Build Anything with Claude 3.7, Here’s How

Claude 3.7 Sonet: Build Anything with the Best AI Model

Key Concepts:

• Claude 3.7 Sonet: Anthropic's new AI model, potentially the best in the world, excelling in software engineering, agentic tool use, and design.
• Cursor: An AI-powered code editor that integrates with AI models like Claude 3.7 Sonet to enhance coding productivity.
• Vectal: An AI startup leveraging Claude 3.7 Sonet and other AI models to provide AI agents for various tasks, including infinite thinking.
• YOLO Mode: A feature in Cursor that allows AI agents to automatically execute terminal commands, requiring careful configuration of command allow/deny lists.
• CLA Code: Anthropic's autonomous coding tool, an AI agent that operates in the terminal.
• MCP (Model Context Protocol): A new standard from Anthropic that allows connecting AI agents to external services like Google Drive and GitHub.
• Infinite Thinking: A feature in Vectal that allows AI agents to constantly work on user-defined tasks.

1. Introduction to Claude 3.7 Sonet

• Anthropic released Claude 3.7 Sonet after four months of silence.
• The video aims to test the model, integrate it with Cursor, and demonstrate building AI agents.
• Vectal has already integrated Claude 3.7 Sonet, offering access to both free and paid users.
• Vectal's new "infinite thinking" feature allows AI agents to continuously work on tasks.
• Anthropic also released a new reasoning model that displays tokens during reasoning.

2. Performance Benchmarks

• SWE Bench Verified: Claude 3.7 Sonet achieves 62% with the default model and 70% with custom scaffolding, surpassing previous models stuck around 49%.
• Agentic Tool Use: Claude 3.7 Sonet dominates benchmarks, outperforming even OpenAI's models.
• CLA Code: Anthropic's first AI agent, an autonomous coding tool resembling Cursor running in the terminal.

3. Significance of Claude 3.7 Sonet

• Claude 3.5 Sonet was already the primary coding model for many developers and startups.
• The speaker used Claude 3.5 Sonet to build Vectal in two months, a task that would have taken over six months otherwise.
• Claude 3.7 Sonet is expected to facilitate the creation of millions of new entrepreneurs by 2025.
• It improves design capabilities, including UI and 3D generation.
• Example: A 3D game generated by Claude 3.7 Sonet in minutes, demonstrating its creative potential.

4. Reduced Censorship

• Previous Claude models were heavily censored.
• Claude 3.7 Sonet significantly reduces unnecessary refusals: 45% in standard mode and 31% in extended thinking mode.
• Example: Claude 3.5 refused to provide a fire emoji, while Claude 3.7 readily does so.
• Example: Claude 3.5 refused to answer what happens when mixing bleach and ammonia, while Claude 3.7 answers with a warning.

5. Anthropic's Practical Approach

• Anthropic focuses on creating practical and useful AI models optimized for real-world use cases.
• They prioritize results over hype and benchmarks.
• The speaker is switching to Claude 3.7 Sonet in Cursor for its practicality.

6. Evolution of Claude

• 2024: Claude as an assistant, helping users get tasks done faster.
• 2025: Claude as a collaborator, saving users hours of time by handling complete tasks.
• 2027 (Prediction): Claude as a pioneer (AGI), inventing new solutions, solving problems, and creating cures.
• In 2025, Claude is like a full employee, capable of solving problems and accelerating business growth.

7. Installing and Setting Up Cursor

• Use Vectal to get instructions on installing Cursor on a MacBook.
• Download Cursor from cursor.sh.
• Create a Cursor account.
• Open a project in Cursor by creating or selecting an empty folder.
• Update Cursor to the latest version.
• The new Cursor UI is powered by the Cursor Agent.
• Select Claude 3.7 Sonet or Claude 3.7 Sonet Thinking as the model in the Cursor Agent.
• Enable the models in Cursor settings under "Models."
• Update Cursor by clicking the update notification in the bottom right corner.

8. Configuring Cursor Features

• Cursor Tab: Enable for code completion suggestions.
• Chat:
  • Auto-apply to files outside of context: Recommended.
  • YOLO Mode: Allows automatic execution of terminal commands. Use with caution and configure command allow/deny lists.
  • Command Allow List: Add safe commands like git status.
  • Command Deny List: Add risky commands like rm -rf and fork bombs.
  • Delete File Protection: Enable to prevent the agent from deleting files.
  • MCP Tools Protection: Configure if using MCP servers.
  • Large Context: Keep enabled.
  • Collapse Input: Keep enabled.
  • Iterate on Lint: Automatically solve linter errors.
• Review Changes: Leave at the default setting.
• Web Search: Improved web search capabilities within Cursor.
• Docs: Add frequently accessed documentation URLs for easy access.
• Editor: Copy the speaker's settings.
• Terminal: Use command K to access the terminal helper. Select Claude 3.7 Sonet in the terminal helper as well.
• Rules: Define rules for the Cursor Agent. Access the speaker's optimized rules in the New Society classroom.

9. Building a 3D Game with Claude 3.7 Sonet in YOLO Mode

• Create a new Python file named game.py.
• Provide instructions to Claude 3.7 Sonet to create a full 3D game in game.py.
• Instruct the model to install necessary libraries using the terminal.
• Emphasize making the game visually impressive and interactive.
• Encourage the model to solve errors using code running or terminal execution abilities.
• The speaker demonstrates the process of using YOLO mode to autonomously build the game.
• The model first lists and activates the testing environment.
• The model installs the necessary packages, including the urisa game engine.
• The model writes the outline into the game file and then builds on top of that.
• The speaker provides feedback and hints to guide the model when it encounters issues.
• The speaker emphasizes that while errors may occur, the key is to provide feedback and allow the AI to solve them.

10. CLA Code: Anthropic's Autonomous Coding Agent

• CLA Code is Anthropic's competitor to Cursor, an AI agent that operates in the terminal.
• Use Vectal to get documentation on setting up CLA Code.
• Install CLA Code globally using npm install -g @anthropic-ai/cla-code.
• The speaker demonstrates using Cursor in YOLO mode to install CLA Code.
• Initialize CLA Code by running the cla-code command in the terminal.
• The speaker attempts to log in to the Anthropic console but finds that CLA Code is currently at capacity.

11. Conclusion

• Claude 3.7 Sonet and AI agents like CLA Code are revolutionizing coding and productivity.
• Tools like Cursor and Vectal empower users to achieve more with AI assistance.
• The speaker encourages viewers to embrace these technologies to stay ahead and conquer the world in 2025.
• The speaker promotes Vectal and its "infinite thinking" feature as the future of productivity.
• The speaker encourages viewers to subscribe for more content on AI tutorials, including a dedicated video on CLA Code and MCP.