Since the launch of GitHub Copilot, which began by completing lines of code and later entire code snippets, many have wondered: when will we be able to simply describe an application in natural language for Copilot to generate it entirely? Numerous experiments have been conducted along these lines in recent months. Now, GitHub is advancing this vision with the announcement of GitHub Spark, presented at the annual GitHub Universe conference in San Francisco.
GitHub Spark, an experimental project launched from the GitHub Next labs, enables the rapid creation of small web applications using only natural language instructions. Advanced developers still have the option to review and adjust the code (the foundation relies on a GitHub repository, GitHub Actions, and uses Microsoft’s Azure CosmosDB as the default database for applications that need it). However, the main objective is to offer a chat-like experience so users can build a prototype and gradually refine it.
To start an application, it’s enough to send an initial message, and Spark will generate a preview in seconds. From there, users can make adjustments and iterate easily in collaboration with the bot.
Thomas Dohmke, GitHub’s CEO, states, “We sincerely believe this is the next step in creating personal software. These microapplications are not meant to replace professional developers but rather to be a tool for exploring ideas, building small assistants, or simply experimenting with development using natural language.”
Although GitHub refers to Spark as a tool for creating “microapplications,” Dohmke mentioned that it’s not yet clear if there are any defined limits on the complexity of applications that can be created with Spark. A GitHub spokesperson confirmed that there are currently no restrictions on prompts or capabilities (though this could change after the technical preview). Spark can use any web API and leverage AI models, and Dohmke is excited to see how users will take this tool to its full potential.
“It will be fascinating to see how far users can take this tool, the size and complexity of applications that can be created just with natural language and this simple interface designed to foster creativity,” he stated.
Since GitHub Copilot allows users to select which large language model they want to use, Spark offers the same flexibility, allowing them to choose between the recent versions of Claude Sonnet from Anthropic and OpenAI’s GPT models.
Spark also facilitates project sharing with customized access controls, and even more interestingly, it allows users to take the code from a shared Spark and continue developing it. Developers can access and modify the code at any time, which is useful when errors occur, as Dohmke acknowledges that AI is not infallible. “You can always see the code base,” he noted. “If the foundation is understood, it’s possible to review and adjust the code directly, which is often necessary when AI makes an error.”
By: Nestor Castillo, ForAllTech Director