context of a coding AI agent like cursor or windsurf which are AI powered code editors that have a agent mode that can autonomously perform coding tasks either with claude sonnet 3. 7 or Gemini 2. 5 pro. Another very common AI agent use case are customer service chat bots. Many companies are now experimenting with customer service agents that
Поиск по транскрипциям
show you exactly how to walk through it. All right, let's talk about what just happened in the AI world. Anthropic made a move. They locked down Claude Code, said you can only use their model in authorized interfaces. Translation: Use our terminal. follow our rules, use it our way. And look, I get it from their side. They...agent like having a company expert that actually sites sources. Third, a data to app pipeline. You paste a CSV, Claude builds you a dashboard. Real stuff, stuff you can use today. Let's start with what Open Code
government, not a company, the actual government, tax records, voter data, employee passwords, all of it. But here's the thing. He didn't write the code himself. He opened Claude AI and asked it to help him break in. And Claude said, "No, I can't do that. It goes against my safety guidelines. " Story over, right? Nope...problems. " And that's when everything changed. Claude said, "Oh, okay. I'll help. " Now, stop here for a second because what happened next is the part that should genuinely scare you. The AI found over 20 security holes in government systems. It didn't just find them. It wrote the attack code. It automated the entire data extraction
Claude code in both your web browser. So, I'm in Google Chrome right here as well as the Claude mobile app. So, I just have my iPhone simulator open here and it has the Claude app here. And what you're able to now do is connect to any of your GitHub repos. So anywhere your code lives
bench. By December 2025, Claude scored 80. 9%. That's not a typo. 4% to 80% in under two years. Cursor hit a $9 billion valuation. Claude Code is doing over 500 million in run rate. Prediction three, China maintains competitiveness. January 2025, Deepseek releases a model matching OpenAI's best trained for $6 million. OpenAI allegedly spent over
Google proof questions Live code generation basically anything you throw at it but here's something fishy that caught my attention they didn't compare it to Claude 35 Sonic in their coding benchmarks and if you know anything about AI coding you know that's the model most developers are actually using right now speaking of benchmarks check this
mechanisms, Claude is learning an objective of how to do these things with basically limited guidance or feedback. - Exactly, we do lots of RL on coding tasks, on search tasks, lots of things for Claude to practice being an agent in different environments. - There's kind of this conception, I think of Claude models that they're really, really strong
abandon claude and switch back? Short answer, no. Unless you specifically need computer use. Here's the breakdown. Chat GPT 5. 3 Garlic use if you want fast blunt responses for surface tasks code snippets quick answers daily admin tradeoffs confident hallucinations labized creativity context failures chat GPT 5. 4 for thinking. Use if you need computer use, legacy software...productionready code, stable, ethical, fast, without performance theater, no surveillance, no weapons deals, tradeoffs, no computer use, no native image generation. For 90% of professionals, Claude is superior. Stability, trust, speed without wait, code you can deploy. Chat GPT 5. 4 has one killer feature, computer use. If you need desktop automation, it's your only option
Quen fixes. You run it. And the beautiful thing is you don't need to be a developer. know how to code. The AIS handle that part. You just need to know what you want to build. And Claude will help you figure that out, too. That's the whole point of giving it the architect role. It takes your...give Claude the architect role and ask it to plan your automation. Step two, paste that plan into Quen and tell it to build. Step three, send the code back to Claude for review. Step four, send the fixes to Quinn to apply. Step five, run it and watch it work. Five steps, two AIs, one automation that runs
other thing on Claude Code that has been another popular feature I've seen a lot of software engineers use lately is Claude MD files. So these are files that you, you know, define within a project and gives Claude relevant information about what your programming style is or like what the layout of the directories are, things like that...what they mean for Agents? - Yeah, so Claude Skills are a very exciting extension of Claude MD files where instead of just giving it notes files, you can give it any sort of file. That can be PowerPoint template files, it can be code and like helper scripts that you want it to use. It can be images or assets
Jensen Wong says it's the number one AI benchmark test he trusts. So, let's see how Cloud Code does here. All right, city flythrough's complete. Let's test this out. Let's see how Claude does. Will they be better than chat GPT? All right, so here we go. Here is the city flythrough from Claude...Some of the windows are off the buildings. That's a issue. Uh and we also go through the buildings, which isn't great. So, this is interesting. Claude Code can't handle the camera too well either. It's struggling with the camera. It is also kind of making me sick. But from a city perspective, a couple positives
Because they're all kind of like at the same level and we've all got like these different ways that we can use autonomous AI agents. So claude code is probably the best solo coder, right? So if you need one AI to write amazing code, this is probably it, right? It runs from your machine, understands your codebase
with some quick scripts Claude 3. 7 Sonic can handle it all with Incredible accuracy and the improvements don't stop with coding Claude 3. 7 Sonet also excels at content generation data analysis planning and more it can understand nuanced in instructions and context recognize and correct its own mistakes and create sophisticated analysis from complex data
Studio Code, free to use. Download it link below. Uh and then I'm going to open up the Claude plugin which is also free. So you can get that installed and start using it. What we're going to do here is we're going to start using Haiku but we're going to use it strategically. So what...second, then I'll show you how good H high coup is at building this out with one shot. But this is how you want to start using Claude Code moving forward. As long as you're not doing something incredibly, incredibly intensive. But if you're just a hobbyist or you're just building
then finally deliver with AI. So this is a whole other thing. Go watch all the videos on this channel. Claude for copy. Gemini for research, brainstorming, Sora V3, cleaning for videogen, uh, NAN for automations or cloud code if you're semi-technical. Now, the process here is you want AI to do 90% of the work. You provide...guys. It's January. Well, it's February now. It was just my birthday. It's February. January went by like that. In January, we got Remotion, Claude Code, Claudebot, which is now OpenClaw. Uh, we got new models. We got Opus 4. 6. We got a new Gemini model. codeex from or from OpenAI. And I'm probably missing
whole bunch of other newsletters I wrote, which is amazing. And what it does is it looks to see how you write. It looks at all the code in your app to see how your app works and then it does research online to find citations just in case it needs to reference other articles online. This is way different...going to write this blog for me. It's going to be excellent as a spoiler. And the difference between doing this normally with chat GBT or just claude code without any skills and doing what we're doing now is the skill describes the quad code exactly how to write compelling content, how to do research online and find
building out this factory. SAS factory. It's going to be sick. is Uh, I'm doing a live boot camp tomorrow. Live boot camp tomorrow in the Vibe Coding Academy. I'll go through all of this in the live boot camp. It's two hours long. You can join, ask questions, talk to me live, ask me questions...join now, you get access to all the other boot camps, recorded boot camps. You also get access to a bunch of video series, uh, social media series, Claude Code series, Open Claw series, whole bunch of series of videos. Link for that down below. Vibe Coding Academy. It's sick. Over 800 people in there. Um, we're back
videos and images. You can literally drop a YouTube video link and have it analyze the entire thing for feedback. I use this daily for content research. Claude is the king of coding and deep thinking. If you're building anything technical or need sophisticated analysis, Claude beats everything else. But here's what nobody tells you. You need
pretty crazy because what we have here is an AI system that is now even better than Claude 3. 7 Sonnet on coding tasks. And I've actually used it to code quite a few things. And I do notice this myself. So, it's pretty crazy that Google are now basically state-of-the-art in basically every single
your Deep Agent builds even better. First, always tell Deep Agent to use Claude Sonic for reasoning. That's Claude 3. 5 Sonet. Just add use Claude Sonnet for all reasoning and code generation to your prompts. Second, tell it to plan first. Say, plan the full architecture with a diagram before building. This saves you credits because Deep Agent