New in Gemini Code Assist: Agent Mode and IDE enhancements

Last month, we released numerous updates for Gemini Code Assist, Google’s AI coding assistant for both individuals and businesses. These updates include the Gemini CLI and Gemini Code Assist agent model, as well as IDE enhancements that make the day-to-day workflow smoother and more intuitive.

Individuals can get started for free by installing the Visual Studio Code plug-in or JetBrains IDE extension. Here’s a closer look at what’s new.

Agent Mode

Google recently released two major AI updates for developers. Gemini CLI brings AI directly into the command line, eliminating context switching by letting developers ask questions, generate code and debug right in their terminal.

Taking this further, Gemini Code Assist's new agent mode acts as an AI pair programmer. It analyzes your entire codebase to plan and execute complex, multi-file tasks — like implementing new features or performing large refactors, all from a single prompt, significantly accelerating the development process. Think of it less as a tool and more as a way to delegate your tasks. Developers in the Insiders channel got access to agent mode last month.

Agent mode's power is its comprehensive project understanding. It analyzes the entire codebase, not just the open file, to model your application. This includes its architecture, dependencies, coding patterns and the relationships between different components.

Because suggestions are now contextually aware, the agent knows how to implement features across all necessary components. It respects your existing coding style, leading to higher-quality, consistent code. This project-wide awareness means fewer errors and less time spent fixing mismatched suggestions or explaining context.

With a full understanding of the project, Code Assist’s agent can then enable multi-file editing. Before this, implementing significant features may have been a piecemeal process. You’d generate code for one file, then another, manually stitching everything together. With agent mode, you can now make a single request, and Code Assist will orchestrate all the changes across the codebase.

Large-scale tasks that once took a lot of manual effort can now be initiated with a single prompt. Consider prompts like these:

• Large-scale refactoring: "Update all API endpoints to use new authentication."
• Feature implementation: "Add a new full-stack user settings page."
• Dependency upgrades: "Update a core library and fix all breaking changes.

Relying on an AI coding assistant may feel daunting — but when using Code Assist, you are always in control of the agent mode experience. The agent never acts blindly. Before modifying any code, it presents a detailed plan, outlining the files and summarizing all proposed changes for your final approval.

You can review the agent's plan, ask for clarification, suggest alternative approaches and approve or deny any change. This collaborative loop combines the speed and scale of AI with your domain expertise and architectural vision. It ensures the final result is not just functional but is precisely what you intended.

And for added peace of mind, the June update introduced the ability to revert to a checkpoint. If you accept a series of suggestions and find they aren't quite right, you can easily roll back all affected files to their state before the changes were applied, encouraging fearless experimentation.

IDE enhancements

Last month, we released an update giving you more granular control over context. Code Assist now automatically enforces your .gitignore and lets you create a .aiexlude file to ignore sensitive or legacy code. Conversely, you can focus the chat on specific code snippets for more precise questions. You can also now attach terminal output directly to the chat to ask about commands or debug errors, eliminating the need to copy-paste logs.

These features give you precise control over the signal-to-noise ratio. By excluding irrelevant files and focusing on specific snippets or terminal logs, you get faster, more accurate and more relevant answers, dramatically speeding up debugging and analysis.

We’ve also made the chat experience more responsive and intuitive. For better readability, code suggestions now appear in clean preview blocks, which you can configure to be collapsed or expanded. Navigating your project is also easier, as filenames Code Assist mentions in chat are now clickable links that instantly open in your editor.

For long responses, the chat now scrolls automatically as Gemini Code Assist types, though you can disable this if you prefer. And if you ever ask a question and realize it was the wrong one, or the response is taking too long, you can now stop the in-progress chat response immediately, saving you time and frustration.

These refinements make Gemini Code Assist smoother and more efficient to use. You spend less time scrolling, parsing text and manually navigating files, allowing you to stay in a state of flow and focus on what matters most: solving problems.

Try Gemini Code Assist for free, and learn more about our recent updates in our release notes.