Google has unveiled ‘Gemini Code Assist Enterprise,’ an extension of its existing Gemini Code Assist platform aimed at tackling developer skills shortages.
Citing a “dearth” of skilled developers, Google said that its new platform will move past AI coding assistance in the integrated developer environment (IDE) to wider application development assistance bolstered by a deep local codebase awareness.
This is made possible by Gemini’s large token context window, allowing developers to create more relevant code with assistance that considers the specifics of codebases and current development sessions.
Code Assist Enterprise understands a user’s local codebase and, through ‘code customization,’ can suggest code based on organizational best practices and internal libraries.
“Developers can stay in flow state longer, bringing more insights directly to their IDEs, while also completing complex tasks like upgrading a Java version in an entire repo,” Google said.
The platform can index repository from both GitHub and GitLab, with support for additional source control systems and self-hosted, on-premises repos coming in early 2025.
Google said that Code Assist Enterprise does not feed proprietary organizational data into the Gemini model and that source code stored for use in code customization is kept in a Google Cloud managed project.
This cloud environment is isolated within each customer’s organization. Customers also have complete control over what source repos are used in customization processes, with the option to purge data at any time.
The tool is now generally available, with prices starting at $45 per month per user, though it is available for a discounted $19 per month per user on a 1-year subscription until March 31, 2025.
The code assistant gold rush
Google first released Gemini Code Assist at Google Cloud Next 2024 in Las Vegas with a huge million token context window that put it way out ahead of other code assistants such as GitHub Copilot.
The other tools are still developing, though – in late April 2024, GitHub rolled out GitHub Copilot Workspace which is designed to be a complete development ecosystem for developers.
There’s also Meta’s Code Llama tool which saw a big uptick in size at the start of this year that scaled it up to 70 billion parameters, and AI startup Mistral’s code assistant competitor, Codestral.
Question marks loom over all these platforms, however, with research suggesting that AI reduces code quality and maintainability despite speeding up code completion times.
