Devin 2.0: First-Ever AI Software Engineer IS TRULY INSANE! (Devin IDE, CLI Coder, & More!)

Key Concepts:

• Devon: An autonomous AI software engineer developed by Cognition Labs.
• Swaybench: A coding benchmark used to evaluate AI coding tools.
• Autonomous AI Agent: An AI system capable of performing tasks independently with tools like code editors, shells, and web browsers.
• Ask Devon: A feature providing accurate question answering across a codebase.
• Deep Wiki: A feature that transforms a repository into comprehensive documentation with diagrams.
• MCP Marketplace: A marketplace to enhance Devon's flexibility with different tools.
• Indexing: The process of familiarizing the AI agent with a codebase.
• Virtual Machine: A dedicated AI work environment for each Devon session.
• Deep Thinking: An option in Ask Devon for providing more in-depth context.

1. Introduction to Devon and its Evolution

• Devon, developed by Cognition Labs, is presented as the world's first AI engineer.
• It achieved state-of-the-art results on the Swaybench coding benchmark, solving complex software engineering tasks autonomously.
• Unlike coding assistants like GitHub Copilot or Cursor, Devon is an autonomous AI agent equipped with a code editor, shell, web browser, and full internet access.
• Since its initial release, Devon has been continuously upgraded, becoming more reliable and proficient.
• Multiple Devons can be run in parallel, each working on different tasks.
• Devon integrates with tools like Slack, Linear, Jira, GitHub, and GitLab.
• A Devon API allows programmatic launching of Devons in response to events like build failures or crashes.

2. New Subproducts: Ask Devon and Deep Wiki

• Two major subproducts have been introduced: Ask Devon and Deep Wiki.
• Ask Devon provides ultra-accurate question answering across a codebase, acting as a deep research tool for engineers.
• Deep Wiki transforms a repository into comprehensive documentation with diagrams, like having a full documentation team.

3. Accessibility and Pricing

• Previously, Devon was inaccessible due to its high cost.
• A new "core plan" is introduced, offering pay-as-you-go access starting at $20, providing access to most features.

4. Using Devon: A Step-by-Step Guide

• To get started, users can click "start building" to create an account.
• Users can connect a Git provider (e.g., GitHub) to import an existing project or create one from scratch.
• The main dashboard allows users to describe tasks or questions in natural language.
• Users can attach files, repos, and macros to provide context.
• The agent can be used to work on tasks, or the "ask agent" can be used to answer questions.
• Devon can be connected to Slack or Linear to automatically pull in tickets or assign tasks.
• The left-hand side allows creating and managing sessions, accessing Ask Devon, and using Deep Wiki.
• Settings allow managing billing, repositories, Devon machines, knowledge bases, and secrets.

5. MCP Marketplace Integration

• Devon now integrates an MCP marketplace to enhance its flexibility with various tools.

6. Indexing and Setting up the Virtual Machine

• To start working with Devon, users should first index the repository.
• The indexed repository can then be added to the Devon machine.
• The Devon machine is a dedicated AI work environment, a virtual development setup for each session.

7. Testing Devon's Capabilities

• To test Devon, the presenter asks it to access a repository, run initialization and lint, and open a dummy draft PR.
• Devon successfully accesses the repo and improves the documentation.
• Devon has an integrated shell and a virtual code editor within the Devon machine.
• Devon creates a draft PR, verifies repo access, improves documentation, and even creates a diagram.

8. Running Multiple Agents Simultaneously

• Devon can run multiple agents simultaneously, each working on different tasks.
• The presenter demonstrates running one agent to open a dummy PR and another to debug the codebase.
• Devon generates plans for each task, outlining the step-by-step process.
• The debugging agent fixes dependency conflicts and missing files, creating a PR request.

9. Ask Devon in Action

• Ask Devon allows users to ask questions about their repository.
• Enabling "deep thinking" provides more in-depth context.
• The presenter asks "What is the deepseek web crawler app?" and receives a detailed response.

10. Deep Wiki Demonstration

• Deep Wiki autonomously creates comprehensive documentation for the repository.
• The documentation includes key components overview, graphs describing the workflow, and instructions on getting started.

11. Conclusion

• Devon has significantly improved over time and is a valuable AI coding assistant.
• The presenter encourages viewers to explore Devon and its features.
• Viewers are encouraged to subscribe to the channel, join the newsletter and Discord, and follow the presenter on Twitter.