Google AI Studio is a great platform. We all agree that it made Android Development and Web Development easier, but it needs a full makeover due to its existing bugs and limitations.
Here is my proposal for the updates:
A: Harness Enhancements
Google AI Studio has a harness and can perform the following:
Read File
Explore Codebase (list directories)
Edit File and Multi-Edit File
Run Command
Set up Firebase and deploy rules
There is one crucial thing missing:
A set of Context Management Tools and Features.
Feature: Context consumption tracking, or you can make a tiny bar that shows the context. This would also be a nice option to see if your context is degrading.
Tool A: Compress Context (similar to Claude Code).
Tool B: Remember. It remembers specific parts of a project, similar to ChatGPT’s memory. This helps the AI pivot its work within its active context.
Tool C, which is the most important: Deploy_Subagents.
The idea is really simple: Gemini, with its large context, acts like a big manager that orchestrates smaller teams of agents. These teams would have a smaller context window and an understanding of the architecture, ensuring nothing problematic occurs.
They can make changes, especially if the user says they want the highest performance or effectiveness.
Tool D: Reset_Context (DANGEROUS). This would be at the user’s discretion. If the user has documentation, Gemini must always obtain user permission before using this tool.
The AI instance or session would be reset. The history would remain, and the code would still be present.
The AI would then start fresh. Gemini is actually very good at understanding codebases; previously, I migrated my project in another chat and managed to wire everything up.
But the problem here is that Gemini doesn’t document.
Even when I tell it to do so, the documentation gets deleted, so there must be a special documentation system that records decisions, etc., for the app. This documentation feature is currently missing. So, I think the Reset_Context tool should only be enabled for people who are aware of the potential consequences.
I also suggest one more thing: The AI might forget how to use its tools, so a small file in its own directory could be created for it to refer to if it forgets how to use these tools.
Other features I recommend for the harness include ‘Google Security’.
It is simply a tool that scans your codebase for vulnerabilities, ensuring its security and addressing even minor or major details, because AI typically doesn’t create secure apps by default; users usually have to explicitly instruct it. For the last feature, I recommend adding software development integrations, such as full integration with Git and Firebase CLI, and the option to integrate with our own CLIs. For instance, if a user is making a project with NeonDB, the same applies to hosting. If the user doesn’t want to host on a cloud service, they could do it in other ways, test their code, and view logs. I know that Gemini is typically within the Google ecosystem, but I don’t think that’s the best option, especially if you want to implement your own integrations or workflows.
What’s left is a remake of the user experience. For instance, the user can see how much quota of the free tier they are spending, how much of the paid ones, etc. I think what would make Google AI stand out is simply that if it can use MCPS and CLIs, and if its harness complements Gemini, it would be the best. Aside from that, I really wish a dedicated part of the upgrade would go into making Google AI Studio the best harness and the best model. Also, maybe in the future, can we get the ability to make and host our backends on Google’s infra using Google AI Studio, or perhaps be able to code in multiple languages (Py, TS, C are the main ones); that would be absolutely great.