?
368. 31:11you know what I'm looking for. Every hour, every  two, if you see something that fits my criteria,
369. 31:15reach out and ask questions. Exactly. Right.  Yeah. I'll show you even step by step how you
370. 31:20would do something like that. So, let's imagine we  had that markdown file. You would literally just
371. 31:23attach it right here as you're getting set up.  And I would just say like create an agent that's
372. 31:26going to watch Facebook Marketplace for sales that  I specify. And we'll just see how this agent gets
373. 31:32spun up. I'll just want you to see the uh the  create agent flow here, which is really, really
374. 31:37simple. So, I've kept this one really general. If  you had like specific products, specific buying
375. 31:42criteria, you would just give all that to the  agent. If you see this little card right here,
376. 31:46we're now going to get what we would call a named  agent. So, a specific agent entity that's going
377. 31:50to have its own name, its own avatar, like all  these ones we've been seeing, signal, funnel,
378. 31:55red line. These are all named agents. So, you can  always use hyper agent just for a general task,
379. 32:02uh, kind of a one-time thing. But exactly what you  started describing there is where we're seeing all
380. 32:06of our customers start to shift. And what they  want is these persistent employee like entities
381. 32:12that actually perform a job on an ongoing basis.  So we've got this agent here called let's see what
382. 32:17it chose to name itself marketplace scout. So  as you see described here this is going to be
383. 32:21a persistent agent browser access listings table.  So it's always going to be remembering what it's
384. 32:26seeing what comes back around. So we can go ahead  and save this agent right here. It's suggesting
385. 32:31that we put the agent into live mode, which is  that feature that I talked about where it's just
386. 32:34gonna wake up every hour, run through its list,  run through its list, and if it sees something
387. 32:39that fits your criteria, you can instruct it on  how to notify you, whether that's through email,
388. 32:43WhatsApp, Telegram, Slack, Discord, whatever.  We'll enable live mode. And then when we go
389. 32:47see that agent, we can see it's given itself the  little shopping cart emoji as an avatar here for
390. 32:53fun. We can uh we can generate one as well. Maybe  friendly robot in a uh noir style. Why not? Let's
391. 32:59see what we get here. And then I'll show you  how we set up those integrations as well. But
392. 33:03that's cool. Yeah, our CEO actually has an agent  that will watch uh vintage sports car listings
393. 33:08for him. Uh which is very similar to what you're  describing. That's a good idea. Beautiful. Well,
394. 33:13how's our pond project doing? Let's see. The uh  the pond dredging lead genen analysis agent is
395. 33:19still working away. Let's see what it's found  so far. You can always uh jump in and just ask
396. 33:24your agent to summarize, but it's been working  hard. you, as you can see, like these runs,
397. 33:28they do get quite long with the more ambitious  tasks. And so, let me just see what it's learned
398. 33:32so far. Just pause work for a second and update  me with a few key bullets on what you've learned
399. 33:36so far. It gives you the whole thought process  here. So, a little out of my depths in terms of
400. 33:41this like geographic data it's looking at, but we  can see what data sources it's been trying. Yeah,
401. 33:46with a little more runtime, what I would expect  to get back from this agent would be basically
402. 33:51a single web page artifact that is kind of  like that command center that I had going for
403. 33:56the solar panel tech company where it's here's  all the leads, here is the buying signal that I
404. 34:01specifically identified, the data sources where  I got it from, and here's exactly the message I
405. 34:06would send to them as a reach out. So here's  what the agent found. Let's have a look here.
406. 34:10Here it picked a specific town in North Texas  looks like based on some population figures.
407. 34:15It's got imagery. It's got a pretty good start  there, it looks like. And then, like I mentioned,
408. 34:20I would just have it keep running and then deliver  its findings basically as a a briefing to me that
409. 34:24uses some of that design and formatting that  we saw from the other agents. Amazing. Well,
410. 34:29thank you, Alex, for your time. Where can we find  you? Where can we find this product? Follow us
411. 34:33on X would be uh the best place to keep up with  what we're doing. Hyper Agent OnX and uh use uh
412. 34:39the link that we'll provide for your audience to  get some initial free credits to get you started.
413. 34:44Give your agents some ambitious jobs. The jobs you  would give, you know, really skilled people that
414. 34:48you were hiring and see what they can do when you  you just let them run. Yeah. Let them help you,
415. 34:53you know, find customers, grow your business, run  operations. There's a ton of potential here. And
416. 34:58we've got some other customer stories you can see  on our socials as well if you want to get inspired
417. 35:01with with what people are doing. Beautiful. Thank  you, Alex. Awesome. Thanks, Chris. All right,
418. 35:06as promised, I'm going to log into Hyper Agent and  build some pretty cool stuff so you can see for
419. 35:11yourself hands-on how this actually works instead  of just talking about it theoretically. Okay,
420. 35:16this is Hyper Agent, which I am an active user  of, and it's built by the team at Air Table,
421. 35:20which I've been an active user of for years. You  give it a job, it goes off, it uses a browser and
422. 35:26real tools and comes back with something that is  actually finished. This thing can run a whole team
423. 35:31of these agents at once. It builds real stuff.  We're talking like entire web pages, dashboards,
424. 35:36not just text or images. There's this one company  that had a single dude build a fleet of 14 agents
425. 35:43that his whole team now uses every day. So,  today I'm going to use it for my favorite thing,
426. 35:48which is finding boring businesses that quietly  make a killing and then actually building one out.
427. 35:53We're going to start right here in the search box  and type in this prompt. Find me 10 boring local
428. 35:58businesses that quietly make over $500,000 a year  and that almost no one talks about. For each one,
429. 36:03tell me what it costs to get started, the  typical profit margins, and why more people
430. 36:07aren't doing it. Put it all on one clean page  that I can read fast. Cool. Okay, let's see
431. 36:12what it does. I'm going to leave it on the model  that it shows by default, which is GPT 5.6. Okay,
432. 36:18let's see what it's doing. Reasoning asking me  a couple questions here. What should make over
433. 36:22$500,000 a year mean? You know what? I'm going  to say owner profit. Which market should startup
434. 36:27costs and economics reflect? United States. Do  you want business models or 10 specific operating
435. 36:32companies? Both. So, if you give it a crappy  prompt, it improves it for you. Okay. I think it's
436. 36:38really important that you actually watch what the  agent does as it's doing it because it's mimicking
437. 36:43human behavior. It's going to an actual browser  in a cloud computer session and browsing like a
438. 36:49human. It's reading documents, searching the web,  portable sanitation. Oh, a little sneak preview
439. 36:54right there. Now, it's suggesting a couple skills  that we can save. This is basically like a skill.
440. 37:00It's a skill that Hyper agent just learned and it  wants to know if I want to save it and recall that
441. 37:05skill at any given point. Kind of like the Matrix,  right? They plug into his brain. He's like, I know
442. 37:09karate now. That's a skill. He doesn't have to use  that skill when he's talking to his grandma, but
443. 37:13if a bunch of bad dudes in suits approach him on  the street, he's going to access that skill that
444. 37:18he had learned previously. Same thing here. We'll  go ahead and save that. Let's go ahead and check
445. 37:24in with it. How's it going? Okay, I went ahead and  approved the plan as presented right here and now
446. 37:29it's cooking. Super cool. Research four boring  US local facility or assetbased business models
447. 37:35capable of exceeding 500,000 and normalize  annual benefit to the owner at established
448. 37:40scale. Multi-tore laundromats, laundry route, self  storage, express car wash, parking lots. Cool. The
449. 37:45initial screen is done. The threshold is genuinely  demanding. Most qualifying operators need several
450. 37:50crews yada yada yada. Tightening the evidence row,  especially the named examples. searching the web.
451. 37:55Bisby by sell. Bisby by sell. Cool. I like how  it's not just being a yes man and saying this one
452. 38:01can and this one can. This one can. It's saying  a $500,000 owner profit threshold eliminates most
453. 38:07easy side hustle examples and usually requires  multi-crew or multilocation scale. Appreciate the
454. 38:13honesty. So interesting watching it learn in real  time. It's an interesting route it's taking. It
455. 38:17says broker reported examples. So it's referring  to a business broker. And it's doing this, I'm
456. 38:23assuming, because I asked for specific examples of  500K SDE, sellers discretionary earnings, aka what
457. 38:30the business owner actually makes, because profit  is not it. You can make a million dollars profit,
458. 38:35but the owner is only able to pay himself a  fraction of that because he's reinvesting the
459. 38:39profit back into growth or he pays 38% in income  taxes or whatever. So, it decided to go look at
460. 38:45business broker listings because that's where  owner reported net profit is super interesting.
461. 38:52It says it's flagging unusually strong  earnings claims rather than treating them as
462. 38:56verified facts. So, it's checking itself. Elevator  maintenance, pest control, parking lot striping,
463. 39:01towing company, commercial cleaning, tree service.  Very cool. Okay, here we go. It's publishing it
464. 39:06on a web page for me. It didn't even ask it to  do that, but I'm down. Okay, research brief. 10
465. 39:11quiet businesses with the path to 500k in owner  earnings. Scale is mandatory. Broker data is not
466. 39:16audited. Okay. Parking lot and striping. I've  posted about this twice. Interviews like long
467. 39:22form interviews. We'll link to them below. 8,000  to 90,000 to start, which is not fully accurate.
468. 39:26You could start this for almost nothing. The  guy that I talked to actually did. You can buy a
469. 39:30$5 paint roller, but regardless, 35 to 40% net  profit. You would need $1.25 to $1.43 million
470. 39:37in revenue to have 500,000 in net profit. It  works because paint fades every 18 to 48 months.
471. 39:43People avoid it because it's night work, weather  windows, aggressive selling, bid accuracy, yada
472. 39:48yada. Septic drainage. Cool. Portable restroom  rental routes, commercial janitorial contracts,
473. 39:54pest control service routes. Number five, elevator  maintenance number six. And it says the downside
474. 39:59is license plus labor fortress, which I would  agree. Fire protection inspection. Number seven,
475. 40:04tree care and arborist. Number eight,  towing recovery and impound. And number 10,
476. 40:09pallet recovery, repair, and resale. Interesting.  And it shows how to verify these earnings claims.
477. 40:15Very cool. Now, I kind of want to know more  about the laundromat business. And I personally
478. 40:19don't really like the elevator maintenance  business. So, I'm going to ask it to swap that.
479. 40:26Swap number six with laundromats.  Okay, look at that. Number six,
480. 40:30multi storere laundromat portfolio. There it  is. Now, let's have it go a step further. Take
481. 40:35number six and build me a page I can actually  publish. Put every laundromat within 30 mi on
482. 40:39a map. I'll put a zip code. Here's the zip code  I grew up in. Put every laundromat within 30 mi
483. 40:43of 32796 zip code on a map so I can see which  towns have a gap. Pull in real photos. Which
484. 40:50neighborhoods? And I'm going to put 100 miles.  Have a gap. Pull in real photos. Then add what it
485. 40:55cost to start. Simple pricing and how I land my  first customers. Pull it all on one page. I can
486. 41:00share. Okay. Centered the study on Titusville,  Florida. Building an overlapping search grid.
487. 41:06Now because map indexes cap results per query  the page will describe. So it's scraping Google
488. 41:10places which is awesome. Looking at census  data population buffalo mess. Yeah. Okay.
489. 41:16Okay. Look at this. It's doing exactly what  I said. There's where I grew up. Tisville,
490. 41:20Florida 32796 within 100 miles. It's making a map.  Okay. I knew I asked it to do this, but I didn't
491. 41:26think it actually would. That's incredible. It's  interactive. It's like based on what seems to be
492. 41:31the Google Maps API. This is incredible. Okay.  Done. It built and published the Space Coast
493. 41:36Laundry opportunity map. Okay. Interactive map  with 78 laundromats and four screening gap areas,
494. 41:43overlapping searches, and then it even has a  design here. Where should the next laundromat
495. 41:47go? Amazing. Click on a couple pins. Pulls in the  reviews, the address, star. Beautiful. Remember,
496. 41:55these agents actually learn as you use them.  They remember things like you remember things.
497. 42:00So each time it runs and builds stuff like this,  it's better than it was before. And guys, this
498. 42:05is amazing. It's interactive. It's handing me the  actual page, not like some theory, not some mockup
499. 42:11or wireframe. This is something I could hand over  to an investor right now. We want to publish it.
500. 42:16We do. I'm just going to hit this drop down right  here. Allow search engines to index this page.
501. 42:21Yes, please. Publish. Done. Done. That's it. Okay.  Now, we want to be sure that this saves. So on the
502. 42:27left sidebar here, we're going to click agents.  Cool. So, I told it that I wanted to save this as
503. 42:32an agent because I want it to update and run every  day, every week, whatever. And it says click this
504. 42:38save as agent. It open this window right here.  We're going to name it uh deal scout. Then we're
505. 42:45going to click invocations right here. And we're  going to create a schedule for it just in plain
506. 42:49English. Run every morning at 7:00 a.m. Oh, got to  be more specific. Run searches for new laundromats
507. 42:55in other Florida zip codes every morning at 7:00  a.m. Boom. Now it is on the team with the rest of
508. 43:01my AI agents. I've got another one that scouts a  Facebook marketplace for me personally every day
509. 43:06looking for skid steers to buy because I'm on the  market for one right now. This is someone on my
510. 43:11quote unquote payroll that never stops working and  costs, you know, 99% less than an actual human.
511. 43:16Boom. That was incredibly easy. Just took four  or five steps. You can let it find an idea for
512. 43:21you. You can give it an idea. Either one. It'll  do the research, create videos, publish stuff,
513. 43:26everything you need. Go try Hyper Agent. Use  the exact same prompt that I used or your own,
514. 43:31whatever, and see what it digs up in your town.  Let me know in the comments what you find.