Google today unveiled updates to Firebase Studio at its Cloud Summit event in London, adding Gemini command-line interface (CLI) integration, initial Model Context Protocol (MCP) support, and “Agent Mode.”
Agent Mode features a combination of familiar and new elements. There are three distinct levels of AI collaboration. “Ask” is a conversational mode for brainstorming and planning. There’s also a human-in-the-loop agent, where Gemini will suggest code changes, but a developer must approve them before they are applied. Finally, there is a second agent mode, which is almost fully autonomous – permission is still needed for sensitive actions – but Gemini can write its own code, fix errors, and build features.
The updates were showcased during the second keynote of Google’s Cloud Summit event in England’s capital city and represented a significant improvement over the April release. Google claimed that “millions” of apps had been produced using the platform, although was more coy when it came to breaking down that figure into developers experimenting with the service versus those using it for production work.
We tried it out and, while we were able to bring up an application using prompts, care was needed in how those prompts were crafted. The service is not yet at the point where a non-engineer can use it to churn out a polished application, though it is edging closer.
As for the other updates, Gemini CLI integration is a welcome addition for developers who prefer to work from the command line and dislike switching windows, and the MCP integration is handy, although arguably something that a developer would expect to be present as part of an AI-powered toolset.
Firebase is all about AI agents – the Google definition of the term in this instance is having the code built autonomously without requiring human approval. Useful for quickly producing a prototype or proof of concept, but still some way from being production-ready. For example, there is no way to apply role-based access to users – an administrator might want some users to have read-only rights.
When we last looked at Firebase, it showed promise, but there was also plenty of hype (as is the norm with AI). This latest release goes some way to making good on the promises. The apps and services that can be created remain relatively simple, although the complexity has increased, and what is created runs in Google’s Cloud. However the pace of development is impressive.
Plenty of coding assistant tools exist, but Firebase’s effort to close the loop hints at a future which could keep developers up at night – or, if the AI evangelists are right, usher in a new era of productivity. ®
The Register attended Google Cloud Summit in London at Google’s invitation.