Kevin Leneway, principal software engineer at Pioneer Square Labs (PSL), has been exploring the potential of AI in software development. This exploration led to the creation of JACoB, an acronym for “Just Another Coding Bot.” Initially introduced earlier this year, JACoB has evolved through user feedback and validation. This week, PSL made JACoB available as an open-source project.
Capabilities and Integration
JACoB integrates seamlessly with GitHub and other widely-used development tools. It can write and review code, convert Figma designs into code, automate routine tasks such as pull requests, and identify potential security flaws and bugs. The bot is designed to learn from patterns, thereby improving the quality of its output over time.
PSL emphasizes that JACoB includes built-in safeguards to prevent runaway processes. Early users have reported increased efficiency and productivity without compromising on quality or security. The tool mimics the workflows typically associated with junior developers, which have been refined over decades.
Open-Source and Customization
One of the standout features of JACoB is its open-source nature. This allows developers to inspect the underlying code and contribute to its improvement. Additionally, users have the option to run JACoB locally, providing extra data protection and code security. A hosted version is also available, free during a trial period.
Currently, JACoB is optimized for TypeScript and JavaScript. Leneway explained that the focus on a narrow use case was intentional to ensure high-quality output. The bot works particularly well with NextJS applications, Tailwind, and Figma for design-to-code transformations. In a video on YouTube, Leneway demonstrated how JACoB works, showing it creating and building its own homepage.
While JACoB has not yet been spun out as a separate startup, Leneway hinted that this could be a future possibility. For now, the focus is on open-sourcing the project and allowing developers to experiment with it. PSL is keen to see how the community adopts and adapts JACoB.
Getting Started with JACoB
Developers interested in JACoB can sign up on the project's website or follow a quick start guide. The bot integrates directly into GitHub workflows, transforming Figma designs into deployable code and understanding the entire codebase. The project is open for contributions, and developers are encouraged to join the community to help shape the future of AI in software development. Leneway posted a YouTube video discussing the setup process for JACoB.
Technical Documentation
The project includes a technical white paper detailing JACoB's methodology and architecture. Evaluations against other design-to-code tools have shown impressive results. The bot is designed to be flexible and extensible, allowing customization to fit specific needs while emphasizing privacy and security.
There wasn't a standard test for how well design-to-code tools translate designs into real code. So, PSL built their own evaluation system to assess JACoB's capabilities in practical situations. This testing and evaluation ground is known as JACoB Arena. During tests, JACoB excelled by beating eight other popular tools and even code written by humans. It scored an ELO of 1183, which is over 100 points higher than its closest competitor. For a deeper dive into how we achieved this, check out the technical whitepaper included in the repository.
