Hey DEV community! 👋
We're the team behind Google AI Studio and the Gemini API at Google DeepMind.
We'll be answering your questions live on August 28, 2025 starting at 1PM ET.
Thank you to everyone who participated in our AMA! We'll do our best to keep answering questions asynchronously throughout the next few weeks so check back later if your question wasn't answered!
Who we are:
- Paige Bailey (@dynamicwebpaige): AI Developer Relations Lead
- Patrick Loeber (@pat_loeber): Developer Relations Engineer
- Alisa Fortin (@alisa_fortin): Product Manager
- Ivan Solovyev (@ivan_solovyev): Product Manager
- Kate Olszewska (@kate_olszewska): Product Manager
- Vivian Jair (@vivjair): Product Marketing Manager
- Dave Elliot (@dave_elliott): Head of Developer Advocacy (Google Cloud)
What we work on:
- 🤖 AI Studio: Our developer platform where you can experiment with Gemini models, including access to our latest experimental releases.
- 🔧 Gemini API: APIs that serve millions of developers and process trillions of tokens.
- 🎨 Multi-modal & Open-Source Models: Advanced AI models including Veo (video generation), Imagen (image generation), Lyria (music creation), and Gemma (open-source language models).
- 📚 Developer Experience: Making Google's most advanced AI models easier to integrate and use.
- 🌍 Community: Building resources, documentation, and support for the global developer community.
Please don't ask us about:
- Unreleased Google products or detailed internal roadmaps
- Proprietary technical implementations
- Confidential business information
- Personal/private information
Get started with AI Studio:
If you haven't tried AI Studio yet, it's the easiest way to start building with Gemini. You can turn on features like code execution, use extended context (2M+ tokens), and access our latest experimental models - all for free to get started!
We'll be rotating through answers throughout the day, so you might hear from different team members. Let's dive in! 🔥
Top comments (161)
Curious if there are upcoming releases for Gemini CLI. In my tests it’s excellent at whole-repo analysis and strategy, but it often stumbles in execution (tools break and it loops).
Are any major releases planned? What kind, and on what timeline?
And will there be multi-agent support?
Hey there! Am glad to hear that you've been using and loving the Gemini CLI (us, too! 😄).
This update is via the Google Cloud folks who are building out the CLI:
Please how do I get to work for Google 🙏.
Vector Databases and VR Question:
Do you forsee AI/vector databases being gamified in such a way that we can throw on a VR headset and 'swim' through the vector database, so to speak, sort of as a fun way to explore and retreieve data?
I'd like to try that, sounds fun.
Thanks.
I love the idea of being "immersed" in your data, and to use 3D space as a path to spot unexpected relationships in datasets! In addition to the recommendations from other folks on this thread, you might also be interested in checking out the Embeddings Projector, as a fun way to view and manipulate data in 3D space.
Please how do I get to work for Google 🙏.
Thats awesome!
Are these mostly used for demo or are they useful for practitioners?
Wow
Excellent idea!
But the main challenge is how to display 512+ dimensions of embeddings in 3D VR space?
Perhaps through interactive projections or using additional channels (color, sound, vibration).
Hi Prema, thanks for your response.
Im assuming it would be approached by taking the overall (x,y,z) of each individual vector and assign it some set volume in space, with some padding, so a user could navigate through the 'cracks'.
It would essentially be like swimming through a gas, but the molecules are ginormous so they are visible to the user... big enough that a user could select each one to see the details.
But small enough to sneak by each one as they gently nudge out of the way but return back to there normal position.
I think this could be done several ways. In my expereince tools that come to mind right away are blender and three.js! Haha.
Could even have a temperature map overlay, so a user could 'jump in' and explore search results based on their custom query and see how closely they are related. Or perhaps a pattern overly, to be accommodating for more users?
You know what. This would be really awesome for music exploration.
Former game dev here. Blender is 3d modeling software, not really ideal for your use case. I just wanted to say that if you have a big idea like this, your often better off to try to make it yourself.
There are a couple of game engines that are free to use such as Unreal and Unity that provide VR support as well as plenty of online resources.
I would recommend Unity for this due to a combination of community support regarding tutorials, and it using C# as it's primary coding language. Most AI is pretty good at writing C# scripts (As long as you keep them modular) so you don't need to be a master programmer.
You might even enjoy learning how to use the game engine. In regards to visuals, you would also want to learn Blender for the 3D assets.
I don't foresee Google making anything like this as it's very niche and they prefer broad strokes, not to mention they had a pretty massive failure in the game industry and likely aren't looking to try again (Stadia).
Thanks for your wide-lensed feedback. I have used unreal engine a bit but not to any major extent. Any reason you would use unity over unreal for something like this? Based on your answer, sounds like my orignial question is at the very least, possible.
I can sort of picture your concept in my head. Unreal is a lot more complex, for me at least, in regards to setting up a system like that cause your options are their visual blueprints or C++ and the engine itself is pretty heavy on resources. Unity is lighter and i think scripting a system like that would be much easier in C# as long as you can optimize it.
You could probably just instantiate new nodes as your going along and cull anything thats out of view. Since its VR its going to be a bit heaftier to run so the smaller engine would likely be more stable for the average person :)
My discord is on my profile if you want to discuss it more over there. I don't really have any other socials lol
When will it be possible to vibe code with Google Apps script? Thanks
Thanks for the question!
You can already use the Gemini APIs and Gemini in AI Studio to generate Apps Script code, which you can then pull into Google Workspace products (like Sheets). The Google Cloud team also has a few codelabs showing how to use the Gemini APIs with Apps Script (example).
Can we have a virtual hackathon solely focused on building AI apps in ai.studio?
I love that! We’re planning to run more hackathons later this year and I'll make sure to forward that idea!
Yes!
On DEV!
I wish to work for Google as a C++ developer scraplinkecomarket.netlify.app
My work with html and css and js.
Stay tuned here - may or may not have something coming soon!
I really have keen interest in drug development and personalized medicine using AI. My master's thesis was on find suitable drug candidates for PSP using graph neural networking and other AI techniques. I did use DeepMind's Alphafold2 also in it. I learnt everything by myself for it through online resources. But I feel overwhelmed with the vast number of online resources, and they are not that helpful to make a proper plan with tangible result to get better in the domain. So, if I want to one day work in DeepMind and be part of novel drug discovery, what are the steps I need to take?
It’s great to hear that you’re interested in AI for drug discovery! Google DeepMind, Isomorphic Labs, and our colleagues in Google Research are all investing very heavily in AI for health and the medical domain.
The skill sets that you would need would depend on the role that you would be interested in taking - for example, engineering, product, research, marketing, and more are all role profiles that we hire for in our AI for health orgs. For each of those focus areas, I would recommend that you continue building your expertise in AI and in the medical / life sciences, and make sure to share your work visibly - either via GitHub for open-source and software projects, or by publishing the research that you've been pursuing.
I'd also recommend building on or evaluating some of the open models that Google has released in the healthcare space, like TxGemma and MedGemma. Good luck, and am looking forward to seeing what you build!
I wish to work for Google as a C++ Developer 🙏.
I am an AI engineer by profession. So any specific guidelines I can follow to attain a position at Deep Mind in the Drug Research group? To attain an interview call, or what all should I prepare etc.
What would it take to intern as a devrel for DeepMind?
We regularly have engineering and product internship roles available in Google and at Google DeepMind! I recommend checking out our careers pages, and searching for "internship".
If you’re interested in a career as a developer relations engineer, I would recommend building in the open - contributing to open-source projects, sharing your work publicly (on social media, and on GitHub) and investing in supporting your local and online developer communities. Many DevRel folks start their careers as software engineers, and then gradually move to a more community-facing role.
On this subject, how do you think the idea of internships will evolve in the future? There's so much written about how AI is particularly affecting entry-level jobs. What do you think needs to change for employers to be able to best support this kind of work?
How does 'Search-grounded' mode work under the hood—are citations confidence-weighted and deduplicated? Can we constrain freshness windows, force certain domains, or provide our own corpus for grounding?
The secret sauce is the same as Google Search because the tool relies on the Google Search Index. Currently, the groundingMetadata does not expose a direct confidence score for each citation. The presence of a citation indicates the model found that source relevant for generating a specific part of the response. In terms of deduping, the system generally attempts to provide unique and relevant sources. While you might see citations from different pages on the same domain if they each contribute distinct information, the goal is to provide a concise set of the most useful sources rather than a long list of redundant links.
For bring your own search scenarios, try using function calling with RAG flows
In terms of working under the hood, the first thing the tool will do is analyze your query. For example, a prompt like "Who won the F1 race last weekend?" will trigger a search, while "Write a poem about the ocean" likely won't. The model then formulates one or more search queries based on your prompt to find the most relevant information from the Google Search Index. The most relevant snippets and information from the search results are fed into the model's context window along with your prompt. The model uses this retrieved information as its source of truth to generate a "grounded" response. The API returns the response along with groundingMetadata. This metadata includes the source URLs for the information used, to build citation links back to the original content for verification.
We are working on a filter to constrain to date ranges. You cannot force certain domains (use URL Context for that), but you can exclude some domains from search. The “Bring your own search” option is available through Vertex.
How influenced are you by the work done from other companies (i.e. OpenAI releasing GPT-5 recently etc)
It's always inspiring to see the recent surge in AI development – both in the modeling and product space! 😀
At Google, we ensure many different closed (ex: Anthropic) and open models are available to our customers on Google Cloud via the Vertex AI Model Garden. We also support many of the research labs via both our open machine learning frameworks (JAX) and hardware (TPUs and GPUs) for training on GCP, and have been excited to see many startups and enterprises adopt the Gemini and Gemma models.
Our DevX team has also been hard at work adding or improving support for the Gemini APIs and Gemma models into developer tools (like Roo Code, Cline, Cursor, Windsurf, etc.) and frameworks (LangGraph, n8n, Unsloth, etc.). More to come, we all go further when we're working together as one community.
What advice do you have for someone who is considering signing up to a CS Bootcamp vs. going all-in on building with AI tools?
Great question, and I know a lot of folks have this top-of-mind. 👍🏻
For programs like a CS Bootcamp or attending a university, I'd say the biggest value that you're really getting is the in-person community. Many educational structures are still catching up to state-of-the-art in AI and in building product-grade software systems, so the coursework you'd be completing might not be aligned with the latest model and product releases - and those features / models are changing minimally weekly, if not daily, which makes it a challenge for educators to keep their curriculum up-to-the-minute.
To build up expertise and the skill set for working with AI systems, I would strongly suggest to just start building: find a problem that really bugs you, use AI to automate it, and then share your work visibly externally -- via GitHub and social media. This is a really useful way to get product feedback, and to get inspired! There are also frequently AI hackathons happening, either in-person or online (ex: the Major League Hacking events list and DevPost are great places to look).
You can also check out DEV Challenges 😇
Hi. I can’t turn my Gemini-powered voice assistant into a real product because the input price for audio on Gemini 2.5 Flash is $1.00. Any plans to make audio input pricing more like text input? (My assistant also pays for Google Cloud TTS). Who should I contact for this and to request higher usage limits?
While we don't have any immediate plans to reduce the audio input price, we're always looking for ways to make our models more accessible.
Feedback like yours directly influences our future plans, so thank you for raising this.
For higher usage limits, the best next step is to fill out the request form here: docs.google.com/forms/d/e/1FAIpQLS...
Some comments may only be visible to logged-in visitors. Sign in to view all comments.