Transcriptly

AI is way Underhyped. He Runs His Entire Marketing Team with 40 AI Agents | Relay.app, Jacob Bank

EO15:44Added Aug 31, 2026

Jacob Bank had zero marketing experience a year ago. Now he runs 40 AI agents that do the work of a $50,000 marketing team, for $500 a month. The ex-Google p...

Watch on YouTube →
Contributed by 刘嘉琪

Transcript

Transcript format

  1. Intro

  2. 00:00We should all be having dozens or
  3. 00:02hundreds of agents working for us. So
  4. 00:03over time, I built up this team of 40 AI
  5. 00:06marketing agents to work with me. I'm
  6. 00:07the only marketing person at Related
  7. 00:08App. I have an AI coach that once a week
  8. 00:11tells me, "Here are the themes that seem
  9. 00:12to be really working well or the styles
  10. 00:14that are working really well for you. Do
  11. 00:15more of this, do less of that." That's
  12. 00:17why I've had a few very successful posts
  13. 00:19on LinkedIn. So for example, the post
  14. 00:21was here are the 40 marketing agents
  15. 00:23that I actually use. And that post so
  16. 00:25far has 1.5 million impressions and the
  17. 00:28most comments ever on a LinkedIn post.
  18. 00:30You know, I've been posting consistently
  19. 00:31on LinkedIn for a year with no results.
  20. 00:33And then all of a sudden, I was like,
  21. 00:34wo. If your job is as a junior content
  22. 00:38marketer to take a YouTube video and
  23. 00:40write a blog post based on that YouTube
  24. 00:41video, that job won't exist anymore. A
  25. 00:44high quality marketing contractor, let's
  26. 00:46say I had four people at $12,500 per
  27. 00:49month each. That would be $50,000 a
  28. 00:51month. And my AI bill is $500 a month.
  29. 00:53Building AI agents is the fundamental
  30. 00:55skill that will define every
  31. 00:57professional's career for the next 30
  32. 00:59years. It's a requirement. In the same
  33. 01:00way that knowing how to use spreadsheets
  34. 01:02like Microsoft Excel was a requirement
  35. 01:04for the last 40 years, knowing how to
  36. 01:06build AI agents is a requirement for the
  37. 01:08next 40 years. We are all using it for
  38. 01:10probably 1% of what we should be using
  39. 01:12it for. Here's what I'll say. AI is way
  40. 01:17underhyped.
  41. 01:23My name is Jacob Bank. I'm the founder
  42. 01:25and CEO of Relay.app. Relay.app is a
  43. 01:29platform to build AI agents and
  44. 01:30workflows that get work done on your
  45. 01:33behalf. So, right now, we're a
  46. 01:34nineperson team. We have five engineers,
  47. 01:36two designers, a product person, and
  48. 01:37myself. If we did not have AI, we would
  49. 01:40probably need to be a 15 person team to
  50. 01:42achieve the same level of accomplishment
  51. 01:44that we have now. Once upon a time, I
  52. 01:46was an AI researcher, then a startup
  53. 01:48founder, and then a longtime product
  54. 01:50lead at Google. And I've been working on
  55. 01:51relay.app since 2021. So, I've seen the
  56. 01:54whole AI boom over the last 3 years from
  57. 01:56a very close vantage point.

  58. Lesson 1 - AI is NOT Your Intern

  59. 02:02My take on AI is [music] it's like an
  60. 02:04intern.
  61. 02:05It's just an intern. It does exactly
  62. 02:06what you tell it to do.
  63. 02:07It's the intern.
  64. 02:08This is something I've changed my mind
  65. 02:10on. Actually, a year ago, I said, "Hey,
  66. 02:12think of an AI as your intern." [music]
  67. 02:14And then I realized that that was the
  68. 02:15wrong mental model. There are certainly
  69. 02:16a class of tasks that AI can do. Those
  70. 02:19are the time-saving kind of tasks that
  71. 02:21you would give otherwise to an intern.
  72. 02:23But AI can also do these strategy
  73. 02:24analyses and competitive analyses and
  74. 02:27content creation tasks that an intern
  75. 02:30definitely could not do. So for example,
  76. 02:32I am not good at sales calls. I do not
  77. 02:34have a background in sales. It does not
  78. 02:36come naturally. But it turns out there
  79. 02:37are some techniques that are pretty
  80. 02:38valuable in sales calls. So, I have an
  81. 02:40AI coach that reviews all the
  82. 02:42transcripts of my meetings and says,
  83. 02:43"Oh, you could have talked about this
  84. 02:44better. You could have articulated this
  85. 02:45value or you were too easy to eager to
  86. 02:47jump into a demo there. You should have
  87. 02:48asked some more discovery questions." If
  88. 02:50you go out and hire a sales coach, a
  89. 02:52good sales coach is like $10,000 a month
  90. 02:55for one meeting a week. You can build a
  91. 02:57pretty good AI sales coach by yourself
  92. 03:00and a little bit of, you know,
  93. 03:01transcript training and it'll cost you
  94. 03:03like five bucks a week to run.
  95. 03:07when there's a new role or function I'm
  96. 03:09taking on that I don't have experience
  97. 03:11in or I don't feel like I'm an expert
  98. 03:13at, how can I figure out the right
  99. 03:14coaching relationship where I can give
  100. 03:16my coach the raw data of what I'm doing
  101. 03:19and then have it give feedback to me on
  102. 03:21the right cadence whether it's after
  103. 03:23every call or once a day with a summary
  104. 03:25or once a week with a summary. That's
  105. 03:27why I'm now realizing that thinking
  106. 03:28about AI as an intern was so wrong.

  107. Lesson 2 - Managing People won't be Your Job Anymore

  108. 03:33So back when I was at Google, there was
  109. 03:35a role called tech lead manager and that
  110. 03:38was a hybrid role where you were both
  111. 03:39supposed to be the technical expert who
  112. 03:41knew the entire codebase and could
  113. 03:43review technical designs and solve the
  114. 03:44hardest problems. But you were also a
  115. 03:46manager. You were also responsible for
  116. 03:48coordinating the other people on the
  117. 03:50team. And it's a very hard role. It's
  118. 03:51hard to do both technical work and
  119. 03:53management work at the same time. But I
  120. 03:55think that's what the future holds for
  121. 03:56all of us. We all need to become super
  122. 04:00IC's. Let me explain why. If your job is
  123. 04:04as a junior content marketer to [music]
  124. 04:06take a YouTube video and write a blog
  125. 04:07post based on that YouTube video, that
  126. 04:09job won't exist anymore. You cannot
  127. 04:11build a job just doing single individual
  128. 04:13content repurposing tasks like that. If
  129. 04:15your job is to work at a large
  130. 04:18corporation and manage a team of a
  131. 04:19thousand people, I don't want to say
  132. 04:21that job is going to go away, but I
  133. 04:22think that job is going to become less
  134. 04:23and less important because companies are
  135. 04:25going to be smaller. We're going to have
  136. 04:26many more leaner, smaller companies. So
  137. 04:29that skill set of managing a thousand
  138. 04:31people, that's going to be less and less
  139. 04:33important. Everyone is going to look
  140. 04:35like me where about twothirds of your
  141. 04:37day is individual contributor work. I'm
  142. 04:39doing stuff. I'm making YouTube videos.
  143. 04:41I'm editing blog posts. I'm publishing
  144. 04:43content. I'm talking to customers. And
  145. 04:45then oneird of my day is coordinating my
  146. 04:47team of AI agents that helps me with
  147. 04:48those things. The thing that's so
  148. 04:50exciting about this super IC [music]
  149. 04:52future is that that's the work most of
  150. 04:55us love. Most people that I talk to who
  151. 04:57are in very junior roles are like, "Uh,
  152. 05:00I wish instead of just writing this blog
  153. 05:01post I was told to write, I had a bit
  154. 05:03more strategic influence on what we were
  155. 05:05doing and why." And everyone I know,
  156. 05:07including myself when I was at Google,
  157. 05:08who was in a complete organizational
  158. 05:10management role was, "Ah, I wish I could
  159. 05:11be closer to the work. I wish I could
  160. 05:13get back into product. I wish I could be
  161. 05:14closer to the customer and the user."
  162. 05:16And so now, I think every role in the
  163. 05:18future will combine the best of both
  164. 05:20worlds. We will all need to be strategic
  165. 05:22enough and senior [music] enough to
  166. 05:24decide what needs to be done and to be
  167. 05:26able to see what good looks like so we
  168. 05:27can give the right coaching, but and we
  169. 05:29need to be still hands-on enough that we
  170. 05:31can actually edit the blog post and edit
  171. 05:32the tweet and make the video ourselves.
  172. 05:34That's the future of the workplace.

  173. Lesson 3 - Two Rules to Build a 40-Agent System

  174. 05:40Yes, it's a bit overwhelming to see 40
  175. 05:42of them all at once, but if you click

  176. Build one simple agent at a time

  177. 05:44into any [music] individual agent,
  178. 05:46they're very simple and they all have
  179. 05:48very specific jobs. One, every time we
  180. 05:51create a new YouTube video, it creates a
  181. 05:52LinkedIn post automatically. Every time
  182. 05:54we create a new YouTube video, it
  183. 05:56creates a tweet automatically. Every
  184. 05:57time the CEO of one of our competitors
  185. 05:59posts on Twitter or LinkedIn, it lets me
  186. 06:00know what they're talking about. Every
  187. 06:02week, it goes and looks at our
  188. 06:03competitor pricing and it lets me know
  189. 06:05if anything important has changed. So,
  190. 06:06one by one, I built up this team of AI
  191. 06:09agents to help me. I have not had good
  192. 06:11luck with trying to build a single agent
  193. 06:13to try to do 25 things at once. That is
  194. 06:16one difference maybe with human
  195. 06:17employees where when you hire a social
  196. 06:19media marketer, you want them to own all
  197. 06:20of social media marketing, especially if
  198. 06:23you're a beginner. Start simple. Create
  199. 06:25one agent that can do one thing, create
  200. 06:27a second agent that can do a second
  201. 06:29thing, create a third agent that can do
  202. 06:30a third thing, and then maybe have one
  203. 06:32agent that sits above them and helps
  204. 06:34invoke them when necessary. And then
  205. 06:36build up over time. Do not start with a
  206. 06:3840 agent or chart. Start with one.
  207. 06:42One very important thing to keep in mind
  208. 06:44is that AI agents are not a set it and
  209. 06:47forget it kind of thing. You can't just
  210. 06:48build an AI agent and say, "Oh, this
  211. 06:50thing is just going to work for me

  212. Fire the agents that don't deliver

  213. 06:51forever." I'm constantly modifying my AI
  214. 06:54agents. For example, I do assessment
  215. 06:56calls with new customers where I'll
  216. 06:58spend 30 minutes with them and we'll
  217. 07:00identify opportunities for AI in their
  218. 07:01business. And I thought, wouldn't it be
  219. 07:03cool if they got a personalized Google
  220. 07:05Doc from me after the call with what we
  221. 07:07had talked about with five specific use
  222. 07:09cases and links to templates? And I
  223. 07:10thought this was so cool. So I built an
  224. 07:12AI agent to help me with it. I said, "AI
  225. 07:14agent, your job is to take this
  226. 07:15transcript of the meeting and turn it
  227. 07:17into a really nicely formatted Google
  228. 07:18doc to send to the customer." And after
  229. 07:20doing that 10 times, I was like, "This
  230. 07:22is not the right format. The customers
  231. 07:23aren't reading the Google doc. I should
  232. 07:24have just included in the email." So I I
  233. 07:27fired that agent and I said, "Uh, we
  234. 07:30don't actually need that job to be done
  235. 07:32anymore." I actually repurposed that
  236. 07:33agent and said, "Can you just take that
  237. 07:35summary and put it directly into the
  238. 07:36follow-up?" Sometimes [music] you just
  239. 07:38don't need that function to be done
  240. 07:39anymore. Like we were very focused on
  241. 07:41SEO a year ago. And then we realized
  242. 07:44that it's just not the important channel
  243. 07:45for us right now. It's such a small
  244. 07:47percentage of our new users and it's
  245. 07:48relatively high effort. So we're not
  246. 07:50going to focus on SEO anymore. So I told
  247. 07:51my AI SEO agents, you guys can stop for
  248. 07:54now. Like I'll come back to you in the
  249. 07:56future, but for now you can stop doing
  250. 07:57what you're doing. In the same way that
  251. 07:58you would tell an SEO agency, we don't
  252. 08:00need SEO at the moment right now. We'll
  253. 08:02come work with you again in the future.
  254. 08:04And the really cool thing is there's no
  255. 08:06emotional baggage. There's no
  256. 08:08disagreements. There's no coordination
  257. 08:11costs. I just have an idea. Hey AI
  258. 08:13agent, can you help me with this idea?
  259. 08:14And they say, "Yeah, absolutely. Let's
  260. 08:16do it." Um, and so I don't want us to
  261. 08:18think about it as replacing people. I
  262. 08:20want to think about it as giving me
  263. 08:22superpowers. I wish I had the time to do
  264. 08:25all these marketing activities cuz I
  265. 08:26love doing them. They're very valuable
  266. 08:28to the company. And now I can do that.
  267. 08:30We all want to do more at our jobs.
  268. 08:32We're all ambitious people. We're all
  269. 08:33motivated people. We all wish we could
  270. 08:35do more and now we have the tools to do

  271. Lesson 4 - Staying at Google is the Riskiest Career Choice

  272. 08:37so.
  273. 08:41It's fascinating to think about how
  274. 08:43perspectives on what a risky career is
  275. 08:46versus [music] what a safe career is
  276. 08:48have changed over the years. In my
  277. 08:50parents' generation, a safe career was
  278. 08:53working at one large Fortune 500 company
  279. 08:55[music] for 40 years. You know, getting
  280. 08:57your watch when you retire or after 50
  281. 08:59years. And a risky career was starting
  282. 09:01your own company or moving between
  283. 09:04companies every couple of years. I think
  284. 09:06that has completely flipped. The
  285. 09:08riskiest career is if your skills are
  286. 09:11too tied to the environment of one
  287. 09:13particular company that could change
  288. 09:15over time. And the most robust careers
  289. 09:18are people who are starting their own
  290. 09:20companies and who are having lots of
  291. 09:22different life experience. I'll give you
  292. 09:23an example. Much of our team I've
  293. 09:24recruited from Google because we all
  294. 09:26work together at Google and now we're
  295. 09:28working on this startup. And if you ask
  296. 09:3010 people what's riskier, staying at
  297. 09:32Google or joining a startup, most of
  298. 09:34them will say it's safer to stay at
  299. 09:36Google and it's riskier to join a
  300. 09:38startup. And in terms of your cash
  301. 09:40compensation for the next year, yeah,
  302. 09:42that's true. Like Google is going to pay
  303. 09:44you more in cash for the next year than
  304. 09:46most startups will. But if you think
  305. 09:47through the long-term lens of your
  306. 09:49career where it's about how broad of a
  307. 09:51network have you built, how many new
  308. 09:53skills have you built? How have you
  309. 09:55pushed yourself to develop new
  310. 09:57experiences? It is way riskier to stay
  311. 10:00at a company like Google than it is to
  312. 10:01join a startup or start a startup.
  313. 10:03There's two ways you can react to the
  314. 10:04current AI agent craze. You can say,
  315. 10:07"I'm skeptical. I don't believe it. I'm
  316. 10:09going to put my head in the sand. I'm
  317. 10:10going to keep working the way I've
  318. 10:11always worked." That's one way to react.
  319. 10:13The second way to act is like, whoa,
  320. 10:16this is a cool tool that's going to
  321. 10:18completely change the way that work is
  322. 10:19done. And I want to be on the cutting
  323. 10:22edge of how I'm using it. Please be
  324. 10:24person number two. That's going to serve
  325. 10:25you much better. I never want to
  326. 10:28communicate you should take more risk.
  327. 10:30That's not the way I frame the
  328. 10:31discussion. It's not you should take
  329. 10:33more risk. [music] The way I frame the
  330. 10:34discussion is you should optimize your
  331. 10:37career for personal growth and learning
  332. 10:40new skills. And often the way to create
  333. 10:43personal growth and learn new skills is
  334. 10:45to go into newer and more uncertain
  335. 10:47environments where you'll have more
  336. 10:49responsibility and you're less sure of
  337. 10:50what's going to happen. And I don't
  338. 10:52think that's risky. I think the riskiest
  339. 10:55thing you can do is to stagnate. And so
  340. 10:57for me, it's all about reframing the
  341. 10:59conversation about what risk is. Risk is
  342. 11:02stagnation. Progress means learning and
  343. 11:04growing and pushing yourself. And so I
  344. 11:06always want people to optimize for
  345. 11:07progress and growth and pushing
  346. 11:09yourself. I have no marketing
  347. 11:10background. I am terrible at marketing.
  348. 11:12I was not even on any social media
  349. 11:14platforms until a year ago. I hate
  350. 11:16sharing stuff about myself in public. It
  351. 11:18was deeply uncomfortable for me to be
  352. 11:20the person responsible for marketing. I
  353. 11:21realized
  354. 11:24that organic social media and marketing
  355. 11:26on LinkedIn was the best channel for us
  356. 11:28to succeed as a company. But I worked
  357. 11:30really hard at it for a year to learn to
  358. 11:32get good at it. And now I love it. And
  359. 11:34so it could have felt, oh, it's so risky
  360. 11:37to put so much effort in this marketing
  361. 11:38strategy that we haven't tried. No, it's
  362. 11:41an opportunity to take advantage of an
  363. 11:43uncertainty to learn and grow in a in a

  364. Lesson 5 - What a founder who uses AI all day wants his kids to know

  365. 11:45new direction.
  366. 11:50And I think often about how different
  367. 11:51the world my kids are going to live in.
  368. 11:53There are certain areas of technology
  369. 11:54where I'm just like, "Wow, that's so
  370. 11:56cool. That's obviously better." I'll
  371. 11:58give you one small example. where we
  372. 11:59live on Puro Hill in San Francisco is
  373. 12:01very close to the main testing area for
  374. 12:03Whimo and Zuks. Zuks is I think it's
  375. 12:06Amazon's self-driving car division. And
  376. 12:08so I take my kids to school. My two
  377. 12:10older daughters are sitting behind me on
  378. 12:11a cargo bike and they're saying Whimo
  379. 12:13Zuks Whimo Zuks. And for them it is so
  380. 12:17normal to see a car with nobody driving
  381. 12:19it. They think it's cool because they
  382. 12:20have all like the cameras and lasers on
  383. 12:22top, but it is so natural to them the
  384. 12:25idea that a car can be driven by a
  385. 12:27human. And so in the particular area of
  386. 12:29self-driving cars, I am so happy that I
  387. 12:32think neither of them will ever need to
  388. 12:33drive a car by themselves [music]
  389. 12:35because it'll be so much safer and so
  390. 12:37much more convenient and so much better
  391. 12:39for the environment when all these cars
  392. 12:40are electric. There are other areas
  393. 12:42where I have more mixed feelings like I
  394. 12:46don't want them to be just scrolling
  395. 12:48YouTube shorts all the time even though
  396. 12:50there's a very very good algorithm. The
  397. 12:52general way I think about it is maybe
  398. 12:54twofold. One [music] is I try to give
  399. 12:57them as many real world challenges and
  400. 12:59experience as possible. Yeah, you still
  401. 13:01got to learn how to ride a bike. You're
  402. 13:02going to skin your knees when you're
  403. 13:03riding your bike. This is what I always
  404. 13:04say to my daughter, skin knees are good
  405. 13:06for kids, right? Cuz if you're if you're
  406. 13:07skinning your knees, it means you're
  407. 13:08trying something and taking risk in the
  408. 13:09real world and dealing with the
  409. 13:11consequences. This is another big
  410. 13:13breakthrough when I was teaching my kids
  411. 13:14to swim. It's okay if they swallow some
  412. 13:16water. That's going to happen when
  413. 13:17you're learning how to swim. Like, it's
  414. 13:19okay if you swallow water from time to
  415. 13:20time cuz that's what it takes to learn
  416. 13:21how to swim. So, I'm trying to have them
  417. 13:23take, you know, some risk in the real
  418. 13:25world. The thing that I haven't thought
  419. 13:26about yet, cuz my kids are too young, I
  420. 13:29don't know how tools like CIABT or AI
  421. 13:32agents like we're talking about now, I
  422. 13:34haven't yet thought about how those are
  423. 13:35going to influence their educational
  424. 13:36journey. But because I'm working on AI
  425. 13:38agents, I do think a lot about what my
  426. 13:39kids careers are going to look like. And
  427. 13:41here's [music] my current perspective on
  428. 13:42that. Number one, the ability to clearly

  429. Articulate clearly what you want AI to do

  430. 13:45[music] articulate what is important,
  431. 13:48what needs to be done, and [music] give
  432. 13:51guidance on how to do it. That will
  433. 13:53always be important. That will only be
  434. 13:54more important in the future because our
  435. 13:56jobs will be more of how do we
  436. 13:58efficiently instruct AI on what it needs
  437. 14:00to do rather than how do we do it all
  438. 14:02ourselves. So, I'm spending a lot of
  439. 14:04time like I really want my kids to learn
  440. 14:06logic and philosophy because I really
  441. 14:08want them to be able to very clearly
  442. 14:09articulate what they want. And then
  443. 14:11number two, I want them to think about

  444. Find where your personality is the value

  445. 14:14areas where their unique personality and
  446. 14:18abilities can shine through in an in a
  447. 14:20way that AI wouldn't. And so that's why
  448. 14:22I record our own YouTube videos because
  449. 14:24it's an expression of my personality and
  450. 14:26people build a social connection. Those
  451. 14:28are going to be the two fundamental
  452. 14:29skills for the future of the workplace.
  453. 14:31How good are you at articulating what
  454. 14:33you want and telling that to AI? And
  455. 14:35then how good are you building social
  456. 14:37connection with others? And those I
  457. 14:39think are going to be durable skills for
  458. 14:40the future of any career.
  459. 14:43There's no doubt that the US is ahead of
  460. 14:46China on artificial intelligence, but
  461. 14:48the claim to technological leadership
  462. 14:50for the United States appears to me to
  463. 14:53be pretty narrow. What are Americans
  464. 14:54going to be using AI for? Well, to make
  465. 14:56a lot more powerpoints perhaps and to
  466. 14:59file a lot more lawsuits against each
  467. 15:01other. Whereas the Chinese are much more
  468. 15:03manufacturing focused. So, China is
  469. 15:05going to use AI to make more iPhones,
  470. 15:08make more electric vehicle batteries,
  471. 15:10make more drones, and make more
  472. 15:11munitions. Last year, the uh United
  473. 15:14States built around [music] five ships.
  474. 15:17China built about 1,500 ships. I've
  475. 15:19spent about equal parts of my life
  476. 15:21between the US, uh, China, and Canada.
  477. 15:25Right now, it looks like the US and
  478. 15:27China are both very intent on losing,
  479. 15:30not on winning. Both of these countries
  480. 15:32making very big mistakes.