Unleash Your Development Superpowers: Refining the Core Coding Experience

Here on the Gemini Code Assist team, our primary goal is to make the moment-to-moment experience of writing and reviewing code more fluid, intelligent, and productive. We believe that a great developer tool gets out of your way while offering powerful assistance right when and where you need it.

This post highlights recent feature additions and improvements that enhance your core coding workflow. From writing the first line of a function to navigating your project, these updates are designed to keep you in a state of flow and help you produce high-quality code faster.

• Agent Mode with Auto Approve acts as an intelligent collaborator that understands your entire codebase, drastically simplifying large-scale changes like adding a new API endpoint across multiple files. The agent analyzes your request, proposes a multi-step plan for your approval and then executes it. To even further improve your efficiency on routine tasks, you can enable Auto Approve Mode, which lets the agent act on its plan without manual checks at each step. This transforms tedious, multi-file updates that once took hours into a single, automated command, powerfully freeing you up to focus on more complex architectural challenges.

• Inline Diff Views: To foster true collaboration between you and the AI, we've enhanced our code review experience with Inline Diff Views. You can now edit AI-generated changes directly in the integrated diff view, giving you precise, on-the-fly control to accept, reject, or modify every line. Furthermore, the feature displays modifications directly within your code, providing a clear, immediate understanding of changes in their surrounding context. This feature transforms the review process into a dynamic conversation with the AI, making it a fast and intuitive part of your daily workflow.

• Revert to Checkpoint: Experiment Fearlessly: Our new Revert to Checkpoint feature encourages fearless experimentation and risk-free exploration of complex changes. If you accept a series of suggestions from the non-agentic chat that doesn't work out, you're not stuck manually undoing changes across multiple files. With a single click, you can instantly roll back all affected files to their prior state before the issued prompt. This creates a powerful safety net, giving you the confidence to try ambitious refactoring, knowing that you can always return to a known good state in seconds.

• Multi-part Chat Code Suggestions We've streamlined how you apply AI-generated code from the chat panel. When Gemini suggests changes across multiple files or code blocks, multi-part code suggestions give you more granular control. You can review each change individually within the chat interface and decide whether to accept a single, specific code block or apply all suggested changes at once. This is especially useful for complex responses where you might only need a portion of the suggested code, ensuring you have the final say with every modification without ever leaving your editor.

• Chat Code Suggestion Preview Readability is key to a fast workflow. We’ve improved how code is presented in chat by displaying suggestions in clean, collapsible preview blocks. By default, you can see a snippet of the code and expand it when you need the full detail. This keeps the chat history clean and scannable, especially during long, complex conversations with the AI. It allows you to quickly scroll back and find previous parts of a discussion without getting lost in large code blocks, helping you manage screen real estate and find information more easily.

• Improved Code Completion Speed We’ve made significant under-the-hood improvements to our models and infrastructure to boost code completion speed. This isn't just about raw speed; it's about maintaining your cognitive flow and reducing frustrating interruptions. By suggesting the relevant code you need quickly, it allows you to complete blocks of code, implement entire functions, and import libraries without disruptive pauses. This quick feedback loop enables you to code closer to the speed of thought, keeping you focused on the problem and not the tool.

Each of the features we've covered was designed with one goal in mind: to make your hands-on-keyboard coding experience as seamless and powerful as possible. By removing friction from your daily tasks and further enhancing Code Assist’s AI-assistive capability, we hope to keep you in that creative flow state where great development happens.

Now, let's zoom out. A truly powerful developer experience is one you can control and tailor to your specific needs. The following features focus on the new tools and settings that allow you to customize your interactions with Gemini, manage your development context with precision, and stay informed about the tool itself.

• Manage Files and Folders in the Context Drawer Gain precise, temporary control over what Gemini sees with the new Context Drawer. This UI allows you to easily view, add, or remove specific files and folders from the context for your current conversation. For example, if you're working on a bug that only involves three files in a massive project, you can limit the context to just those files. This ensures that the AI provides highly relevant answers without being confused or cluttered by unrelated parts of your codebase, leading to faster and more accurate assistance.

• Create Custom Commands Take control of your workflow by creating your own custom commands. You can now define custom slash commands or save complex prompts that perform routine actions, such as generating a test file with your preferred boilerplate or refactoring a code snippet according to your team's style guide. This is about encoding your personal and team-specific knowledge into the tool itself. By doing this, you transform Gemini from a general-purpose assistant into a highly specialized tool that understands and automates your specific, recurring tasks, saving time and ensuring consistency across your projects.

• Add Code Snippets to the Chat Context Ask more targeted and effective questions by attaching specific code snippets to the chat context. Instead of describing a complex piece of code, you can highlight a function, class, or even a single line of code to focus Gemini's attention. This is far more efficient than pasting code into the chat box, as it preserves the context of the file it came from. This leads to more accurate explanations, refactoring suggestions, or bug fixes because Gemini has the exact code and its surrounding environment right in front of it.

• Add Terminal Output to the Chat Context Seamlessly debug command-line issues by adding terminal output directly to the chat context. Instead of copying and pasting long error messages or logs, you can now attach the output from your integrated terminal with a single click. This gives Gemini the full, unaltered picture—including error codes, stack traces, and command history—to help you understand and resolve the problem much faster. It streamlines the debugging loop for CLI tools, build scripts, and server logs by bringing the analysis right into your IDE.

• Configurable Codebase Awareness For more permanent context management, you can use AI exclusion files (like .aiignore) to configure your local codebase awareness. This allows you to permanently instruct Gemini to ignore specific files or directories—such as build artifacts, node_modules, or files containing sensitive keys. This not only improves suggestion relevance by focusing on your source code, but it also enhances security and privacy by ensuring Gemini only ever learns from the code you explicitly want it to see, giving you total control over your project's context.

• Stop In-progress Chat Responses You now have more control over your chat interactions. A new stop button allows you to halt an in-progress chat response at any time. This is useful when you realize you've asked the wrong question or the answer is heading in a direction you don't need, saving you time and letting you quickly pivot without waiting for the full response to generate. It’s a simple feature that respects your time and makes the conversation with Gemini more efficient and responsive to your immediate needs.

• Thinking Tokens To provide more transparency into what Gemini is doing, we now show "thinking tokens" while it's processing your request. This visual indicator gives you confidence that your query is being actively worked on and improves the interactive feel of the chat experience. It removes the guesswork of whether the assistant has stalled or is working on a complex problem, providing the same kind of feedback you'd get from a human colleague who is pausing to think, which makes the interaction feel more natural and less like a black box.

• Release Channel Name in Chat Banner Stay informed about the tool you're running with the release channel name now displayed in the VS Code chat banner. This small but important detail helps you know whether you're on the Stable or Release Candidate channel. This transparency makes it easier to track when new features will become available to you, understand potential behavior changes between versions, and provide more accurate feedback to our team, which is crucial for bug reports and feature requests.

We’ve explored the evolution of your AI collaborator—from an assistant that speeds up your core coding tasks to a fully customizable partner that adapts to your unique workflow. Our mission is to build Gemini Code Assist that not only understands code but understands how you code. By refining the immediate, in-editor experience and providing deep customization, we aim to make your development process more intuitive, creative, and powerful.

We're incredibly excited to get these tools into your hands. Please dive in, experiment with the new features, and continue to share your feedback. The future of software development is collaborative, and we can’t wait to see what you build with Gemini Code Assist.