Devin AI: The First AI Software Engineer

Devin AI: The “First AI Software Engineer” – A Game Changer or Just Hype?

by

Devin AI: The First AI Software Engineer. Remember the buzz around Devin, the self-proclaimed “first AI software engineer?” It garnered significant hype and even secured funding, with many speculating it could at least replace junior software engineers.Remember the buzz around Devin, the self-proclaimed “first AI software engineer?” It garnered significant hype and even secured funding, with many speculating it could at least replace junior software engineers. But now that Devin is publicly available, how does it really stack up? We dive into its capabilities, a hands-on experience, and the broader implications for software development jobs.

What is Devin AI?

Devin positions itself as an AI software engineer capable of crushing your backlog with a personal AI engineering team. It claims to automate tasks like creating GitHub pull requests, handling entire workflows, and performing as a full-fledged software engineer. Interestingly, Devin’s capabilities have been enhanced by its acquisition of Windframe, another AI coding product by Cognition Labs, meaning Devin now incorporates Windframe’s functionalities.

A key differentiator for Devin is that it operates on its own virtual machine, running on Ubuntu 22.04. Unlike some other AI coding tools that make changes directly on your machine, Devin has its own Command Line Interface (CLI), terminal, Integrated Development Environment (IDE), and even an interactive browser to perform and test its work. The service is paid, with the reviewer initiating their use by paying $20.

A Hands-On Journey with Devin

Let’s walk through the experience of using Devin to understand its practical application:

  • Getting Started & Setup:
    • Accessing Devin requires a login, which can be done via Google.
    • A crucial step involves connecting your GitHub account, with options for administrators or solo developers. This process requires authenticating your GitHub account on a separate device using a provided code.
    • Users grant Devin permissions for specific GitHub repositories, not necessarily all of them, for security reasons.
    • Devin also offers integrations with tools like Slack and Linear.
  • Repository Integration:
    • The reviewer created a new GitHub repository named “PAP ToDo”.

    • This repository was then indexed and added to Devin’s virtual machine, where Devin clones it to begin working. Devin advises against moving or deleting the cloned directory to ensure smooth operation.
  • Building a PHP To-Do Application:
    • The reviewer tasked Devin with building a PHP To-Do application from scratch, starting with a minimal README.md repository.
    • Devin’s process included:

      • Installing PHP on its own machine.
      • Managing dependencies by setting up Composer and running composer install.
      • Creating the basic application structure, including the main index.php file, which serves as the entry point.

    • Users can define commands for linting, syntax checks, and how Devin should run the application locally.

    • Interactive Testing: Devin uses its built-in interactive browser to run and test the application, allowing the user to see the output and functionality directly.

    • Iterative Improvement: When the initial UI of the To-Do app wasn’t appealing, the reviewer simply told Devin, “UI doesn’t look good“. Devin took this feedback, edited the index.php file, and significantly improved the application’s user interface, making it look much more polished and professional.
  • Automated Pull Request Creation:
    • One of Devin’s most anticipated features is its ability to directly create pull requests on GitHub.

    • Upon instruction, Devin executed a series of git commands on its virtual machine: it checked the git status, created a new feature branch, added the modified files (including a newly generated .gitignore file), committed the changes, pushed the branch to GitHub, and finally created a pull request using the gh pr create command.

    • The pull request appeared on GitHub, allowing the user to review all the changes made by Devin (e.g., additions to .gitignore, composer.json, and index.php with new CSS). If satisfied, the user can then merge the pull request.

First Impressions and Key Features

The reviewer noted that while Devin shares similarities with tools like Cursor AI and Windframe, it elevates the experience to “another level”.

  • Own Machine & Browser: Devin’s ability to operate on its own machine with its own terminal, CLI, and especially its interactive browser, simulates the workflow of a real software developer, which is seen as a significant advantage. Unlike Cursor AI, which makes changes on the user’s local machine, Devin isolates its work.

  • Direct Pull Requests: The capability to directly create pull requests on GitHub is highlighted as a “dangerous” (in a good way) and exciting feature, making it feel like someone is doing the work for you.

  • Transparency: Devin also displays its thinking process and the commands it executes, similar to Cursor AI, providing transparency into its operations.

  • Self-Adaptability: It can even update its own setup configurations based on the environment.

  • Solving Real-World Problems: Devin is lauded for its potential to solve real-world software development problems, earning praise for its development team.
Devin AI: The First AI Software Engineer

The AI-Job Replacement Debate: A Balanced View

The emergence of tools like Devin inevitably sparks conversations about AI replacing human jobs.

  • AI as a Tool, Not a Replacement: The core message is that AI, including Devin, is fundamentally a tool. These AI tools are created by real software developers themselves.

  • Upskilling is Key: Rather than being replaced, those who learn to use AI effectively and understand the underlying concepts of machine learning, AI, and data science will be the ones who succeed. AI’s advent can be a “boon” rather than a “curse” if embraced and understood.

  • Consequences of Stagnation: The threat of replacement primarily applies to individuals who are unwilling to step out of their comfort zones, lack essential skills, or perform only simple, repetitive tasks. Employers will deploy AI if it can perform tasks more efficiently than an employee.

  • Human Intervention Remains Crucial: The idea that AI can build entire companies or complex applications autonomously is dismissed. Human intervention is still essential for building, deploying, and managing beyond just code generation. If AI could do everything, major tech companies would not be hiring AI engineers and software developers.

  • Job Numbers May Shift, Not Disappear: While the number of engineers required for a project might decrease (e.g., from 25 to 15), the number is unlikely to reach zero in the foreseeable future. A zero-human scenario would imply the arrival of Artificial General Intelligence (AGI), which currently remains in the realm of fiction.

  • No Job Is Truly “Safe” (if one thinks broadly): The argument that if AI can replace software developers, it could logically replace any other job (like cooking or driving), highlights the oversimplification of the fear of complete human replacement.

Conclusion

Devin AI is an impressive product that shows the potential of AI in software development, particularly with its own execution environment, interactive browser, and direct pull request capabilities. It effectively demonstrated its ability to understand instructions and iteratively improve a web application.

However, the consensus is that AI is a powerful tool to augment, not entirely replace, human software developers. The key takeaway for developers is to adapt, learn AI and data science skills, and embrace these tools to enhance their productivity and problem-solving abilities, rather than fearing job displacement.

What is Devin AI and what are its core claims?

Devin AI has been introduced as the "first AI software engineer" and claims to automate software development tasks, acting like a full-fledged software engineer. It positions itself as a tool that can help users "crush your backlog with your personal AI engineering team" and is designed to automatically create GitHub pull requests. Notably, Devin has integrated the functionalities of Windframe, an AI coding product acquired by Cognition Labs, meaning Devin now possesses Windframe's capabilities.

Devin AI operates on its own dedicated virtual machine, specifically running on Ubuntu 22.04. Unlike some other AI coding tools that might make changes on a user's local machine, Devin provides its own Command Line Interface (CLI), terminal, Integrated Development Environment (IDE), and an interactive browser for performing and testing its work.

To begin using Devin AI, users must first log in, typically via Google. A critical step involves connecting a GitHub account, with options provided for administrators or solo developers. This process requires authenticating GitHub on a separate device using a provided code and granting Devin permissions for specific GitHub repositories for security. Devin is a paid service, with the reviewer initiating use by paying $20. The service also offers integrations with other tools like Slack and Linear. Once a GitHub repository is selected, it is indexed and cloned onto Devin's virtual machine, where Devin advises against moving or deleting the cloned directory for smooth operation.

Yes, Devin AI demonstrates the capability to build applications from scratch and iterate based on feedback. For example, when tasked with building a PHP To-Do application, Devin successfully installed PHP on its own machine, managed dependencies by setting up Composer and running composer install, and created the basic application structure, including the main index.php entry point. Users can define commands for linting, syntax checks, and how Devin should run the application locally. A significant feature is Devin's built-in interactive browser, which allows it to run and test the application directly, enabling the user to see the output and functionality. When the initial UI of the To-Do app was deemed unappealing, the reviewer simply provided feedback, and Devin edited the index.php file, significantly improving the application's user interface to be more polished.

Devin AI can directly create pull requests on GitHub. Upon instruction, Devin executes a series of git commands on its virtual machine: it checks the git status, creates a new feature branch, adds modified files (including a newly generated .gitignore file), commits the changes, pushes the branch to GitHub, and finally creates a pull request using the gh pr create command. The created pull request then appears on GitHub, allowing the user to review all the changes made by Devin (e.g., additions to .gitignore, composer.json, and index.php with new CSS) before merging.

The key advantages and impressive features of Devin AI, as observed by the reviewer, include its ability to operate on its own machine with its own terminal, CLI, and interactive browser, which simulates the workflow of a real software developer. The capability to directly create pull requests on GitHub is highlighted as a "dangerous" (in a positive sense) and exciting feature, making it feel like someone is doing the work for you. Devin also displays its thinking process and the commands it executes, providing transparency, similar to Cursor AI. Furthermore, it can update its own setup configurations based on the environment. Overall, Devin is lauded as an amazing product for its potential to solve real-world software development problems, earning praise for its development team.

The overarching message from the source is that AI, including Devin, is fundamentally a tool. These AI tools are, in fact, created by real software developers themselves. The view presented is that those who learn to use AI effectively and understand the underlying concepts of machine learning, AI, and data science will be the ones who succeed. The advent of AI can be a "boon" rather than a "curse" if embraced and understood. The threat of replacement primarily applies to individuals who are unwilling to step out of their comfort zones, lack essential skills, or perform only simple, repetitive tasks, as employers will deploy AI if it can perform tasks more efficiently than an employee. The idea that AI can build entire companies or complex applications autonomously is dismissed, as human intervention remains crucial for building, deploying, and managing beyond just code generation. While the number of engineers required for a project might decrease (e.g., from 25 to 15), the number is unlikely to reach zero in the foreseeable future, as a zero-human scenario would imply Artificial General Intelligence (AGI), which is currently considered fiction.

In conclusion, Devin AI is considered an amazing and impressive product that showcases the potential of AI in software development, particularly with its own execution environment, interactive browser, and direct pull request capabilities. It effectively demonstrated its ability to understand instructions and iteratively improve a web application. The consensus, however, is that AI is a powerful tool to augment, not entirely replace, human software developers. The key takeaway for developers is to adapt, learn AI and data science skills, and embrace these tools to enhance their productivity and problem-solving abilities, rather than fearing job displacement.

Leave a Comment