The Beast keeps on getting better.
Claude Sonnet 4.5 + Claude Code 2.0: Agentic Coding Masterclass
A comprehensive crash course on Anthropic's latest Claude Sonnet 4.5 and Claude Code 2.0, covering everything from basic setup to advanced features for AI-powered development.
Published Oct 3, 2025 by Leon van Zyl
Key Insights
Claude Sonnet 4.5 is positioned as the best coding model in the world according to Anthropic, with significant updates to both CLI and VS Code extension interfaces.
Claude Code offers different operational modes (ask before edits, edit automatically, planning) and thinking modes that can be toggled to optimize for speed or reasoning depth.
The system uses 'memories' to store context about projects and user preferences, enabling personalized interactions across conversations.
Advanced features like custom commands, checkpointing, and subagents allow for specialized workflows and improved productivity.
Claude Code can run background tasks and long-running processes while continuing other work, significantly enhancing multitasking capabilities.
Integration with GitHub enables remote development and asynchronous code changes without needing to be at your computer.
MCP servers extend Claude's capabilities by providing up-to-date documentation and tools for specific libraries like Shad CN components.
0:00
Claude Sonnet 4.5 Release and New Features Overview
“According to their article, Claude Sonnet 4.5 is the best coding model in the world right now. And in addition to the model, they also released a major update to Claude Code CLI and the new extension for VS Code.”
The video begins with an introduction to Anthropic's latest release - Claude Sonnet 4.5, which they claim is the world's best coding model. This update comes alongside significant improvements to the Claude Code CLI tool and a new VS Code extension.
Leon outlines that this comprehensive crash course will cover everything from basic setup to advanced features like subagents, MCP servers, custom commands, background tasks, checkpointing, and GitHub integration. He promises that by the end of the video, viewers will know exactly how to leverage Claude Code 2.0 with the new Sonnet 4.5 model in both new and existing projects.
Takeaways
Anthropic has released Claude Sonnet 4.5, positioned as the world's best coding model
The update includes major improvements to the Claude Code CLI and a new VS Code extension
The video will cover both basic and advanced features of Claude Code 2.0
The goal is to enable viewers to fully utilize Sonnet 4.5 in their projects
1:03
Setting up VS Code for Claude Code
“I recommend downloading and installing a code editor like VS Code. Of course, you can use Claude Code in the terminal without a coding editor, but I think it makes it way easier to use Claude Code within a code editor where you can see the files being changed.”
Leon begins the practical portion by recommending VS Code as the preferred environment for using Claude Code. He notes that while Claude Code can be used directly in the terminal, using it within an IDE makes it easier to visualize file changes.
After installing VS Code, users need to open a folder where they'll work with Claude Code. He explains that there are two main ways to use Claude Code: through the CLI tool or via the VS Code extension, and he'll demonstrate both approaches throughout the video.
Takeaways
VS Code is recommended for using Claude Code, though terminal-only usage is possible
The IDE extension appears to work in VS Code forks like Cursor
Users can choose between the CLI tool or VS Code extension depending on their preference
Using an IDE makes it easier to visualize file changes made by Claude Code
1:51
Installing the Claude Code VS Code Extension
“In VS Code, go to extensions. In the search bar, simply search for Claude Code and then look for this extension and install it.”
This section walks through the process of installing the Claude Code extension in VS Code. Leon demonstrates how to search for the extension in the VS Code marketplace and install it, which adds a Claude Code icon to the sidebar.
Once installed, clicking the Claude Code icon opens the conversation window where users can interact with the coding agent. The installation process is straightforward and follows the standard pattern for adding extensions to VS Code.
Takeaways
Search for 'Claude Code' in the VS Code extensions marketplace
After installation, a Claude Code icon appears in the sidebar
Clicking the icon opens the Claude Code conversation window
The interface is clean and integrated directly into the VS Code environment
2:18
Login and Authentication Setup
“Initially, we will have to sign into our Claude account. To do that, enter front slash and from this list, select login.”
This section covers the authentication process for Claude Code. Users need to sign in to their Claude account by typing '/' in the extension and selecting 'login' from the command list. Leon explains the two billing options available: signing in with an existing Claude Code subscription (which requires a paid plan) or using a pay-as-you-go approach with an Anthropic Console API key.
He suggests that the subscription option is likely the more economical choice for most users. After selecting the preferred billing method, users need to authorize the application by opening a URL and completing the authentication process, after which Claude Code becomes fully operational.
Takeaways
Authentication requires either a Claude subscription or an API key with pay-as-you-go billing
A paid plan is required to use Claude Code
The subscription option is generally more cost-effective than pay-as-you-go
After authorization, Claude Code becomes immediately available for use
3:08
Installing Claude Code CLI Tool
“In order to install Claude Code, all you have to do is run the following command: npm install -g @anthropic-ai/claude-code and press enter.”
This segment focuses on installing the Command Line Interface (CLI) version of Claude Code. Leon shows how to open a terminal in VS Code and run the npm installation command npm install -g @anthropic-ai/claude-code to globally install the tool.
He notes that Claude Code automatically updates itself, but if users want to manually update to the latest version, they can run the same command with @latest appended. After installation, users can simply run the claude command in the terminal to start using Claude Code in CLI mode.
Takeaways
Install the CLI tool with: npm install -g @anthropic-ai/claude-code
Claude Code automatically updates itself in most cases
Manual updates can be done by appending @latest to the installation command
Start the CLI version by simply typing 'claude' in the terminal
4:33
Windows vs WSL Installation Differences
“Contrary to popular belief, you do not have to use WSL to run Claude. I just ran Claude Code on my machine without the need for WSL. But I feel that I do have to mention that I actually get more functionality out of Claude Code in WSL, specifically on checkpointing.”
Leon clarifies that Windows users have two options for running Claude Code: directly on Windows or through Windows Subsystem for Linux (WSL). He dispels the myth that WSL is required, confirming that Claude Code runs on Windows natively.
However, he mentions that some advanced features, particularly checkpointing, work better in WSL. For those interested in using WSL, he suggests using ChatGPT to walk through the process of installing Ubuntu and setting up WSL, noting that it's relatively straightforward. After setup, users can install Claude Code in the WSL environment using the same npm command.
Takeaways
Claude Code can run natively on Windows without requiring WSL
Some advanced features like checkpointing function better in WSL
Setting up WSL with Ubuntu is relatively straightforward
The same installation commands apply in both Windows and WSL environments
5:09
Basic Claude Code Usage and Modes
“In the IDE, you can change the different modes by holding shift and pressing tab. You can see we now switch to 'accept edits on', which means Claude Code will make changes to the project without asking our permission to do so.”
This section explains the different operational modes of Claude Code. Leon demonstrates how to cycle through these modes by pressing Shift+Tab in both the IDE extension and CLI versions. He covers three main modes: 'accept edits on' (where Claude makes changes without asking permission), 'planning mode' (where Claude won't make any changes but can discuss solutions), and the default mode (where Claude asks permission for everything).
He clarifies that even in 'accept edits on' mode, Claude Code will still ask permission for certain actions like running commands or accessing the network. This gives users flexibility in how they want to interact with the AI assistant based on their comfort level and specific needs.
Takeaways
Switch between modes using Shift+Tab in both IDE and CLI versions
Accept edits mode allows Claude to make changes without asking permission
Planning mode prevents Claude from making changes, ideal for discussions
Default mode requires permission for all changes
Claude still asks permission for running commands or accessing the network regardless of mode
6:16
Bypass Permissions and Thinking Modes
“If you want Claude Code to run without ever asking your permission to do so, we can simply exit out of Claude Code. Then we can run 'claude --dangerously-skip-permissions'. When we run this, we're now in the special mode called bypass permissions mode.”
Leon introduces two additional modes that further customize Claude Code's behavior. First, he demonstrates the 'dangerously-skip-permissions' mode, which allows Claude Code to run commands without ever asking for permission - a powerful but potentially risky option for users who want maximum automation.
Second, he covers the 'thinking' mode, which can be toggled by pressing Tab. With thinking mode on, Claude provides more reasoning in its responses, while with thinking off, it delivers faster responses with fewer tokens for simpler tasks. The VS Code extension doesn't have a dedicated thinking toggle, but users can force thinking by adding phrases like 'think hard' to their prompts.
Takeaways
Use 'claude --dangerously-skip-permissions' for complete automation without permission prompts
Toggle thinking mode with Tab in CLI (not available in VS Code extension)
Disable thinking for faster responses on simple tasks
In the VS Code extension, add 'think hard' to prompts to force thinking mode
7:48
Adding Images to Prompts
“We can include images in our prompts in both the extension and the CLI tool.”
This section demonstrates how to incorporate images into conversations with Claude Code across both interfaces. Leon shows several methods for adding images: in the VS Code extension, users can reference files using the '@' symbol and selecting the image from the project, or they can paste screenshots directly into the chat window.
In the CLI tool, users have more options - they can drag and drop images while holding shift, reference images using the '@' symbol similarly to the extension, or paste screenshots with Alt+V. This feature allows users to ask Claude to analyze or describe visual content, expanding its utility beyond just code.
Takeaways
VS Code extension: Reference images with '@' or paste screenshots directly
CLI tool: Drag and drop with shift, reference with '@', or paste with Alt+V
Images can be from project files or screenshots from anywhere
Claude can analyze and describe visual content when provided with images
9:00
Managing Conversation History
“In order to go to a previous conversation in the extension, you simply click on this dropdown and here you can see all your previous conversations.”
Leon explains how to access and manage conversation history in both the extension and CLI versions of Claude Code. In the VS Code extension, users can simply click on a dropdown menu to view and select previous conversations.
In the CLI tool, users need to type '/resume' to see their conversation history and jump back to any previous conversation. This feature allows users to continue work from previous sessions or reference earlier discussions, making Claude Code more useful for ongoing projects.
Takeaways
VS Code extension: Use dropdown menu to access previous conversations
CLI tool: Type '/resume' to view and select conversation history
Conversation history is preserved between sessions
Easily resume work from earlier conversations
9:36
Understanding Memories and Project Rules
“An important concept in Claude Code is memories. This is how you can give the agent context about your project and tech stack as well as very specific instructions it needs to follow.”
This section covers the concept of 'memories' in Claude Code - a powerful way to provide persistent context to the AI. Leon demonstrates how to add memories in the CLI by typing the '#' symbol, followed by the memory content, and choosing whether it should be a 'user memory' (spanning all projects) or a 'project memory' (specific to the current project).
These memories are stored in a claude.md file and are automatically loaded into context at the start of new conversations. He shows how the agent remembers to refer to him by name after adding this as a memory. While the extension doesn't allow adding memories directly, it does respect the memories defined through the CLI, making this a crucial feature for guiding the agent's behavior, tone, and understanding of the project.
Takeaways
Memories provide persistent context and instructions across conversations
Add memories with '#' in CLI (not available in extension)
User memories apply to all projects, project memories are specific to current project
Memories are stored in claude.md file and automatically loaded into context
Use memories to guide agent behavior, tone, and project understanding
10:23
Making Changes to Existing Projects
“Let's set up a basic project. I'm going to open up a new terminal and run npx create-next-app@latest.”
Leon demonstrates how Claude Code can be used with existing projects by creating a simple Next.js application as an example. He sets up a boilerplate Next.js project using the create-next-app command and verifies it's working by running the development server.
He then shows how to use Claude Code to analyze the project by asking it to review the homepage for Shad CN components. In planning mode, Claude examines the code without making changes and reports back that the homepage isn't using Shad CN components. This section also introduces the ability to search previous prompts with Ctrl+R, allowing users to find and modify earlier commands rather than retyping them completely.
Takeaways
Claude Code works seamlessly with existing projects like Next.js applications
Use planning mode for code review without making changes
Search previous prompts with Ctrl+R to reuse or modify earlier commands
Claude can analyze code and provide feedback on specific aspects (like component usage)
11:59
Creating Custom Commands
“If you are planning to reuse a prompt in your project, then there's actually a better way than searching for it. What we can do is create custom commands.”
This section covers how to create reusable custom commands in Claude Code. Leon demonstrates the process of creating a directory structure (.claude/commands/) in the project root and creating a custom command file with a .md extension. He shows how to structure the command file with sections for description, goal, and user input, including the special $arguments placeholder for dynamic arguments.
After creating the file, users can access the custom command by typing '/' in Claude Code and selecting it from the list. The command appears with the description that was defined in the file, and any dynamic arguments can be provided when executing the command. This feature works in both the CLI tool and VS Code extension, allowing for standardized and efficient workflows.
Takeaways
Create a .claude/commands/ directory in your project
Command files use .md extension with sections for description, goal, and user input
Use $arguments placeholder for dynamic input in commands
Access custom commands via '/' in both CLI and extension interfaces
Custom commands standardize workflows and save time on repetitive tasks
14:05
Checkpointing and Rewind Feature
“Another really cool feature is the ability to create checkpoints. This allows us to rewind all of the changes back to a previous message.”
Leon introduces the checkpointing and rewind feature, which is exclusive to the CLI tool. He demonstrates this by having Claude make a cosmetic change to the website (changing the background to purple), then shows how to revert this change without using version control.
By entering the rewind command or pressing double Escape, users can access a message history with code changes tracked. When selecting a message, users get three options: restore both code and conversation to that state, restore only the conversation, or restore only the code. This feature provides a powerful way to experiment with changes and easily roll them back if needed, functioning as an integrated version control system within Claude Code.
Takeaways
Checkpointing is exclusive to the CLI tool (not available in VS Code extension)
Access rewind with the 'rewind' command or by pressing double Escape
Three restoration options: code+conversation, conversation only, or code only
Functions as an integrated version control system for experimental changes
Eliminates the need to manually track and revert changes through git
16:20
Background Tasks and Bash Management
“Claude Code can run long-running processes in the background - things like running the dev server, running build commands, extracting data, etc. that will free Claude Code up to continue with other work.”
This section demonstrates Claude Code's ability to run background tasks and manage multiple processes simultaneously. Leon shows how to instruct Claude to run commands like build processes in the background, allowing the main conversation to continue while those tasks execute.
Users can monitor background tasks by typing /bashes to see a list of all running processes and their status. For each task, users can view detailed logs and even kill processes that are no longer needed by pressing 'K'. This multitasking capability significantly enhances productivity by allowing users to continue working with Claude while resource-intensive operations run in parallel, with the agent maintaining access to logs from all background tasks.
Takeaways
Run resource-intensive commands in the background while continuing the conversation
Monitor background tasks with '/bashes' command
View detailed logs for any background process
Kill unnecessary processes by pressing 'K'
Claude maintains access to logs from all background tasks
18:10
Creating and Using Subagents
“Everything comes down to context windows. The more context you throw into a conversation, the worse the output is going to get. So, it's always a good idea to start new conversations with smaller and focused context. And that is kind of what sub agents do.”
Leon introduces the concept of subagents, one of Claude Code's most powerful features. He explains that the default 'general agent' handles everything, but this can lead to context overload. Subagents solve this problem by creating specialized agents that focus on specific tasks, each with their own context window.
Using the CLI tool, he demonstrates how to create a specialized subagent for writing Shakespearean-style text. The process involves running the agents command, configuring the system prompt, selecting available tools, choosing a model, and assigning a color for visual identification. When a task requires this specialization, the general agent can hand it off to the subagent, which completes the work with a clean context and returns a summary to the general agent. This modular approach significantly improves performance by keeping each agent's context focused and relevant.
Takeaways
Subagents solve the problem of context overload by specializing in specific tasks
Each subagent has its own context window, keeping contexts clean and focused
Create subagents with the 'agents' command in the CLI tool
Configure subagents with specific system prompts, tools, models, and visual identifiers
General agent can intelligently delegate to specialized subagents or be directed to do so
20:38
Setting up MCP Servers with Shad CN
“We can also give our agents superpowers by assigning MCP servers. There are many different servers out there, but effectively it's a fantastic way to provide tooling to our agents.”
This section explores how to extend Claude Code's capabilities using Machine-Controlled Program (MCP) servers. Leon uses Shad CN as an example - a popular component library that most AI models don't have up-to-date training on. The Shad CN team provides an MCP server that gives Claude access to current documentation and component listings.
He demonstrates how to create an mcp.json file with the Shad CN server configuration and explains platform-specific differences in the setup commands. Once configured, Claude Code detects and loads the MCP servers, giving the agent access to these external tools through the /mcp command. Leon then shows Claude using this capability to replace components on a homepage with Shad CN components, retrieving up-to-date documentation and installation instructions directly from the MCP server rather than relying on its training data.
Takeaways
MCP servers extend Claude's capabilities with external tools and up-to-date documentation
Create an mcp.json file to configure server access
Claude automatically detects and loads MCP servers from this configuration
Particularly valuable for accessing current documentation on frequently updated libraries
Allows Claude to use external tools beyond its training data
25:02
Adding MCP Servers via CLI
“You can also add MCP servers by running the following command: 'claude mcp add' followed by the name of the server then '--' and after these double dashes you will pass in the command.”
Leon explains how to add MCP servers using CLI commands rather than manually editing the mcp.json file. He demonstrates the command structure: claude mcp add [server-name] -- [command] [arguments], using Shad CN as an example with claude mcp add shad-cn -- npx shad-cn@latest mcp.
He also covers additional configuration options like specifying different transport methods (standard IO, SSE, or HTTP) and setting the scope for server availability (user-level for all projects, project-level for the current project, or local for personal use only). This provides flexible ways to manage and configure MCP servers based on specific needs and sharing requirements.
Takeaways
Add MCP servers with 'claude mcp add [name] -- [command] [arguments]'
Configure transport method with -t flag (standard IO, SSE, or HTTP)
Set scope with -s flag (user, project, or local)
User scope makes server available to all projects
Project scope adds server to project's mcp.json file
26:34
Hooks for Notifications and Automation
“Hooks are incredibly powerful. During our conversation with Claude Code, certain events are triggered like the start of a new conversation, maybe the end of a tool call, or an event could also be once the agent completes its work and generates the response back to the user.”
This section covers hooks - a way to trigger custom actions in response to Claude Code events. Leon demonstrates setting up a hook that sends a notification to his phone via Telegram when Claude completes its work, allowing him to step away from the computer during long-running tasks.
He shows how to access hooks with the /hooks command, which displays all available events like pre-tool use, after tool use, notifications, user prompt submission, session start, and response conclusion. For his example, he creates a 'stop' event hook that calls an N8N webhook, which in turn sends a Telegram message. The hook is configured through a settings.json file and can be set at either project or user level, providing a powerful way to integrate Claude Code with external systems and automation workflows.
Takeaways
Hooks trigger custom actions in response to Claude Code events
Access available hooks with the '/hooks' command
Events include tool usage, notifications, prompt submission, session lifecycle
Hooks can be configured at project or user level in settings.json
Integrate Claude Code with external systems like messaging platforms or automation tools
29:46
GitHub Integration for Remote Development
“You can even use Claude Code while you're asleep. What we can do is integrate Claude Code into GitHub.”
The final major feature covered is GitHub integration, which allows users to leverage Claude Code remotely without being at their computer. Leon demonstrates the process of deploying a project to GitHub, then running the install-github-app command in Claude Code to authorize the integration and create a long-lived token.
After setup, users can create issues in GitHub with instructions for changes, tag @Claude in the comments, and Claude will automatically implement those changes through pull requests. This enables truly remote development - users can request changes from their phone or while traveling, and Claude will handle the implementation. The entire process, from code changes to pull request creation and merging, can be performed through the GitHub web interface, making it a powerful solution for continuous development even when away from the development environment.
Takeaways
Integrate Claude Code with GitHub using the 'install-github-app' command
Create issues with change requests and tag @Claude to trigger automated implementation
Claude creates pull requests with implementations that can be reviewed and merged
Enable remote development from any device with web access
Continue development work while traveling or away from your computer
32:43
Series Wrap-up
“I really hope you enjoyed this video. If I missed anything, please let me know down in the comments.”
Leon concludes the masterclass with a brief wrap-up, asking viewers to comment if he missed any features they'd like to learn about. He encourages engagement through likes and subscriptions to his channel, where he regularly posts similar content on AI tools and automation.
This closing section serves as a call to action for viewer engagement while also directing them to additional resources. The concise nature of this wrap-up reinforces that the video has comprehensively covered Claude Code's capabilities from basic setup to advanced features.
Takeaways
The video provides a comprehensive overview of Claude Sonnet 4.5 and Claude Code 2.0
Viewers are encouraged to comment with any questions or topics that weren't covered
Additional content on AI tools and automation is available on the channel
The channel focuses on regular updates about AI automation tools and techniques
Conclusion
Claude Sonnet 4.5 and Claude Code 2.0 represent a significant leap forward in AI-assisted software development, transforming how developers can interact with their codebase. The combination of powerful coding capabilities with intuitive interfaces (both CLI and VS Code extension) makes this technology accessible to developers with different preferences and workflows.
The most revolutionary aspects are the advanced features like subagents, MCP servers, background tasks, and GitHub integration, which go beyond simple code generation to create a truly agentic development experience. These tools allow for sophisticated workflows, specialized task delegation, and even fully remote development that continues while you're away from your computer.
So what? AI-assisted development is evolving from simple code completion to full development partnership. Developers who master these tools will gain significant productivity advantages by automating repetitive tasks, delegating specialized work to subagents, and maintaining development momentum even when away from their workstation. As these technologies continue to improve, the line between human and AI contributions to software development will increasingly blur, creating new workflows that leverage the strengths of both.