{
  "video_id": "iltdFNpl73I",
  "channel_slug": "andrew-codesmith",
  "channel_handle": "Andrew Codesmith",
  "title": "900+ hours of Learning Claude Code/Cursor in 10 minutes",
  "duration_seconds": 593,
  "url": "https://www.youtube.com/watch?v=iltdFNpl73I",
  "upload_date": "20260126",
  "transcript": "All right, so this video is going to be\nthe anti-vibe coder's guide to building\nwith AI, and it's basically going to be\na lot of things I wish I knew earlier\nabout using Claude code and cursor. So,\nhacks and a few things I've not seen\nonline that have really helped me when\nI've been using AI to code. Because\nthere's two ways, I think, that you can\nuse AI. You can be like a pigeon that's\njust had brain surgery, and you're just\nlike, \"Continue, continue.\" And then you\nget all these errors and you have to\nkeep restarting your project or rolling\nback. Or use it in a more considered way\nand get the AI to work for you. Because\nthese LLMs, basically, they're just\nguessing. The human's job is to make\nsure those guesses are better. So, in\nthis video, I'm going to show you how I\ndo that. And I am a senior software\nengineer. So, over the last year, I've\nseen my role just totally change to the\npoint where this has become like one of\nthe main skills people have to learn now\nif you're an engineer. So, things like\nprompt and context, like memory, and how\nto basically use AI effectively. Or if\nyou're watching this and you just want\nto make a better app and just spend less\nmoney on tokens, these are hacks and\nthings I wish I knew earlier. So, let's\nget stuck in. Now, I make content for\ndevelopers, programmers, people who work\nin tech, and a lot of people get pissed\noff when you talk about AI. They feel\nthat you become stupid by coding with AI\nand that it's an either/or situation. I\nthink you can use AI, do it quickly,\nlike build faster, but also learn at the\nsame time. So, later on in the video,\nI'm going to share how I do that because\nI thought about this loads. All right,\nbut let's start with the number one\nmistake I made and I see people making\nwhen they're using Claude and cursor to\ncode. And that is not using plan mode.\nBecause when I started coding, you wrote\na lot of the code manually. So, you\ndidn't have to plan too much. You just\nwrite a few notes and then build out the\nfeature.\nBecause you had a lot of control and you\nwere doing it step-by-step. [music]\nSo, basically, you couldn't it up\ntoo much. With AI, it's a lot different\nbecause it's like [music] a train, and\nit codes so fast, it can go in the wrong\ndirection really quickly. And the way\nyou like circumvent that is focusing a\nlot on getting the AI to plan the\nproject for you. And it really can feel\nlike magic seeing this LLM just like\ngenerating all this code. The reason why\nthat it's better to get the AI to do it\nwith plan mode, with things like Ultra\nThink, get it to plan out the project\nfor you, is that you can go back and\nforth with the AI, and also it will just\ndiscover edge cases and side effects\nwhich you not thought about. So, things\nwhich will pop up, and you'll be like,\n\"Oh, that's actually a good question.\"\nSo, it will just save you a lot of\nmoney. So, start every project, make\nsure you plan it. Now, the next one is\none I struggled with for ages, and I\njust put it off for a while because I am\na software engineer, and I like typing,\nbut it's better to just dictate your\nprompts. So, to speak, use the voice\ncontrol feature, and then just talk your\nvoice prompts rather than typing them\nout. [music] And the reason why it's\nbetter to do voice prompts, cuz I've\ngone back and forth with this, but I've\njust noticed when I do voice prompts,\nwhen you're typing, you're kind of\nlimited, and it's time-consuming. With\nvoice prompts, they're usually longer,\nI've noticed for me, they're more\nspecific, I add more context, and also\nit's just a bit more human as well. So,\nthis gives the AI better instructions\nthan these like shorter typed prompts.\nSo, when you can, obviously, if you're\nin a cafe or whatever, it's a bit weird,\nbut when you can, just try out doing\njust dictating your prompts and see how\nit goes. If you're building an app in\n2026, then you want to set up some\nautomations on a platform like N8N. And\nit's best to self-host these. That way,\nyou'll have control of your workflows\nand your data. And you don't have to pay\ncloud pricing for something that you can\njust run yourself. And you can do that\nwith hosting a VPS. So, we're sponsoring\nthis part of the video. N8N is\naddictive. And if you're a programmer\nand you're not making automations,\nyou're missing out. There is a lot of\nmoney to be made making these right now,\nand it's fun. Let me show you. So, I\nwant to send one WhatsApp message like\nthis, and the automation does the rest.\nIt all starts with this, the trigger, my\nWhatsApp message. That message is\ntranscribed, then the AI detects my\nemail intent, identifies a contact and\nthe Gmail thread, generates my reply,\nand sends the email automatically. And\nthen, boom, it sends me a WhatsApp\nconfirming it. Then, you got to host it.\nGo to this URL for hosting a VPS. It's\nup to four times cheaper, it's fast,\nunlimited workflows, and one click and\nyou're done. They've even got a sale on\nnow. You'll get 10% off with this code.\nTake this as your sign, get involved in\nthis AI automation error.\nOkay, next is to use MCP servers. And\nthe best way I found to think about MCP\nservers is think of it like a USB port\nfor AI to connect to the real world. So,\nyou can use it to pull in like live\ndata, APIs, and documentation into your\nproject. And a really good one is called\nContext 7. And what this will do, it'll\njust pull in all up-to-date\ndocumentation for you. And if you've\nlike coded professionally, you've worked\non a professional code base, one of the\nbiggest pain in the\nis when, for example, in JavaScript,\nthere's a lot of packages, and when one\nupdates, it can like affect others, and\nparticularly stuff like routing. So, MCP\nservers will give your project just\nbetter context. But, there is some\ncaveats here. You want to be careful\nabout what you're bringing into your\nproject. For example, there's an amazing\nSupabase MCP server, but you're giving\nthat a lot of access to your project,\nand in theory, it could delete and tank\nyour whole database and schema, but it's\nprobably worth the trade-off. So, yeah,\njust be careful about the access you're\ngiving them, but long story short,\nbetter context equals better code. Okay,\nnext one, treat AI like a junior\ndeveloper with ADHD. Let me explain, no\ndisrespect to anyone with ADHD,\n[music]\nand but I think it's just like a nice\nlike metaphor and it does make sense.\nIt'll do things really quick and really\ngood as well and you're just like,\n\"Whoa.\" You just generated a load of\ncode, a feature, and it's just like\namazing. That's exactly what I wanted.\nOther times it'll do things really\nquick, but you've got to like\nguide it a lot. You've got to You need\nto constant guidance, really tight\ncontext, and also just give it specific\ntasks. Otherwise, it will guess, but it\nwill guess really confidently and it\nwill take you to like error hell. When\nyou think about it, how LLMs work is\ntheir job is to please you by trying to\nfill in the blanks. And your job in the\nwhole of this, your job as the human is\nto try and make those blanks better. And\nby using MCP servers, you give it more\ncontext, and thus you try and make the\nblanks better. Okay, next, if you're a\njunior developer and you're watching\nthis and you're like, \"I like coding\nwith AI, but I'm worried that I'm losing\nmy coding ability.\" Or maybe you're like\na senior and you're just in the same\nboat where you're like, \"I don't think\nI'm learning as much. Maybe I'm losing\nsome of the craft of coding.\"\nThis is the mindset I would have.\nObviously, if you're building your own\napp and your goal is to build an own\napp, this this isn't important, I'd say.\nThe way I treat it is like pair\nprogramming and treat it like a\ncollaboration. So, pair programming, if\nyou don't know, I've spent probably\nthousands of hours of my life pair\nprogramming, and it's where you buddy up\nwith someone who's usually like senior\nand just way better than you at the job.\nYou get a problem together, you share a\nscreen, and then you go back and forth,\nand usually they code a lot, you ask\nquestions, and you just watch how they\ndo things. If you really want to learn\nand grow, then treat coding with AI\nusing Cursor and Claude like this. So,\nyou're asking questions. You're like,\n\"Why did it do that?\" This is going to\ntake longer, but you're going to learn\nmore. So, it's that balance of being\nproductive, but also learning. This is\nWith AI now, this is, I think, massive.\nIf you're working for a company, don't\nsacrifice your learning for their\nproductivity. And another really\nimportant and helpful concept is this\nidea of compound engineering. So, like\nmoney, your codebase and your AI should\nget smarter and smarter with each prompt\nand each iteration. So, a big part of\nyour role as the human being in all this\nis to just give it as much context to\nallow it to compound engineer as a\nproject goes on. So, by the end of it,\nit knows what you want, it knows what\nwhat you want the project or the app to\nlook like. And a big way you can do this\nis by providing and using Claude MD and\ncursor rules. So, what kind of stuff do\nyou put files? The way I think about it\nis like documentation for your app. So,\nstuff that you keep repeating again and\nagain, like themes, how you want the\ncomponents and stuff to look. If you\nfind yourself repeating it again and\nagain, just put it in these files. Start\nbasic and then grow from there. And\nwhenever your AI is generating code,\nit's going to refer to these this\ndocumentation in the context. You want\nto be careful though. You You don't want\nto put too much in there because it'll\nblow up the context window. Anything\nthat you think will help, anything that\nyou're repeating again and again, put it\nin these files. So, you're helping\ncompound engineering by using Claude.md\nand cursor rules. Okay, the last one,\nthe last hack I really like, it's for\nanyone who's designing an app and they\njust suck at design. Because what you\ncan do, you can actually just steal\ndesigns and use AI to implement it. And\nactually works pretty well, I found. So,\nthere's a website called Mobbin. It's a\ngood way to like browse different\ndesigns. If you see one you like that\nyou think's cool, take a screenshot of\nit. Go back to Claude or Cursor and just\nsay, \"I want you to update the design of\nthis app using this image.\" Then drag it\nin, but give it a constraint. Tell it,\n\"Only use the color scheme and the font\nfrom this design.\" That's it. Otherwise,\nit's just going to like change loads of\nstuff. But I did this the other week and\nit worked really well. And that is it.\nIf you enjoyed the video, if you got any\nvalue whatsoever, all I ask is free for\nyou, just like the video if you want and\nsubscribe if you enjoy this kind of\ncontent. I've also got a newsletter\nwhere I'm going to be posting like\ndeeper insights into the app I'm making\nright now using AI. So, subscribe for\nthat. I'll see you in the next one.\nHappy coding. Ciao.\n[music]",
  "transcript_chars": 10900,
  "ingested_at": "2026-05-21T19:33:35.961602+00:00",
  "source": "discover",
  "yt_meta": {
    "view_count": 459654,
    "like_count": 20973,
    "channel_id": "UCcgNiiErDGHICbNYtuyxW4A",
    "categories": [
      "People & Blogs"
    ],
    "tags": []
  }
}