“Imagine a world in which AI is generating a lot of software, you need to then spend a lot of time reviewing software,” Codeium CEO Varun Mohan says.
Cody Pickens for Forbes
R
ohan Choudhury, a Ph.D student in robotics at Carnegie Mellon University, writes code every day— whether that’s for training a machine learning model or running experiments. But since he started using Codeium’s AI coding assistant on his former colleagues’ suggestion, the usually arduous process has become magnitudes faster and simpler.
Tasks like debugging and tweaking code that would once take hours, now only take a few seconds, he said. All he has to do is highlight the part of the code that he wants to change and Codeium’s autocomplete tool churns out suggestions that he can directly plop into the code. “Most of the time it just works out of the box,” he said.
Featured on this year’s Forbes Next Billion Dollar Startups list of 25 companies most likely to reach a unicorn valuation, Codeium’s software is widely popular in the world of software, used by some 600,000 developers (for free) and 1,000 enterprises (like Zillow, Dell and Anduril, who pay by the user) to speed up production of features and applications. The company, which has raised $93 million at a $500 million valuation, made about $1 million in 2023 revenue.
Now, Codeium is releasing a new coding engine called Cortex, which it claims can process more data — up to 100 million lines of code — at once. That’s useful for a number of reasons: for one, because a single piece of code is “interrelated with hundreds of millions of other lines,” having fuller context about a company’s entire codebase helps Cortex make better suggestions, Coedium CEO Varun Mohan told Forbes.
It also means a single update in one part of the code can be automatically applied throughout all the files in a codebase in as few as six seconds, Mohan said. It’s a capability that is handy for Codeium’s enterprise customers, where an edit needs to be reflected across hundreds of thousands of code repositories. For instance, if Zillow wants to add a new type of information for each property listed on its site, Cortex could make it easy to apply one change system-wide.
The market for AI coding assistants has seen massive investor interest in recent months as startups have cropped up with towering valuations. Cognition Labs, the creator of a viral “AI software engineer” called Devin, raised $175 million at a $2 billion valuation in April; coding automation company Magic is in talks to raise more than $200 million at a $1.5 billion valuation. Then, there’s the giant in the room: Microsoft’s GPT-4-powered GitHub Copilot, which has 1.8 million paid subscribers and crossed $100 million in annual recurring revenue in 2023.
Compared to its startup peers, Codeium has picked up significantly less capital. But CEO Mohan only considers Microsoft GitHub Copilot as a real competitor.
“None of the [other] companies have a product app,” he said. “I can’t fight ghosts.”
Codeium’s new coding engine, however, could give it an edge against its competitors. By allowing more data to be consumed at once, Codeium’s new coding engine is a step toward what’s called “advanced reasoning”: AI systems that can use logic to better solve complex, multi-step problems, unlocking the potential for artificial intelligence to be applied for things like scientific discovery or building software. The industry is increasingly inching towards this type of AI. OpenAI, for instance, is reportedly working on “Strawberry,” an initiative to build models capable of human-like reasoning abilities.
Founded in June 2021, Codeium was previously called Excel Function. Mohan and his cofounder and MIT classmate Douglas Chen had built software to manage 10,000 GPUs for companies, making them more efficient for running applications. In 2022, before the launch of ChatGPT popularized generative AI, the duo pivoted their startup to offer their own AI models for coding, a space where they thought they could better differentiate. Had they not pivoted back then, “I think we would be a worse company,” Mohan says now.
As more startups pile into the space, Codeium is looking to stand out by making tools that developers can interact with and give feedback to. Cortex, for example, allows coders to accept or reject its suggestions rather than fully completing the task on its own.
“Instead of being a copilot, we want to be a cockpit,” he said. “And what I mean by that is we provide the ultimate amount of leverage, helping developers review, navigate and deploy code 10 times faster.”
MORE FROM FORBES
