Software code quality startup CodiumAI Ltd. is exiting stealth mode today with the beta test launch of its generative artificial intelligence-powered code integrity tool, having closed on an $11 million seed funding round.
Today’s round was co-led by Vine Ventures and TLV Partners and saw participation from angel investors including some unnamed executives from OpenAI LP, VMware Inc. and Snyk Ltd.
CodiumAI has created an AI model called TestGPT that’s powered by OpenAI’s GPT-4 large language model. Its job is to assist developers in testing their code interactively, which it does by generating tests on the go.
As CodiumAI explains, testing is a key part of the coding process. Errors in software code can be expensive, and the company claims that software of questionable quality cost U.S. companies a combined $2.08 trillion in 2020 alone.
Given those costs, it makes sense that it’s better to prevent errors than fix them after they’ve already moved through the software development lifecycle. To prevent errors, regular and thorough testing is required, but for developers this can be very time-consuming. Moreover, it’s an imperfect process, since they don’t always know what they should be testing for.
TestGPT works by analyzing the newly written code, docstring and comments as the developer is working, then suggests which tests should be performed to ensure the functions and integrity of that code. The developer only has to accept and commit those tests when TestGPT suggests them to ensure the integrity of their code.
Co-founder and Chief Executive Itamar Friedman said his company is working at the confluence of new developments such as ChatGPT and coding assistants such as GitHub Copilot. It has taken the generative AI capabilities of ChatGPT and fine-tuned them for testing code logic.
CodiumAI’s tool is available now in beta as an extension for popular integrated development environments such as PyCharm, VSCode and WebStorm, and will integrate with more IDEs and additional programming languages in future. Since launching in a closed alpha release in January, it has already been installed by thousands of developers, the company said.
TLV Partners’ Brian Sack said he’s backing CodiumAI because it addresses a real problem for developers. Although they have access to tools for testing performance, checking software vulnerabilities and generating code, there’s no viable solution for testing code logic.
“It is not a coincidence as it is a very challenging task, almost like getting into the mind of the one who wrote the code,” Sack said. “CodiumAI is a creative, targeted and focused use of generative AI and an application that provides value immediately.”