If artificial intelligence can learn to write poems, it should also be able to write computer code.
That's the hope of San Francisco-based Magic.dev. The generative AI company has raised a $23 million Series A, bringing its total funding to $28 million. The round was led by Alphabet-affiliated fund CapitalG with participation from investors Nat Friedman, Elad Gil and Amplify Partners.
The company uses large language models like with OpenAI's ChatGPT to develop a software that can write lines of code based on a text prompt. The program is meant to make software engineers more efficient and not, at least for now, to replace them altogether.
"I think building an AI software engineer, which is the primary product objective that we have, is both an avenue to something that's incredibly useful for companies, for developers and for people who don't know how to code," said Magic.dev co-founder and CEO Eric Steinberger.
OpenAI and Github have already developed a similar product called Copilot, which automatically completes code written by programmers using AI.
However, Steinberger says Magic.dev differentiates itself from by being more personalized to teams and projects by remembering the project and building off its knowledge as opposed to generating new code based on a fixed data set.
"With Copilot, you have one completion at a time and after every completion, it forgets what it's done before," Steinberger said. "Whereas with Magic, it's just like a colleague, it remembers what it's done before. It remembers your codebase, and it can act within that much more complex setup of information."
While Copilot remains more or less a competitor, former Github CEO Nat Friedman, who oversaw the creation of Copilot, is an investor in Magic.dev.
Jill Chase, a partner at CapitalG who led the investment into Magic.dev, says her firm was actively looking for a company building generative AI for software development and was introduced to Steinberger through Friedman.
"The market for software engineers is supply constrained, and people will pay for productivity, and so if you could give them a generative AI tool to make them more productive, that would be really compelling and businesses would be willing to pay a lot of money for that," Chase said. "Then when we looked into the market at what people were doing today, all we saw was really sort of this GitHub copilot situation, which is more of this autocomplete, and which is really helpful. But it's not quite the same as having sort of a true AI software engineer colleagues that you trust as if you're pair programming with them."
Steinberger says Magic.dev is currently pre-revenue and will not disclose its planned revenue model, although he says it will be B2B. The company has six employees working remotely and Steinberger splits his time between London and San Francisco.
