Verdent: NEW Agentic Coder IS POWERFUL! Can Build ANYTHING! (Sub Agents + SOTA)

Key Concepts

• Verdant: A full-stack AI engineering system designed for planning, coding, verification, and multi-agent execution with production-level reliability.
• Agent UI: A user interface for interacting with AI agents.
• Warp: A terminal emulator with an overhaul supporting multiple agents.
• Multi-agent execution: The ability for multiple AI agents to work concurrently on a task.
• Isolated Git worktrees: Separate, clean environments for each agent session to ensure reproducibility.
• Swaybench verified test: A benchmark for evaluating AI coding agents.
• Multimodal orchestration: Coordinating different AI models and capabilities.
• Long context + persistent memory: The ability of an AI to retain and utilize a large amount of information over time.
• Automated verification: AI automatically checking the correctness of generated code.
• Automated code review: AI automatically reviewing code for quality and potential issues.
• Verdant Deck: The desktop application for Verdant, featuring a multi-agent platform.
• Verdant VS Code Extension: An extension for VS Code that integrates Verdant's AI coding agent.
• Planner agent: An agent responsible for generating detailed plans for execution.
• Sub-agents: Specialized agents within Verdant that handle specific tasks (e.g., verifier, explorer).
• GBC 5 and Cloud 4.5: Powerful AI models integrated into Verdant.
• Presets (Performance, Balance, Efficiency): Configurable settings for AI models to optimize for different needs.

Verdant: A Next-Generation AI Engineering System

This video introduces Verdant, a new state-of-the-art AI coding agent that aims to unify planning, coding, verification, and multi-agent execution with production-level reliability. It is presented as a significant advancement beyond existing tools like Cursor 2.0 and Warp.

Comparison with Existing Tools

• Cursor 2.0: While offering an agent UI, it is fundamentally a single-agent experience.
• Warp: Its multi-agent workflow is powerful but confined to a terminal-first environment, not optimized for full lifecycle software delivery.

Verdant distinguishes itself by being a "full stack AI engineering system" with distinct tool modes.

Verdant's Architecture and Capabilities

Verdant offers two primary access points:

1. Verdant Deck (Desktop App):

   • Multi-agent platform: Runs multiple agent sessions in parallel, asynchronously.
   • Isolated Git worktrees: Each agent session operates in a clean, reproducible environment.
   • Context awareness: Advanced context integration, including MCP integration.
   • Ideal for: Asynchronous multi-project workflows.

2. Verdant VS Code Extension:

   • Integrated AI coding agent: Plans, writes, tests, and verifies code directly within the editor.
   • Production-level reliability: Focuses on generating correct, testable, maintainable, and production-ready code.
   • Full control over sub-agents: Users can create custom sub-agents or utilize built-in ones for specialized tasks.
   • Comparable to: Tools like Cursor, but with enhanced capabilities.

Performance and Verification

• Swaybench Verified Test: Verdant achieved an impressive 76.1% pass rate, outperforming Cloud Code and Codeex. This positions Verdant as a highly promising agent for solving real software engineering problems.
• Key to success: This performance is attributed to:
  • Multimodal orchestration: Coordinating various AI models and functionalities.
  • Long context + persistent memory: Enabling agents to retain and utilize extensive information.
  • Automated verification: AI automatically checks code correctness.
  • Automated code review: AI performs code quality checks.

Getting Started with Verdant Deck

1. Download: Access Verdant Deck via a provided link for a limited-time free trial.
2. Sign Up: Create an account to access state-of-the-art models.
3. Agent Selection: Choose agent types, such as the "planner agent" for detailed task structuring.
4. Context Attachment: Attach files and add context to agents or specific files/folders.
5. Project Management: Manage projects and tasks efficiently.

Example: Web Scraping Agent Enhancement

• Task: Prototype a new feature for a web scraping agent by creating a CLI-integrated interface.
• Process:
  • Request Verdant to create a CLI interface for the web scraping agent.
  • Tag specific folders and files for the agent's reference.
  • Simultaneously: Initiate another agent task to find and fix errors in the project files.
  • Tag the overall folder and specific files (e.g., main.py) for the debugging agent.
• Outcome: Verdant successfully created the CLI interface and fixed associated errors, demonstrating asynchronous multi-agent workflows within isolated Git worktrees.

Verdant Deck Features:

• Terminal Panel: Directly accessible within the application.
• Editor View: Provides logs, diff changes, and an overview of all modifications.
• Step Changes & Diff Length: Visualize and generate diffs for changes.
• Code Review: Access code review functionality.
• Commit Changes: Commit generated code directly.

Getting Started with Verdant VS Code Extension

1. Install: Search for "Verdant" in the VS Code extension store and install it.
2. Sign In: Authenticate using the account created for Verdant Deck.
3. Prompting: Use natural language prompts to instruct the agent.

Example: Creating a Crypto Exchange

• Prompt: Request the creation of a professional crypto exchange inspired by Hyperledger, with numerous features.
• Modes:
  • Agent Mode: Autonomous building.
  • Manual Review: Option to manually accept or skip permissions.
  • Auto Mode: Fully automated execution.
• Sub-agents: Utilize built-in sub-agents (verifier, explorer) or create custom ones by defining their task, description, and example prompts.

AI Model Customization and Presets:

• Swaybench Performance: Achieved through customizing AI models using presets.
• Presets:
  • Performance: Maximizes context and uses frontier models.
  • Balance: Blends performance with cost and efficiency.
  • Efficiency: Prioritizes fast, responsive, and cost-efficient tasks.
• Model Integration: Combines capabilities of models like GBC 5 and Cloud 4.5.
• Free Tier Access: All models are accessible for free during the trial period.

Execution and Outcome:

• Live Preview: Observe the agent's progress in real-time, including plan creation and project structure setup.
• Rapid Development: Verdant quickly generated a fully developed crypto exchange, including backend server and frontend application.
• Comprehensive Output: Provided a visual dashboard layout, component breakdown, and a guide for getting started.
• Startup Script: Included a start.bat file to launch backend, frontend, and server components.
• Live Demo: The generated crypto exchange was demonstrated, showcasing features like market orders, recent orders, and open orders, with simulated live data.
• End-to-End Functionality: The agent successfully configured both the backend and frontend without errors.

Conclusion and Recommendation

The speaker highly recommends Verdant, calling it their "favorite AI coding agent" due to its state-of-the-art capabilities and its approach to software development as a full-stack, end-to-end process. Key strengths highlighted are:

• Multi-agent orchestration
• Agent capabilities
• Code generation

The speaker encourages users to try Verdant, especially during the limited free trial, to access advanced AI models for free.