The latest Gemini model dropped as a preview yesterday, and we need your help on three things:
Testing our Gemini AI chat capabilities in our code-first workspace
Using it to vibe code an app
Sharing the app you created so that we can potentially use it as part of our internal and external “look at the cool thing the community has built” kind of content
To easiest way to get started is to start your app from your favorite template and change the model to the latest (gemini-2.5-pro-preview-06-05):
What is the “code-first” workspace? I’ve tried googling and searching here on the forum and I don’t come up with anything. Do you have a link on this, or is this just the Gemini chat tab?
Should I really have to tell Gemini which file I want the change in based on this scenario? It is friction like this that I find it hard to use Gemini in the code-first workspace since it doesn’t seem intuitive and know what the focus is and doesn’t line up with the visual aspect of coding. As a creative technologist, I don’t want to have to flip to the actual file, to be honest, I’m not really sure where that code is since it was just created by my initial prompt; so having to go find it to tell Gemini is bothersome and I would rather just go back to the prototyper if Gemini is going to behave like that.
ok, is there a firebase article to guide on how to troubleshoot this step as it relates to firebase users? The error message getting directed to a Google Developer article for Rate Limits isn’t helpful.
Remember, the users for a Vibe Coding platform is a little less technical, so the documentation is going to be a little less “Google-ify” and more n00b friendly. The experience of going to the Google Docs feels very enterprise and sterile; which is the opposite vibe of Firebase Studio; so the context switching of working in Firebase for software development, getting hit with not knowing how to do something, and then going to the Google Docs, its like you’ve been taken out of a vortex and hit back into reality. Nothing makes me want to stop prototyping more than getting to a point where I have to read the docs, generally speaking, but especially for these docs.
Ok, when I first started, I tried using Gemini chat capabilities in the code-first workspace because I couldn’t figure out how to use the prototyper. I had built an app in the normal Gemini App and got to the point where I needed more than a SPA so I found my way over here; and it was a full day to figure out how to get the same flow that I naturally was getting in the Gemini App for building an app.
Its because the transition from Gemini App to here, it wasn’t obvious the difference between Prototyper and Code-First workspace and that they would act so differently. And since I already had an app built, I naturally tried to import it and that is why I was doing so much in the code-rist workspace. But it felt insurmountable to get the app I had developed to work in the code-first workspace.
Now that I’m back to trying it again, I’ll give you some of my previous observations and let you know if they seemed to have changed with the 06-05 model.
I was constantly getting hit with the code not being able to apply the activity in the code. I basically revereted to use Gemini as a pair programmer at that time.
There felt like more tension/friction in the code-first workspace compared to prototyper because I would ask it to do something – it would prep to do something – and then annoyingly it would then ask me if it was OK to do the thing that we had just discussed … for every single little thing
– (I would prefer to have that behavior as something I turn on through a command function if there is a particular feature that needs more scrutinty vs simple UI change that is easily reversible)
The user experience with every time opening the code-first workspace, it was a brand new Gemini chat; as opposed to the prototyper has a record displayed of our chat history everytime I go to it makes me feel like I’m starting over every single time instead of just continuing from where I left off.
=================================================================
Thoughts while using:
I suggested an edit and then it just replied with the updated files; with no natural language description of what is changing
– Ok, so this is the annoying thing; as a user … I end up with at least 2 actions to complete per instruction that I give it. Once I clicked Update File, it did it, and THEN gave me info on what it did and then I had another action to do. It makes me wonder if there are certain actions like “deleting files” that are super sensative that whenever that kind of action occurs, since it is considered a destructive action, that the vibe flow gets interrupted so that the user takes some responsibility for it. If so, it would be more helpful to understand why the flow is getting interupted if the “Delete File” button was red as an indication of the destructive action and make me less annoyed about my flow being interrupted since I would have a mental model that “flow gets interrupted for red action things”
Preview 06-05 in Firebase Studio is worse in coding than 2.5 Pro in Gemini app.
Case in point: Gemini in FS keeps using Genkit 0.5.x API syntax and concepts while package.json points at 1.2.1, airules.md includes direct links to Genkit.dev pages and I included a sample app code to guide the model.
Is the Gemini in FS capable of reading external documentation? It is a hit and miss now.
Played around with both prototyper and latest Gemini model in code-first workspace to vibe code a Splitwise clone that leverages Firestore. Here are a few things I noticed:
The chat clears in code-first workspace when you toggle back and forth between Prototyper and Code. Not sure this is that big of a deal.
There was a general feeling that Prototyper was more helpful in terms of being able to operate with very vague and non-specific prompts.
At least in code-first workspace, it didn’t seem to understand that we were operating in a non-standard development environment. It took more effort than I had expected to connect to Firestore. This was in part my fault due to the original starting prompt. But even so, getting it to use Firestore took some wrangling.
I’m still blown away how quickly I was able to get a functional app hosted and up and running using Firestore all without even needing to run firebase init. Will definitely be trying to vibe code something else to see if I can get something up and running even faster.
For #1, preserving the context between prototyper and code-first workspace is something we are actively working on addressing.
For #2, that is very fair feedback. This is something we’ll look into further refining.
For #3, I’ll flag this for @davideast to be aware of
Do you have a link to your app that you can share? We’d love to be able to highlight this as part of apps that are created using our latest under development Gemini model.
Great question! It is the toggle we provide for you to load your under-development app in our VS Code environment, terminal, and other related capabilities you’ll want to do more coding-related activities.
It’s still very much a proof-of-concept (the UX was entirely conceived by Gemini); no polishing has been done as of yet. A tight integration with Stitch could potentially be quite powerful though it’s hard to conceptualize exactly how that might look in practice.
I think there’s potentially a missed opportunity here to acquire/source new customers. I can imagine a “Copy to Firebase Studio” or similar so users can keep the momentum towards building out their idea.
Tried it, didn’t realize I was going to get charged for trying it. Worked okay, but cost me $30 for marginally better results. Switched back immediately. Unless I’m wrong about the charges, you should warn people of the costs.