Whether it’s predictive text, real-time language translation, or voice assistants, code is the foundation of every AI-driven application.
While AI plays a critical role in the output of the software development lifecycle (SDLC), AI also can accelerate it. Tabnine, an AI-powered code completion assistant, is helping developers get next-generation applications from planning to development and release much faster — one line of code at a time.
Empowering developers with generative AI
Generative AI — AI technology that allows developers to create new content such as text, audio, images, code, and video based on historical data and trained models — underpins Tabnine’s key function and features. Tabnine uses generative AI in several ways to make developers more productive. The tool predicts and suggests developers’ next lines of code based on context and syntax. It works with developers in flexible ways based on their preferences, whether a developer wants to use a conversational style where larger snippets of code are generated or prefers to stay in flow with shorter inline suggestions that reduce some of the hassle associated with boilerplate functions. As a result, the tool improves code quality and helps developers produce anywhere from 25% to 40% of their code.
Brandon Yung, vice president of ecosystems at Tabnine, says the company uses both customized and open-source AI models, which enable it to innovate faster. Tabnine has made the ethical decision to train its models using only fully permissive open source code, meaning the code isn’t copyrighted and other developers can use it freely. Tabnine also doesn’t aggregate code across its customers.
“We’re able to train on your code base,” says Yung. “Really good data from your specific company, from your specific use cases, and from your specific languages are what makes this the most powerful.”
To drive all these capabilities, Tabnine needs high-end computing power and the ability to easily segment data in multiple ways. All of these factors led Tabnine to turn to Google Cloud.
Harnessing the processing power and security of Google Cloud
Tabnine, which serves over 1 million active monthly users across several integrated development environments (IDEs), relies on Google Cloud for its scalable, fast, and cost-effective AI infrastructure, services, and solutions.
Tabnine uses high-performing graphics processing units (GPUs) on Google Cloud to run data-intensive machine learning workloads with the low latency developers need when coding.Google Kubernetes Engine, a scalable, fully automated Kubernetes service, manages the infrastructure, ensuring greater operational and cost efficiency.
Google Cloud enables Tabnine to apply its models on companies’ private codebases while securing the underlying code. It also facilitates privacy and portability. Tabnine doesn’t store any of its customers’ data, and companies are able to run the AI-powered tool in different environments, including on their own virtual private cloud or on Google Cloud as a software-as-a-service (SaaS) application. Yung says that Google Cloud’ composable capabilities streamline new developer onboarding and allow Tabnine to deliver higher-quality services.
“[It] leads to efficiency. It leads to much faster speed, and it allows our customers to take a smaller chunk of compute while they’re getting started with Tabnine,” according to Yung.
How CI&T accelerated development 11% with Tabnine and Google Cloud
“We produce a lot of code,” says Luis Ribeiro, head of engineering and digital solutions at CI&T. “Additional efficiency and finding ways to improve the speed of how we develop solutions is crucial for us. That’s why AI is a really important and hot topic for us as we try to find more ways to bring innovation to our clients at a faster pace.”
Tabnine has boosted developer productivity for CI&T: The company’s developers accept 90% of the tool’s single-line coding suggestions, resulting in an 11% productivity increase across projects.
Ribeiro expects Tabnine to deliver even better results as it continually trains its models with CI&T’s data. He says the tool will be particularly useful for CI&T in highly regulated industries such as healthcare, life sciences, and financial services because of the data sovereignty it provides and the strong foundation of security and privacy Google Cloud delivers.
“We see the roadmap and the possibilities of improvements here with the technology,” Ribiero says. “As with anything related to machine learning, as much as you can give back to the modeling, it will have an opportunity to deliver more and more code. This is one of the secrets to keep using [the technology] in a better way.”