User: Make a checkpoint.
Gemini 3 Pro Preview
Thought for 58 seconds
To establish a solid checkpoint for your application and ensure the layout meets your requirements (Header, Left Sidebar, Main Area, Spacer), I have enhanced the App.tsx file.
I interpreted the incomplete sentence "The sidebar ca..." as a request for the sidebar to be collapsible. I have added a toggle button to the header and logic to show/hide the sidebar and its resizer (spacer), giving you a flexible workspace.
Here are the updates:
Sidebar Toggle: Added a PanelLeft button in the header to toggle the sidebar's visibility.
Conditional Layout: The Sidebar and the Resizer (Spacer) now gracefully hide when toggled off, allowing the Main Area to expand fully.
I don’t get it. I want the app to just set a checkpoint to keep my code save as I might want to return to it. (If only AI Studio had an option to set specific versions as restore points…)
What does Gemini do? It tries to add functionality that I had previously removed yet it somehow remembered that I once asked for a resizable layout. (It was ugly, so I had it removed.) And instead of just setting a checkpoint, it started changing app.tsx which tends to fail sometimes. The AI tends to mess up the code and then my app is broken and I need to go back to the previous checkpoint. Which is why I wanted the checkpoint in the first place…
So I wonder… I ask for a checkpoint and the AI decides to do something completely different, with the potential of destroying my work. I stopped the AI in time, but this is annoying.
The problem is that when I start with a clear chat, I immediately want a checkpoint as the first prompt I make could damage the code so I want to go back to a restore point. But it only makes restore points after my first prompt. This is a serious dilemma…
Hey! The way vibe coding in AI Studio works today is that each conversation turn acts as a checkpoint. You don’t have the ability to manually set a checkpoint. The best way to get what it seems like you are going after is copying the app you are working with in the top right corner.
The problem is when you clear the chat so all checkpoints are also gone. When you then ask the AI to do anything, there’s a risk that the code gets damaged by the AI and then there’s no checkpoint to restore, as checkpoints are set after the prompt executed, not before. I’ve lost several features in this way.
Not cleaning up the chats also causes problem as the AI looks at the whole history and might hallucinate and do something you asked before. It can get quite messy after a while.
Hi Logan, I would like to propose a new feature for AI Studio: an “approve before commit” mode (can be enabled/disabled).
Currently, the AI often moves directly to implementation (due to its “completion bias”). While efficient and nimble, introducing an optional checkpoint where the model must first state its intent (the why, how, and what) before modifying the codebase could be a significant workflow improvement for many vibe coders.
This would be a first-mover advantage for Google AI Studio, as very few, if any, other vibe coding tools (apart from professional IDEs) offer such a feature.
Motivation
Predictability and control: By forcing the AI to outline its plan first, users can catch logic errors or architectural misalignments before a single line of code is written. This reduces the “trial and error” loop often found in AI-assisted development. Would also avoid having to roll back to a restore point in many cases.
A bridge for “vibe coders”: For users who lead with high-level intent rather than deep technical syntax, this mode would serve as a powerful educational tool. Seeing the AI explain the implementation details helps these users bridge the gap between “vibes” and concrete engineering principles.
Reduced technical debt: An explicit approval stage encourages more intentional coding, preventing the AI from introducing unnecessary dependencies or convoluted workarounds that can happen when it acts autonomously. Would also encouage users to be more precise in their prompts.
Proposed workflow
Drafting: The AI analyzes the prompt and outlines its proposed changes (how & what) in plain language or pseudocode.
Justification: The AI explains why it chose this specific approach.
Approval: The user reviews the plan and provides a “go-ahead” or requests an adjustment.
Execution: The AI commits the code changes only after receiving explicit confirmation.
This feature would make AI Studio not just a faster way to code, but a smarter, more transparent partner in the development process.
I discovered that AI Studio has a button “Open version history” hidden between the other buttons in the top-right. Is that new?
Anyways, this button allows me to go back to a previous version of my code, which is more or less what I want. It could use a nicer user interface, though but it works, so I’ll stop complaining.
Anyways, I can now clear the chat history regularly, which is needed as the AI gets confused by it and then tends to destroy my code. So now, when this happens, I can return to a previous version, clear the chat and try again…
