YouTube2Text

Claude Code + AIsa : j’ai créé un agent IA qui automatise ma prospection commerciale — Transcript

by Ethel Cédric · 3,468 words · 510 segments · language en · Watch on YouTube

Full transcript

  1. 0:00One of the questions you ask me most on
  2. 0:02this channel is: how do you find
  3. 0:04clients when you offer artificial
  4. 0:06intelligence services? And that is
  5. 0:08exactly what we are going to look at in
  6. 0:10this video. If you are new to the
  7. 0:12channel, I’m Etel Cédric, actor and
  8. 0:13data scientist. And in this video, I
  9. 0:16will show you step-by-step how to use
  10. 0:18Claude Code to create an AI agent
  11. 0:20capable of searching for clients 24/7
  12. 0:23for your business. So, get comfortable,
  13. 0:26let’s get started. OK, so we’re
  14. 0:29going to head over to Claude Code and
  15. 0:31start the project together. For this
  16. 0:33video, I assume everyone has used
  17. 0:35Claude Code at least once. If that’s
  18. 0:38not the case, it’s very simple to
  19. 0:40install. You simply go to Google and
  20. 0:43search for Claude Desktop. You will
  21. 0:46install the application, create an
  22. 0:48account, and once you open the
  23. 0:50application, you will see this
  24. 0:52interface. So, we can start the project
  25. 0:55. On my PC, I created a "client
  26. 0:58prospecting" folder for this project
  27. 1:00with a "Claude Project" directory. And
  28. 1:03in that directory, I put two Markdown
  29. 1:05files. The first file is the "claude.md
  30. 1:08" file, which represents the entire
  31. 1:10memory and context of my project. I’m
  32. 1:13going to open it so we can read through
  33. 1:15it briefly together. So, in my "
  34. 1:17claude.md" file, what did I specify?
  35. 1:20First, I presented the project to it. I
  36. 1:23told it that this project consists of
  37. 1:25developing a sales prospecting agent
  38. 1:27for Neflow AI. So, Neflow AI is my
  39. 1:31artificial intelligence agency. I
  40. 1:35defined the goal, which is to create an
  41. 1:37application that will automate part of
  42. 1:39the research and qualification of local
  43. 1:42prospects. Basically, we are going to
  44. 1:46provide a set of information to our
  45. 1:47application, specifically the business
  46. 1:50sector. It could be the hospitality
  47. 1:53sector, for example, the restaurant
  48. 1:56industry, or SMEs specialized in
  49. 1:58digital services. The second piece of
  50. 2:02information we will provide to our AI
  51. 2:04agent is the city or geographic area.
  52. 2:06It could be a country, a city, or a
  53. 2:08small neighborhood so that it can
  54. 2:10conduct a targeted search. The third
  55. 2:13piece of information we will ask the
  56. 2:14agent for is the number of prospects we
  57. 2:16want. Once we provide this information
  58. 2:19to the agent, it will then search for
  59. 2:22companies. It will quickly analyze
  60. 2:24their digital presence and generate
  61. 2:26personalized prospecting emails.
  62. 2:29Secondly, I will specify the main
  63. 2:31objective, the target workflow. OK? In
  64. 2:35a third step, I describe here the user
  65. 2:38inputs. OK, in a fourth step, I will
  66. 2:40therefore specify the data sought. What
  67. 2:43am I looking for in companies, exactly?
  68. 2:45The name, address, phone number, rating
  69. 2:48, number of reviews, website, social
  70. 2:51media, and useful information to
  71. 2:53understand their digital presence. In
  72. 2:57fifth, I will therefore specify some
  73. 2:59qualification criteria for the leads.
  74. 3:01So each company is filtered into a
  75. 3:04priority of either high, medium, or not
  76. 3:07relevant. So that is generally what my
  77. 3:09cloud file looks like. MD looks like.
  78. 3:11The second file is the prompt. It is
  79. 3:13the prompt that will allow us to create
  80. 3:15the AI agent. So I am opening it so we
  81. 3:17can see together what it looks like. So
  82. 3:19here, what is the agent's mission? You
  83. 3:22are the sales prospecting agent for
  84. 3:23Neflow EAI. So your mission is simple.
  85. 3:26Find local companies matching the
  86. 3:28requested criteria. I would like to
  87. 3:32clarify that I will put both files in
  88. 3:33the description so that you can
  89. 3:35download and modify them according to
  90. 3:37your own context. Now that we have
  91. 3:40presented the context of this project,
  92. 3:42we can move on to practice. We are
  93. 3:44going to go into Claude. I am opening
  94. 3:45it here. You can see that this is the
  95. 3:47chat section. So you can write here,
  96. 3:50you hit enter, and you get a response.
  97. 3:52That is not what we are going to use.
  98. 3:54What we are going to use is the Claude
  99. 3:55Code section. So you are just going to
  100. 3:57click there. OK. So these are the
  101. 4:00different projects I have worked on.
  102. 4:02The first thing to do when in Claude
  103. 4:04Code is to provide it with the working
  104. 4:06directory. It is a very simple thing to
  105. 4:08do. You are simply going to click here
  106. 4:10on the folder. OK. You go to your work
  107. 4:13folder, cloud tuto. I have the sales
  108. 4:17prospecting, and I will provide it with
  109. 4:19the Claude Project directory. I click
  110. 4:22on select a folder. OK. So here, we
  111. 4:26have just provided Claude with access
  112. 4:28to our working directory. To launch the
  113. 4:30creation of the AI agent, what we will
  114. 4:32do is very simple. We are going to copy
  115. 4:35this prompt file. We go into Claude
  116. 4:37Code and we are going to paste it there
  117. 4:39. OK, we have our prompt file here. We
  118. 4:43are just going to tell Claude, execute
  119. 4:47these instructions please. OK. So here,
  120. 4:50we have provided the prompt file to
  121. 4:53Claude. Knowing that in our working
  122. 4:57directory structure, we have the
  123. 4:59cloud.mmd file. OK. Which is right
  124. 5:03there. So Claude will be able to read
  125. 5:06it. The second super important thing to
  126. 5:08do is to choose the model. So here, we
  127. 5:10have different models. You can click
  128. 5:12right there on Opus 4.6. You have
  129. 5:14different models there. Uh, Fable 5.1,
  130. 5:17which is currently the most
  131. 5:18high-performance model from Anthropic.
  132. 5:21We have Opus 5, we have Sonnet, we have
  133. 5:23Haiku, and you can click here for more
  134. 5:26models. So you have different models.
  135. 5:29Now, for really specialized tasks that
  136. 5:32will require coding, I recommend you
  137. 5:35use at least Opus 4.6, really Opus 4.6.
  138. 5:40And for the most relevant tasks, you
  139. 5:42can go up to Fable 5.1. But one thing
  140. 5:47to keep in mind is that Fable 5.1 is
  141. 5:49the most powerful model, but it is also
  142. 5:51the most expensive. So if you want good
  143. 5:55value for money, I recommend using Opus
  144. 5:584.7 or 4.8, for example. Okay? So for
  145. 6:02this video, I'll stick with Opus 4.6
  146. 6:04because we're working on a fairly
  147. 6:06simple task. We want to create an AI
  148. 6:08agent that will perform online company
  149. 6:10research. So I've chosen Opus 4.6 here.
  150. 6:14Once that's done, I can click on
  151. 6:17execute. And here, Claude will first
  152. 6:20try to gain access to the working
  153. 6:22directory. Then, it will read the
  154. 6:25various files. Okay, perfect. So now,
  155. 6:27it is iterating. You can see the
  156. 6:30iteration time is currently 48 seconds
  157. 6:32and the number of tokens consumed is
  158. 6:35184. Okay, so now it's in a reflection
  159. 6:38phase. We'll wait just a little bit.
  160. 6:41Okay, perfect. So Claude Code has
  161. 6:43completed a certain number of
  162. 6:44iterations. It has almost finalized the
  163. 6:47application creation, but it needs one
  164. 6:49important piece of information. You can
  165. 6:52see here, it's asking me a question. It
  166. 6:56says dependencies are installed. Before
  167. 6:59starting the server, you need to
  168. 7:01configure your Anthropic API key. Do
  169. 7:04you already have a .env.local file with
  170. 7:07your Anthropic API key? This is a very
  171. 7:11important step because since your
  172. 7:14application is supposed to search for
  173. 7:17companies online, you need an Anthropic
  174. 7:20key to perform this specific task. That
  175. 7:24is why Claude Code is asking if you
  176. 7:26already have a local environment file
  177. 7:28with the key provided. So we'll just
  178. 7:30tell it no, I don't have a .env.local
  179. 7:33file with my Anthropic key entered. I
  180. 7:36execute. It will then create this file
  181. 7:40itself. We will just need to input our
  182. 7:43API key. Okay. So now, it has created
  183. 7:46the local .env file and is simply
  184. 7:48asking me to enter my API key. And if
  185. 7:51you've never done it, it's very simple.
  186. 7:54You just go to Google and type in Cloud
  187. 7:57Platform. Click on the first link. OK,
  188. 8:04perfect. In your case, if you've never
  189. 8:07created an account in Cloud Console,
  190. 8:08you will simply need an email address
  191. 8:10to create an account. Once that's done,
  192. 8:14you can create an API key by clicking
  193. 8:17on get an API key. Here, in my specific
  194. 8:19case, I had already created an API key.
  195. 8:21So I can go to the API keys section and
  196. 8:24my API key appears there. So if I
  197. 8:27remember the key, I can copy it and
  198. 8:29come paste it into Cloud Code. And if
  199. 8:33not, I can simply create a new key. So,
  200. 8:37I remember my API key. I will therefore
  201. 8:40just come here and paste it. OK. So as
  202. 8:45soon as I provide my key to Cloud Code,
  203. 8:47it can continue creating the
  204. 8:49application. OK, Cloud Code has created
  205. 8:52the application, it is accessible at
  206. 8:55this URL. We will simply copy the link,
  207. 8:58go to Google, and paste it there. OK,
  208. 9:02so here we have our Neflori application
  209. 9:04. So here we can define our different
  210. 9:07criteria. We will, for example, put
  211. 9:09restaurant, the city, we will put Paris
  212. 9:11. And the number here, we will, for
  213. 9:13example, put one restaurant. OK, we
  214. 9:14launch the search. So when we click on
  215. 9:17search in the application, our AI agent
  216. 9:21begins to look for prospects. So here,
  217. 9:24in this instance, it will look for a
  218. 9:27restaurant in Paris, OK, that has a
  219. 9:30poor online presence. It will then see
  220. 9:33if it is possible to offer services to
  221. 9:36this restaurant. It will then generate
  222. 9:39an analysis report for us. So we will
  223. 9:41wait a bit for the search to be
  224. 9:43finalized. OK, perfect. The application
  225. 9:46has finished the search. It found a
  226. 9:49restaurant, in this case, the Suite
  227. 9:52Insulty. OK, so we'll go to view, and
  228. 9:57here we see a 4.6 rating, 692 reviews,
  229. 10:02at € 30. OK. Is that what the
  230. 10:05application is telling us? So here, the
  231. 10:08application tells us 4.7 with 410
  232. 10:10reviews. So that is problematic. There
  233. 10:16is truly a discrepancy between what the
  234. 10:18application provided and the reality.
  235. 10:21OK. And that is a major limitation
  236. 10:24because the application does not
  237. 10:26actually use a specific real-time
  238. 10:29search API. The application used
  239. 10:32Anthropic's search API, but this API
  240. 10:35likely pulled from Claude’s training
  241. 10:38data, which led it to provide this
  242. 10:41incorrect information. To address this
  243. 10:45issue, we need to use specialized APIs,
  244. 10:48and in this case, we could use several
  245. 10:51of them. The Serper API, which allows
  246. 10:55for real-time search, and the Tavily
  247. 10:58API, which allows for searching and
  248. 11:01returning well-structured data. The
  249. 11:05Twitter X API and other specialized
  250. 11:07APIs that will allow us to perform
  251. 11:10instant searches to see if a restaurant
  252. 11:13has an online presence. Now, the
  253. 11:17problem with this approach is that for
  254. 11:19an application like this, you would
  255. 11:22typically need at least 5 different
  256. 11:24APIs to ensure the most optimal search
  257. 11:27possible. And what you would have to do
  258. 11:31is go to each API's site, retrieve the
  259. 11:33documentation, understand how it works,
  260. 11:35and integrate it into the application.
  261. 11:38That is, therefore, the first
  262. 11:39constraint. If we need five API keys
  263. 11:42for our application, we have to go to
  264. 11:45five different platforms. You have to
  265. 11:48create accounts, generate API keys, and
  266. 11:50read five sets of documentation to
  267. 11:53understand how to call these different
  268. 11:56API keys in our application. The second
  269. 11:59major constraint is cost tracking. Five
  270. 12:03platforms means five different ways to
  271. 12:05track costs. And that can quickly turn
  272. 12:08into a mess in terms of cost monitoring
  273. 12:10. And personally, to address these two
  274. 12:13issues, the solution I use daily is the
  275. 12:16One API platform. Okay, you simply go
  276. 12:19to Google. Search. There you go, hit
  277. 12:25enter, and open the first link. Okay,
  278. 12:29so I’m going to quickly introduce you
  279. 12:30to this platform. It is a platform that
  280. 12:34aggregates various API keys so that,
  281. 12:37when you develop your application, you
  282. 12:40only need to provide a single API key
  283. 12:43—your only API key—and you provide
  284. 12:46that API key to your application. And
  285. 12:51this API key is connected to a set of
  286. 12:53tools, namely APIs (there are over
  287. 12:565,000) connected, tools (there are over
  288. 12:5940), LLM models (there are over 110),
  289. 13:01and payment methods. So, to enrich my
  290. 13:06application, what am I going to do?
  291. 13:09Since it is currently limited by the
  292. 13:12use of the Anthropic API key which can
  293. 13:14only use web search, which is very
  294. 13:17limited, what we are going to do is
  295. 13:19generate an API key for API EA, which
  296. 13:22is automatically connected to a set of
  297. 13:25APIs. I can show you here the various
  298. 13:29APIs that are aggregated. There are
  299. 13:31more than 5,000 of them. For example,
  300. 13:33we have Twitter here, which is super
  301. 13:35interesting when checking an company's
  302. 13:37presence. We have Reddit, super
  303. 13:38interesting when doing information
  304. 13:40research. We have YouTube, we have
  305. 13:41Tavily search. That is for retrieving
  306. 13:44information from the internet in a
  307. 13:46structured way. We have financial
  308. 13:49information, we have Instagram to know
  309. 13:50if a company has a presence on
  310. 13:52Instagram. Uh, we call it Perplexity
  311. 13:55for in-depth research. So personally,
  312. 13:58AISA is really a platform that allows
  313. 14:01me to overcome this major issue of
  314. 14:03online research and also the multiple
  315. 14:06use of different API keys. So to use it
  316. 14:10, it's very simple. OK? The first thing
  317. 14:13to do is simply create an account. OK?
  318. 14:15You click there on get started. I was
  319. 14:17already connected. So it automatically
  320. 14:19opens a dashboard. The first thing to
  321. 14:22do is to create an API key. So you will
  322. 14:25click there on new API key. OK. So I go
  323. 14:29to API keys and I can see there I had
  324. 14:32already created an API key. OK, so once
  325. 14:35you create your API key, how are you
  326. 14:38concretely going to use this platform?
  327. 14:42You go to get started. OK, here you
  328. 14:45have the three steps to follow. Really
  329. 14:47very simple. Personally, I find this
  330. 14:49platform to be a real game changer for
  331. 14:52people who want to build tools that use
  332. 14:54different API keys from different
  333. 14:56platforms. It is really super efficient
  334. 14:59for tracking. Everything is centralized
  335. 15:02with a single API key and it is just as
  336. 15:05interesting for cost tracking. So here,
  337. 15:09you will simply go to your usage log
  338. 15:10and you can see here the costs and
  339. 15:12token usage. No need to look at five
  340. 15:15platforms at the same time to be able
  341. 15:16to evaluate your costs. So you will
  342. 15:18come here to get started. OK, I will
  343. 15:21copy this command and I'm going back
  344. 15:23into Cloud Code. I will ask it to
  345. 15:26enrich my application using Ace One.
  346. 15:29enrich my application using Ace One and
  347. 15:38particularly use two tools, namely
  348. 15:42DataForSEO and Apollo. DataForSEO will
  349. 15:48allow for a complete search of the
  350. 15:50internet based on certain keywords and
  351. 15:53Apollo will allow for finding leads. So
  352. 15:56we are going to specify that. Data, for
  353. 16:03and Apollo please. OK, I'll put a dot
  354. 16:08and paste the previous command. Once
  355. 16:11that's done, I'll execute it. I'm
  356. 16:15running it and waiting. Cloud code is
  357. 16:19now getting to work. And there, it
  358. 16:21clearly tells me, I will first save the
  359. 16:24new key and read the documentation. to
  360. 16:28understand how to use DataForSEO and
  361. 16:31Apollo via Eisa. So we wait. It has
  362. 16:35retrieved the doc, it is currently
  363. 16:38retrieving the doc. So it tells me I
  364. 16:41have a clear picture of the available
  365. 16:43APIs. DataForSEO for business listings.
  366. 16:46But I'm not localizing well, but Apollo
  367. 16:48is excellent for prospecting. Here is
  368. 16:50my integration plan. Apollo mixed
  369. 16:53company search. Find companies by
  370. 16:56sector plus location. Apollo
  371. 16:59Organizations Enrich to enrich each
  372. 17:02company. Description, headcount, phone,
  373. 17:05network. Perplexity sonar existing.
  374. 17:08Let's find Google ratings, reviews,
  375. 17:10additional info. Tavily search existing
  376. 17:14web presence, Cloud, which will do the
  377. 17:17analysis plus email generation. So now
  378. 17:20it's starting the implementation. OK,
  379. 17:22perfect. So the other code has
  380. 17:23finalized. the registration of the
  381. 17:26application, it tells me the Apollo
  382. 17:28integration is working perfectly. OK,
  383. 17:31so we're going to test this. We'll go
  384. 17:34to Google within the application. We're
  385. 17:37going to redo the search. We'll put in
  386. 17:40a restaurant in the city of Paris.
  387. 17:43We'll look for one. We'll launch the
  388. 17:46search. OK. So what is it telling me?
  389. 17:49Company search with Apollo plus
  390. 17:51Perplexity. So Apollo is a really
  391. 17:55powerful tool when you need to find
  392. 17:57leads on the internet. So it tells me
  393. 18:01searching web presence using Tavily,
  394. 18:03analyzing five websites. OK. OK.
  395. 18:07Perfect. The search is finished. So the
  396. 18:11application preselected a certain
  397. 18:14number of companies and finally
  398. 18:17selected the restaurant Crocus Paris.
  399. 18:21OK. So the restaurant Crocus Paris.
  400. 18:24Here we have the address, the phone
  401. 18:27number, we have the website. Uh, so the
  402. 18:31priority is medium. So in terms of
  403. 18:34findings, the site is built on the Wix
  404. 18:36platform. OK. And the email sent. Hello
  405. 18:41, I am contacting you regarding the
  406. 18:43restaurant Crocus located on Rue Didot
  407. 18:46in the 14th arrondissement. While
  408. 18:49checking your online presence, I
  409. 18:51noticed that your current site is
  410. 18:53hosted on a generalist platform, Wix,
  411. 18:55which often limits customization and
  412. 18:58can provide a less distinctive image
  413. 19:00than a bistronomic and Mediterranean
  414. 19:03cuisine like yours deserves. At Nexi,
  415. 19:06we support independent restaurants with
  416. 19:08the creation of custom websites
  417. 19:10designed to reflect their culinary
  418. 19:12identity and facilitate customer
  419. 19:14contact. Menu, ambiance, reservation.
  420. 19:18Would you be available for a quick chat
  421. 19:20in 15 minutes to discuss what could be
  422. 19:22envisioned for Crocus? Best regards,
  423. 19:25Nexi, and my email address. So that is
  424. 19:27perfect. So here, typically, I can copy
  425. 19:30the email by clicking here, copy the
  426. 19:34email, and go to any email tool to send
  427. 19:37my message. With a tool like this,
  428. 19:40frankly, you can actually automate your
  429. 19:44client search. And here, I started with
  430. 19:47a tool where I have three parameters.
  431. 19:50The industry sector, the city, and the
  432. 19:52number of leads. You can add as many
  433. 19:55criteria as you wish, for example, the
  434. 19:57number of company employees, revenue,
  435. 20:00etc. I assure you, with such a tool,
  436. 20:02you can truly automate your prospecting
  437. 20:04. So, rather than spending hours every
  438. 20:07day manually searching for clients on
  439. 20:08the internet, you launch your tool. The
  440. 20:11tool loads clients for you
  441. 20:13automatically, and you can go further
  442. 20:15by connecting an automated email
  443. 20:17sending system. So it is truly a game
  444. 20:19changer. So definitely do not hesitate
  445. 20:23to test it on your end. I will put all
  446. 20:26the resources in the description that
  447. 20:28will allow you to reproduce this AI
  448. 20:30prospecting agent. I assure you, once
  449. 20:33you have managed to develop such an
  450. 20:35agent, you will have the ability to
  451. 20:37automate your client search. Now,
  452. 20:41speaking of the different resources, I
  453. 20:43will send you the Claude MD file, okay?
  454. 20:47Which represents the memory of your
  455. 20:48project. Then, the Prompt Claude Code
  456. 20:51file, okay, which will allow you to
  457. 20:53specify your instructions to let Claude
  458. 20:56know exactly what it needs to do. So,
  459. 20:59there is the possibility to customize
  460. 21:02these two files based on your company
  461. 21:04and its characteristics, and also what
  462. 21:07you wish to use for the prospecting
  463. 21:10application. In addition to that, in
  464. 21:13the description, I will also add the
  465. 21:15URL for the AI One platform. So, to use
  466. 21:18it, it is very simple. You simply type
  467. 21:20EASA. Do not hesitate to give me
  468. 21:25feedback on your use of this platform.
  469. 21:28It aggregates thousands of APIs. So you
  470. 21:30can truly use it for all types of
  471. 21:33projects. And what is really very
  472. 21:35practical with this platform, besides
  473. 21:37the fact that it aggregates several
  474. 21:39thousand APIs, is that you have the
  475. 21:41possibility to track your various costs
  476. 21:43on a single platform. So if I go here
  477. 21:47into my dashboard, okay, you can see
  478. 21:50the number of API keys I have, the
  479. 21:52requests, and the expenses I made today
  480. 21:55. Okay? So for the search I did earlier
  481. 21:59, it cost me 24 cents. Okay. So if I go
  482. 22:03into usage logs, I can see here for
  483. 22:05each of the APIs used in my application
  484. 22:08how much it cost me. This is a real
  485. 22:12game changer because it will actually
  486. 22:15allow me to compare different APIs. If
  487. 22:19I find that one API is too expensive
  488. 22:21compared to another, I can therefore
  489. 22:23use the ones that are cheaper. So here
  490. 22:26for example, I can see for the search
  491. 22:28from earlier, here are all the APIs
  492. 22:30that were launched. So here, I can see
  493. 22:32right here tav extract, how much did
  494. 22:34that cost me? I go all the way to the
  495. 22:37right, it cost me 0.054. OK. Uh, Tavily
  496. 22:41search 0.03. So this is super
  497. 22:46interesting if you are a business and
  498. 22:48you want to track your various costs in
  499. 22:50an ultra-precise way because for a
  500. 22:52company in the service sector or in the
  501. 22:54SaaS industry, you need to be able to
  502. 22:57save every penny. So, being able to
  503. 23:00track the different APIs you use for
  504. 23:02your applications on a single platform
  505. 23:04is a real game changer. If you found
  506. 23:07the video useful, don't forget to leave
  507. 23:09a like, subscribe so you don't miss the
  508. 23:11next tutorials on Claude Code and agent
  509. 23:13creation. See you very soon for a new
  510. 23:16video. This was Etel. Peace out.

About this transcript

This page contains the full transcript of Claude Code + AIsa : j’ai créé un agent IA qui automatise ma prospection commerciale by Ethel Cédric, generated from the public captions YouTube serves with the video. The transcript has 3,468 words across 510 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.