I Shut Down My AI Agency. Here's What's Worth Building — Transcript

Full transcript

1. 0:00I shut down my AI agency about nine months  ago, and in this video, I'm gonna show you
2. 0:05what is actually worth building today.  The reason I shut it down was simple.
3. 0:10The margins were low, and the client  results weren't quite there. Since then,
4. 0:14of course, AI has improved tremendously,  and so today you can build so much more
5. 0:18without hiring more people. But this  raises another important question,
6. 0:23what is actually worth building today? Or even  more important, what is not worth building?
7. 0:28Because I see so many people  simply focusing on their own
8. 0:31thing, and I get asked this question  constantly. So by the end of this video,
9. 0:36you'll know the exact three things  that are not worth building today,
10. 0:39the three things that are worth building today,  and how you can get started over the weekend.
11. 0:45And if you get this right, I guarantee that you  will be able to make a steady income stream in the
12. 0:50AI era without being constantly afraid of losing  your job or not having any work in the future.
13. 0:57But first, quickly about me. My name is  Arsenii, and I've been building in AI
14. 1:01and working for myself for over five years  now. I started as a freelancer on Upwork,
15. 1:06and then I scaled my AI agency to  $80,000 per month before I shut it down.
16. 1:11Then I also scaled AppointlyFi, which is my  fully automated AI lead generation company,
17. 1:16to $15,000 per month, and my open source  projects also have 7,000 stars on GitHub.
18. 1:22Now let's start with the number one  thing that's not worth building anymore,
19. 1:26which is custom automations for every client if  you're still building custom automations agents
20. 1:32or other solutions for every single client, , I  urge you to stop doing this as soon as you can.
21. 1:37This is completely fine at the beginning for  your first one or two projects, but then as
22. 1:42soon as you get some case studies and as soon  as you get some experience in a certain niche,
23. 1:46please find a service that you can  productize and deliver consistently.
24. 1:51Today, AI is such an integral part  of every single business that you
25. 1:55cannot simply outsource AI development  to some team overseas. It would be kinda
26. 2:00like hiring contractors for some of  the most key roles at your company.
27. 2:05Contractors simply don't know how your company  runs, and if you don't have a clear process
28. 2:10for hiring and training employees, well,  your business is just never gonna scale.
29. 2:14and so that's what we noticed  in our own AI agency as well,
30. 2:18is that we would build an AI agent for  a business, and then they would start
31. 2:22using it. They would be excited about it for a  week or two, and then they would simply stop,
32. 2:28and they would either switch to something  else or they would stop using it completely.
33. 2:32And the reason is because an agent is  never perfect. There is no perfect skill,
34. 2:36there is no perfect agent, and there is  no perfect automation that you can build
35. 2:40for a business. every single  business evolves over time.
36. 2:43And so every agent that you build for  that business must evolve with it as well
37. 2:48So here's what actually got my clients results.  Now I sell a productized offer that customizes
38. 2:56itself. Essentially, now I offer this AI OS  install, which is the exact system that I built
39. 3:02myself first in my own business over the last  four years, and then I provide training videos
40. 3:07and AI skills that allow business owners  to customize the same system themselves.
41. 3:13So this solves two issues. Number one,  it solves my time, and I don't have to
42. 3:17hire anyone to deliver it because the business  owners literally customize it themselves. They
43. 3:23watch the video, then they run the skill.  The skill asks them certain questions,
44. 3:27and then based on these questions,  this skill either builds new agents,
45. 3:30new skills, or customizes the  system for them specifically.
46. 3:34And secondly, they now understand how  to actually run the system themselves.
47. 3:39so if there's some adjustment or  something that isn't perfect yet,
48. 3:43they don't have to come back to me every single  time, and I don't have to do the work for them.
49. 3:48They can simply adjust it themselves, and this  way they keep evolving their own agents and AI
50. 3:55systems as their company grows. And this  has completely changed everything for me.
51. 4:00The client results that I'm getting right  now from these productized AI OS installs
52. 4:05are just on a completely different level from  what we used to get from custom solutions
53. 4:10and I don't even have to do any kind of outreach  simply because clients refer me new clients,I have
54. 4:11two case studies on my website already,  and I have two more coming soon.
55. 4:15Okay, now if building is so cheap, another trap  that I see people fall into is going too big,
56. 4:22and I actually personally fell for that myself
57. 4:25And even Sam Altman said that you  can now build all thirty things,
58. 4:28and you can get people to use them all.
59. 4:30So today, you really don't have  to be laser-focused on one thing
60. 4:35to make it successful. You can build  three, four things at the same time,
61. 4:38although I certainly do not  recommend this at the beginning.
62. 4:42But where most people fall short is instead of  focusing on a small solution that people really
63. 4:48need, they focus on this billion-dollar problem  even before they made their first 10K online. And
64. 4:55I actually felt for the same trap myself as well.  I was trying to build a platform called Agency,
65. 5:00which is a billion-dollar vision, and  unfortunately, for now it didn't work.
66. 5:06for now, I paused this project. I don't think  it will be over yet, but I'm currently trying
67. 5:11to find product market fit in a different niche,  which is exactly what I'm gonna talk about next.
68. 5:17And so what I'm doing now is instead I'm focusing
69. 5:20on more niche use cases where it's  really clear who the end user is
70. 5:25So the thing worth building number two  is niche software products for niche
71. 5:30underserved communities. While everyone  is chasing this billion-dollar vision
72. 5:35and everyone is going after the same  thing, which is kinda obvious today,
73. 5:39you can serve a very specific niche market  that no one else is chasing right now.
74. 5:44And that's exactly what I'm doing right now with  my own framework and the platform. Instead of
75. 5:50building a product for everyone, which basically  means for no one, I'm focusing on one specific
76. 5:56community called Well-Founded Creatives, and we're  building a very specific set of agents for them.
77. 6:02And what I'm seeing is that the feedback,
78. 6:05kinda like with custom solutions versus  productized offerings, is just so much
79. 6:09better and so much clearer. Even with smaller  models, even with not fully fine-tuned agents,
80. 6:15they are already getting a ton of value because  these agents are made specifically for them.
81. 6:21They help them to create their decks,
82. 6:23nail down their messaging, and get  to the right offer a lot faster.
83. 6:27what also surprised me is that because  these agents are so tightly fine-tuned
84. 6:32for this specific community on these  specific resources for these people,
85. 6:36we also don't even have to use  these latest crazy frontier models.
86. 6:41They get results even with  smaller ones just as well
87. 6:45But the best part is that because you're  building for the specific niche market
88. 6:48or a niche community, you don't have to build  anything upfront. You can simply ask them what
89. 6:54is their problem and how much would they pay  to get this problem solved. And then you can
90. 6:59charge a deposit and promise them that you're  gonna help them to get this problem solved,
91. 7:03and only then you actually build the  product and then charge the full price.
92. 7:07This way, you de-risk yourself from building
93. 7:09something for three months that  eventually no one even needs
94. 7:13However, even when you do know who this is for,  there is one more trap that a lot of people fall
95. 7:18into. Building things that consume your usage but  do not move any of your goals forward. a clear
96. 7:25example of such a thing and what is not worth  building, number three, is your own custom AI OS.
97. 7:32So I have seen a ton of people vibe  code their own AI OS dashboards that
98. 7:36look like this with all these lines  and moving parts. They look amazing,
99. 7:40and I would love to play with one myself.  But the question is, what problem does it
100. 7:45solve? Like, what, what does it actually  do? What is on this dashboard right now?
101. 7:50that I cannot get from Claude Code? What actually
102. 7:53matters when you're building  your AI OS is the structure.
103. 7:56It's organizing all your context and  making it actionable for your AI,
104. 8:01not connecting some random dots and lines  together. That literally does not bring any
105. 8:06value Here is another example that I recently saw  on Twitter. Mark Lou, a very famous indie hacker,
106. 8:12recently vibe coded his own Beehive to go  from about $4,000 a year to under $300 a
107. 8:19year. And Beehive obviously worked for him, but  he wanted to cut costs, and that's what he did.
108. 8:26And the founder of Beehive also posted  all of the features that you would need,
109. 8:30which is obviously like a crazy amount  of things that aren't even related to
110. 8:34vibe coding and that you can't get just  from maxing out your usage on CloudCode.
111. 8:39Like for example, 99% deliverability  and avoiding spam filters
112. 8:44If he built it for some niche community where
113. 8:47some features in Beehiiv were clearly  missing, then I would understand that.
114. 8:51But rebuilding an entire platform because  you want to cut 3.5K per year and adding all
115. 8:58that maintenance depth on top just doesn't make  sense to me So what is actually worth building?
116. 9:04The last thing and the most important thing,  number three, that is worth building is real
117. 9:09connections with real people. Two things that AI  will never be able to automate are, number one,
118. 9:16building connections with people, because people  build connections with other people, not with AI.
119. 9:21And number two, taking accountability for  the results, because whether you get the
120. 9:26result or not, you're gonna have to pay for  the tokens. AI does not guarantee you any
121. 9:30kinds of results. It simply charges  you per token. And so of course,
122. 9:34there are no molds today. You  can rebuild pretty much anything.
123. 9:38But if you have built a specific connection  with someone who is already successful in their
124. 9:43own niche, and if they know that you care  about the results that they're gonna get,
125. 9:47they will still choose to work  with you instead of working with AI
126. 9:52This is why consulting firms like  Deloitte and McKinsey still exist.
127. 9:56They don't necessarily do anything better than  Claude Opus 5.5, for example, but many Fortune
128. 10:03500 companies still go to them whenever something  happens for two reasons. Number one, they simply
129. 10:09already have the connections with those consulting  firms, and it's kinda like automatic for them.
130. 10:14It's almost in their SOPs, And number two  is because they actually take responsibility
131. 10:21for the outcomes of the case. they  don't just hand you over a document,
132. 10:26but if the deal falls through because of this  document, because it's incorrect or something,
133. 10:31they're actually gonna reimburse you  for all the expenses and the costs.
134. 10:35So again, this is not something that AI is gonna  be able to automate anytime soon. This is where
135. 10:40people will still be needed whatever happens  next. Even if models become smarter than all
136. 10:44humans combined, you will still need someone  to blame for the outcomes of the actions.
137. 10:44And so if you wanna stand out today in 2026,  the most important thing is to build real
138. 10:49connections with real people and guaranteeing  a real result. And then you use AI on the back
139. 10:52end yourself to deliver that result. And I  know it feels very tempting to keep building.
140. 10:53I fall for this all the time myself  as well. But unfortunately, most of
141. 10:53the time it's the least productive  thing you can do. So what you should
142. 10:53do instead is stop building products  and start building real connections.
143. 10:58Reach out to a few people first.  And then preferably even charge
144. 11:02upfront before you build anything. And  if you wanna see exactly how to do this,
145. 11:06how to actually connect with someone and send  an outreach message live, I recorded my entire
146. 11:12LinkedIn outbound system in this video. I show  you everything, how to get to the right offer,
147. 11:17how to write your messaging, how to send an  actual message that cuts through all the AI slop.
148. 11:21I actually even record a message live  and then send it during this video. So
149. 11:26check it out, it's completely  free, and I'll see you there