Claude Advisor Agent: tous les développeurs doivent le tester !

I just installed the latest Claude update, and it's reduced my token consumption by two-thirds. The new Intelligent Boost Agent system is simply brilliant. You absolutely must install this latest update. I'll explain how it works, which models it's compatible with, and the performance gains you'll experience. The idea is simple: can I use Claude to work completely autonomously and save even more money? The clever trick Anthropic has implemented is that Claude is taking control of my screen. But the exceptional aspect is n't Claude Opus, it isn't Claude Sonnet. No, it's the cheapest model, the one that costs less than 20 cents. In short, we can now configure the Advisor system with a new architecture that will allow even the smallest models to be among the best. And that's how I got Claude to take control and developed a Dark Web version of France Travail. I'll tell you all about it in this video. Let's focus on the new Claude Advisor Strategy system. It's as if your agent can call upon an orchestrating agent, Claude, to help them complete tasks. This is the new tool developed by Anthropic to allow Claude, Claude Sonnet, Claude Haiku, or Claude Opus to rely on a very simple agentic system to set up. Simply activate the Advisor Tool function, and the system will automatically call Claude for assistance. This, of course, marks the beginning of using the new Claude Mythos architecture. You'll have an assistant available who can help you remotely, take control of your agent, and give them instructions for their work. The first step, of course, is to install the Chrome extension and activate it on the interface. With a simple click, you return to your terminal, type /mcp, and check that Claude in Chrome is connected. If it's not connected, press Enter and perform a Reconnect function. Our goal is to use the most cost-effective tool possible and make it work as efficiently as possible. This new combination will allow us to do just that. To achieve this, you'll select the Claude Haiku model from the template system. Press Enter. From now on, you'll use the new Advisor command. From there, you'll choose the model that Claude Haiku or Claude Sonnet will call when it needs strategy and decision-making support. Press Enter. And now, every time you work, you'll automatically pair the model with the new Advisor interface, which will call Claude Opus, likely to be replaced by Claude Mythos in the next update . So we're at the beginning of a new way of working, a system that will be overseen by an AI agent that will provide him with advice on how to work. From now on, we'll be able to transform Claude Haiku into a true AI agent. I asked him: "Go find me some data engineer job postings in Paris. You'll extract the data and information and organize it all into a CSV file, making sure to list the highest salaries first." The model is running completely automatically and extracting the data for me. In short, we've converted the lowest-end model into one of the best. He has all this for three times less than Claude Sonnet. The goal? To optimize our $20-per-month subscription so we can work even better with Claude. And as you can see, I'm not doing anything. He's navigating, extracting, retrieving, and formatting all the data for me. But today, what's truly incredible is that we can take its operation even further. While he's working—and he works quite fast—he generated the two files that we're going to... Let's analyze. First observation: well, Claude Haiku did give me a little glitch. It put this in CSV format, but it's not very readable. So what I'm going to do is use the Claude extension function within Excel, and simply with a prompt to reformat the columns and layout, in less than 5 seconds, all the data is reformatted. You have the option to activate a function called automated modification permission. This will allow you to take control of Claude even within the CSV and make all the necessary modifications. Now that we've proven we can transform Claude into an AI agent system, even with the most basic model, would it be possible to have it perform a automatic mode so it can perform all the connection tests for the coded web page ? In short, I want it to code for me and even run the tests for me. That's exactly what we can do by adding a few extra functions. If you want to learn the best of AI, the description explains how to master the best AI in 2026 and integrate Claude AI agents and workflows into all your businesses. Even if you're a beginner, this training is designed to give you the best professional practices used in entrepreneurship to integrate agentic systems. What you saw is that I made a request with the /opus-plan function. By using the opus-plan function, I'm asking Claude Opus to generate a plan and enabling it to perform agentic tasks. This means that a very small and simple model is now capable of executing the most complex work in companies. But what if I told you it's possible to do even better in terms of agentic architecture? That's what I'm going to reveal in this video. There's the possibility of combining lines of code that will activate the Advisor function even more effectively. And that's what I'm going to show you. The Advisor function is now active by default in my system. So I have my Claude Sonnet system in the medium position, which will be overseen by a Claude Opus 4.6, and this will be the system that plans the entire writing process for my web page. In concrete terms, this means you have optimized combinations between AIs. You can use Claude Haiku and put it in a combination system with a Claude Opus 4.6 or Claude Sonnet, running with supervision by a Claude Opus 4.6. The advantage is that I'm going to tell it: "OK, I have the France Travail page, now I want you to make a Dark Web version of it." The idea is to have the entire directory structure. So I'm going to tell him which directory and how I want the model to develop. This is different from what you've seen previously in all those online videos about launching a project... We're going to make our first website together: "Create a portfolio for Melvin. Find information about him, everything you can, and create a modern portfolio." What is "information about him"? What is "everything you can"? What is "a modern portfolio"? What relationships should the model understand between the information blocks? We 'll specify the tech stack. If you're not a developer, I recommend using Vite.js and TypeScript. So you can just use these keywords: Tailwind CSS, React, and that's it. And you can—I think this is the only correct thing I've seen so far— you can also say Shadcn/ui if you're unsure. So: "Create a portfolio, use..." and here you can maybe type $. I don't know if we can easily add skills. Okay, to create skills, we'll use the $. And here, we can choose "frontend skill". So there, That's the one we added just before. "Use frontend skill to have a modern interface, a minimalist portfolio that represents Melvin. Full dark mode, primary blue, simple and use original, use common." I'm going to ask you another question: what is a "modern interface"? What is a "minimalist portfolio"? What is "full dark mode"? I'm fine with us throwing things out there like that. The model hasn't understood anything anyway, but he'll do something. He'll produce something. But that's really what we shouldn't do. What we need to do is provide the file architecture. Claude is going to help us set it up. We'll give him a color palette, the font types, the libraries we'll use in the interfaces. And from there, as I told you, 98% of the work is done. Quite honestly, the added bonus is that we're simply going to tell the AI: "When you're finished, you're going to run a series of tests for me." We define the tests: the buttons, the recording, the page scrolling. We'll even get a final report. And we could have done this with Claude Opus, and it would have cost us six times as much. But with the /advisor function, well, it really costs us next to nothing. What's happening is the shift from traditional work to an orchestration system. The way to use artificial intelligence today isn't through the prompts you've seen on social media, but rather the ability for each of us to be an architect and model our work to actually delegate it to the AI. That's exactly what the biggest American companies are doing today. And that's why you absolutely must integrate it, because it will only take Claude 10 minutes, by switching to the Claude Haiku model, to complete the entire task by regularly making calls to Claude Opus. So imagine, as soon as we're able to install the latest version of Claude Mythos, this is definitely the new way of working that will become essential: the ability to structure calls to an AI that will complement your workflow. You 'll see, just over here. We've given Claude Haiku the green light to code the page , and it will create all the directories as we initially asked Opus Plan to plan. We have the entire strategy plan. All that's left is to let Claude Haiku set up all the code for about ten minutes, and you'll even have all the assets, HTML, CSS, and JavaScript pages in place, and you'll have the whole suite of tests. You have index.html on the local port. You click on it and it's simply brilliant. It created a Dark Web version of France Travail for me. Take a look, and at first glance, it works perfectly, and it ran all the tests for me. This means that Chrome's new feature, combined with the Advanced function, will allow you to execute complete workflows. You code, you let the AI ​​check that the application is working, it gives you a report, an audit, it fixes bugs, it tests them. In short, I think the developer's job is completely changing, and you'd be wise to learn how to use this new feature. But in any case, what you absolutely must do right now is integrate the new /advisor function into all your workflows. It increases Claude Haiku's capabilities by 50%, and it reduced my token consumption by two-thirds. Let me know in the comments if you've already installed it. Do you use Claude CLI? If you 're not using it, let me know in the comments. I'll make a tutorial on how to set it up and use it because, honestly, I've never seen an AI as powerful as Claude. And if you're still using Claude Cowork and you're a business, you absolutely must Switching to the latest Claude Code CLI version with Advisor features is simply the future of AI taking shape. So, are you going to be part of the game?