$25 million in Series B: Tabnine – AI assistant for programming

Developers are ready to embrace AI tools that give them productivity improvements and accelerated learning. According to a recent survey, 77% of developers have a positive attitude toward using AI in their workflows, and 70% say they are using or plan to use AI programming tools this year.

Investors also see potential in generative programming tools – especially at enterprise scale. And that enthusiasm is translating into new funding for startups like Tabnine, which today announced it has raised $25 million in a Series B funding round led by Telstra Ventures and with participation from Atlassian Ventures, Elaia, Headline, Hetz Ventures, Khosla Ventures and TPY Capital.

Dror Weiss and Eran Yahav co-founded Tabnine in 2012 with the goal of creating a platform to enable the use of generative artificial intelligence at various stages of the software development lifecycle. Yahav was and still is a professor at the Technion (Israel Institute of Technology), while Weiss is a computer science graduate of the Technion.

Among other programming tools based on its own and third-party generative AI models, Tabnine offers Tabnine Chat, a “code assistant” that writes code and answers questions about an organization’s codebase – a kind of ChatGPT for code.

Tabnine has competitors in the form of GitHub Copilot and Amazon CodeWhisperer. But Weiss says the company offers more control and personalization than competing systems, such as allowing customers to deploy their tools both on-premises and in a virtual private cloud.

“Our flexible architecture allows us to change code-generative AI models relatively easily and thus not compete with the big generative model vendors,” Weiss told TechCrunch. “This is how we ensure we are protected for the future when new models from other vendors emerge as AI evolves. Tabnine can provide those models to developers wherever they write code.”

Weiss also argues that Tabnine is less legally risky than its competitors – at least from a commercial standpoint.

A class action lawsuit is currently pending against Microsoft, GitHub and OpenAI accusing them of violating intellectual property law because Copilot, trained on billions of public code samples from the Internet, some of which were created under a restrictive license, reproduces sections of copyrighted code without attribution. Some legal experts believe that AI like Copilot could put companies at risk if they unwittingly incorporate copyrighted sentences into their off-the-shelf software.

Tabnine, Weiss notes, exclusively uses AI models trained on code with permissive licenses, or works with clients to train them on their own codebases.

“We use a curated dataset and know what went into it, so we have much better control and security,” Weiss says. “This is also the basis for our customers to use private models that are trained on their own code and run in their own virtual private clouds and data centers.”

Tabnine’s approach seems to be working for it, which is all the more impressive in light of the collapse of one of its competitors, Kite, late last year. Tabnine claims to have more than a million users and 10,000 customers, which, while falling short of Copilot’s million paying users and 37,000 enterprise customers, is still a pretty solid user base.

Weiss said the Series B funds, which ultimately increased Tabnine’s total funds raised to $55 million, will be used to expand Tabnine’s generative programming capabilities and further develop its sales and global support team. Tabnine expects to end the year with a staff of 150 employees, up from about 60 today.