Magic AI Inc., a startup that’s building an artificial intelligence platform to generate code, said today that it has raised $23 million in new funding to build out its team and prepare its product for launch.

The Series A funding round was led by Alphabet’s CapitalG and was joined by Elad Gil, Nat Friedman and Amplify Partners, bringing the total raised by the company to $28 million.

Magic’s tool, which is not available yet, operates by allowing software engineers to describe what they want in natural language and it generates code — what the company calls “software that builds software.” In this fashion, the AI and the developer can work together to find code, consider its use, reuse it and then collaborate on code changes.

In essence, the idea is that the AI would operate essentially like a “colleague inside the computer” that acts as a smart agent that can do part of the work of generating and helping edit the code.

It is a common step for many of today’s automated code generation tools to use less complex interfaces that take away the tedium of code writing, suggesting code based on what developers are already writing.

In this fashion, Magic is similar to Microsoft’s GitHub Copilot, which uses the AI model based on GPT-3 to write software code automatically. As developers write code, Copilot suggests code that will automatically complete what they are working on, and it’s capable of completing even lengthy pieces of code.

Magic takes that a giant step further by allowing developers to describe what they want in English. That sets it in a completely different space than providing suggestions as developers write code.

“Magic aims to drastically reduce the time and financial cost of developing software,” Magic Chief Executive Eric Steinberger told TechCrunch. “Giving teams access to an AI colleague who can understand legacy code and help new developers navigate it will enable companies to scale the impact of their current employees and train new employees with less personal coaching. In turn, employees will grow their skills faster and will be able to move among high-impact projects with increased agility.”

In order to do this, Magic developed its own neural network architecture that can rapidly read code, the company claims that it “can read 100 times more lines of code than Transformers.” An AI Transformer is a type of machine learning model initially developed by Google researchers and is primarily used in natural language processing and computer vision. Transformers have been implemented in standard frameworks such as TensorFlow and PyTorch.

Magic’s AI code generation platform will do more than just generate code based on what developers ask it, according to the company. The company also plans to use the power of AI to allow it to describe what a snippet of code does by looking at it, adding elements to the code based on what the developer asks, and even detecting, warning about and overcoming potential bugs in the code.

Because Magic is an AI that developers can chat with like a colleague, they can send it messages telling it what changes they want to be made to the code. The AI will take their comments and during the back-and-forth, the code will be modified to fit the parameters of the conversation.

The company said it intends to use the funding to grow its team to 25 people from its current six, focusing on engineering and development, product and go-to-market efforts. The service is currently not available, but it’s launching soon. For those interested, there is a waitlist signup on the website.

Image: Magic AI