How did Google achieve "AI counterattack"? CEO Pichai discusses Gemini 3, Vibe Coding, and AI full-stack strategy

On November 26, Logan Kilpatrick from Google DeepMind had an in-depth conversation with Google CEO Sundar Pichai, discussing the release of Gemini 3, Nano Banana Pro, and Google's overall momentum in the AI field. They talked about Google's long-term investment in infrastructure and the rise of Vibe coding, with Sundar also sharing insights on the future of quantum computing.

Pichai revealed that Google established its AI-first strategy as early as 2016, when breakthroughs based on Google Brain, the introduction of DeepMind, and the success of AlphaGo led the company to adopt a full-stack approach. This long-term perspective allows Google to achieve an innovation stacking effect at every layer of the technology stack, from infrastructure optimization to model training and product application, forming a complete technological closed loop. Pichai emphasized that Google's full-stack innovation from chips to applications is creating a multiplicative effect, with Gemini becoming the core link across all product lines.

The newly launched Nano Banana Pro has sparked enthusiastic responses in the market, with users showcasing incredible creativity through infographics created using the model. Pichai believes this reveals the latent creativity within people, and Google is providing tools for more individuals to express their ideas. He specifically mentioned the phenomenon of "Vibe Coding"—AI is lowering the barriers to programming, enabling non-technical individuals to create applications, a capability comparable to the impact of blogging and YouTube on writing and video creation.

Additionally, Pichai mentioned progress in quantum computing, stating that in about five years, we will be incredibly excited about quantum technology, hoping it will be treated as we do with artificial intelligence today.

Here are the key points from the conversation:

• In 2012, Google Brain achieved breakthroughs in image classification with the famous "cat paper"; in 2014, we introduced Google DeepMind; January 2016 was the moment of AlphaGo. Then in May 2016, we announced our first TPU. Looking back at 2016, I was very clear that we were about to experience another platform shift. That was the bet on the full-stack approach, positioning Google as an AI-first company.
• When you adopt a full-stack approach, the innovation at each layer permeates the entire system up to the top layer, creating this multiplicative effect.
• From Cloud to Waymo, to Search, and all other products like Gmail. And now it feels like Gemini is the thread that genuinely connects every Google product
• Nano Banana Pro has crossed the chasm, especially in infographics and fact-checking in conjunction with Google Search. I can clearly see how this could become part of Google's mission—to organize the world's information and make it universally accessible—through those infographics.
• Developers are excited about Flash because it allows you to serve more people. I'm also excited about Flash 3.0; I think it will be a very excellent model and possibly our best one yet.
• Quantum computing is an excellent prediction. I believe that in about five years, we will be as excited about quantum computing as we are today about artificial intelligence. By 2027, we might be able to deploy some TPUs in space, and who knows, we might even encounter a Tesla sports car flying in space.
• Vibe Coding is almost like the rise of the internet, with blogs suddenly appearing and more people becoming writers, along with YouTube, where more people became creators. You can feel this change in the programming field, and even within Google itself.

Here is the full conversation:

Logan Kilpatrick:

Hello everyone, welcome back to Release Notes. I'm Logan Kilpatrick from the Google DeepMind team. Today we have invited Sundar Pichai, the CEO of Google and Alphabet. We are in Mountain View. Gemini 3 has been released, and Nano Banana Pro has also launched with very positive feedback. So I was wondering if you could summarize this moment of progress we've made? Nowadays, we not only have top models like Gemini and Nano Banana Pro, but also Vo and other music models, flourishing all around. It feels like the longer time goes on, the more things emerge. So, yes, I don't know. Would you like to paint a picture of this moment for us?

Sundar Pichai:

First of all, it's great to be here. I want to say it's been an extraordinary week. You know, when you're doing R&D internally, you envision that moment when you can truly share all the results with the public. There's nothing more exciting than that moment when you're committed to a product. That's what this week represents.

But I think this is built on years of foundation and all our deep investments. It's always been clear to me that you can see the speed at which we are making progress. But all the seeds have come together, and that is indeed very special. Just in the past few weeks, I've been reflecting, and I feel like we are releasing new things almost every day. So it's a wonderful feeling.

Logan Kilpatrick:

Indeed. I remember chatting with you about a year and a half ago when I was complaining about something. I was definitely complaining about something, and then you said something that impacted me to look at things from a long-term perspective. I'm curious, for me, the story of reaching this moment—having top models and all the infrastructure to scale Gemini across Google—clearly embodies that long-term perspectiveEspecially in this highly competitive moment, how do you maintain this long-term perspective? Because it feels like an endless competition right now, struggling to maintain a 1% lead on the leaderboard, but clearly, long-term vision is crucial.

Sundar Pichai:

I have always forced myself, obviously you are in the moment. The pace of our industry is fast, you want to iterate quickly, and I really enjoy that. But being able to step back, make long-term plans, and stay very focused on a long-term goal during that period, I think that has always been crucial.

In 2016, I wanted the whole company to be AI-first. A big part of what facilitated that moment was: Google Brain in 2012, that famous "cat paper" made breakthroughs in image classification; in 2014, we brought in Google DeepMind; January 2016 was the moment of AlphaGo. Then people would notice—many people indeed noticed in May 2016—we announced our first TPU.

Yes, so in 2016, seeing all of this, I was very clear that we were going to experience another platform shift. That was the bet on the full-stack model, positioning Google as an AI-first company. Since then, we have made a lot of progress. There have been so many breakthroughs from Google, including Transformer. We applied it to our products, such as Bert and MUM, making search better, launching Google Photos, and so on.

But obviously, with the arrival of the generative AI moment, I realized that this opportunity window is even larger. People are ready to adopt this technology on a large scale, whether they are consumers, developers, and so on. So how do we respond to such a moment? For us, AI means we launched the Gemini project, spanning Google Brain and Google DeepMind. As part of that, we decided to merge the teams into Google DeepMind, and we really increased our investment in infrastructure, data centers, TPU, GPU, and so on.

Next, you know you need to get the company into a faster pace, right? Now that you have the technology, once the GDM team starts releasing Gemini, you can talk about the series of Gemini milestones we have experienced, and it's great that you have played a role in many aspects of driving this journey. Now, how do you ensure it is reflected in all our products? There are many products that reach billions of users, right? How do you make search iterate on the capabilities that these models can achieve? That is our journey. But you know, you can step back and understand this framework. It is very exciting because for the first time, when you adopt a full-stack approach, every layer of innovation permeates the entire system, all the way to the top layer.Logan Kilpatrick:

This is how I explain pre-training to people. DeepMind's pre-training has shown significant results on Gemma. My model, like post-training, sees reinforcement learning as an accelerator for underlying capabilities. I feel the same way about our infrastructure.

Sundar Pichai:

Absolutely right. You improve the infrastructure, optimize the model's performance before training, after training, during testing, and in terms of computation. Where do we make the model better in various aspects? Or how do you acquire these capabilities and reflect them in the products, right? How does Nano Banana appear in your products? Generative UI with AI modes in search, right? So you've made improvements at every level, not to mention being able to provide these improvements to developers so they can innovate on top of that, right? That's where the multiplier effect comes from. And all of this is always incredibly exciting. It creates this multiplicative effect. Watching all of this is always incredibly thrilling.

But, you know, we've always had a long-term plan, thinking about how to achieve our goals. Some aspects take time because we take a full-stack approach. When we needed to tackle the challenges brought by the AI era, I didn't know; our capabilities at that time were indeed insufficient. So, we needed to invest to scale up, ensuring that all aspects reach a certain scale to guarantee a certain fixed cost. Therefore, if you look from an external perspective, you might feel that we are progressing slowly or lagging behind the times. But in reality, we've been building all the necessary frameworks and then advancing execution based on that. Now we have succeeded, and you can see that various teams are moving forward rapidly.

Logan Kilpatrick:

Yes, it's incredible to see all of this. You mentioned Gemini appearing in all our products. I think I was talking to Josh and Tulsi about this anchor point, about this challenge—I feel that some of the challenges faced by some of the releases now are about synchronized releases, and it may not even be from a product perspective, but from a capacity perspective, and how to ensure that the model performs well across all different product experiences. I feel this introduces a new... we almost... I've commented on similar things, we've nailed the model itself, and clearly, there's more work to do, but deploying them across all of Google's product interfaces is extremely challenging, which brings this to mind. I had a realization at I/O this year, and I want to confirm it with you because maybe you have a different perspective, but historically, aside from your Gmail or your Google account, there seems to be no thread that connects the entire suite of products owned by Google, from Cloud to Waymo, to Search, and all other products like Gmail. And now, it feels like Gemini is that thread, genuinely connecting every one of our products, and it feels like something magical is happening. I don't know how you react to that.**

Sundar Pichai:

I think it's Gemini. I know your perspective is great. But for me, the significance of Gemini goes far beyond that. It clearly embodies the essence of the "AI-first" strategy. Indeed, because now we have tangible and easily understandable products like Gemini. And you're right, Gemini can indeed enhance services across various aspects, from search, YouTube, cloud services to Vemo.

Regarding the launch of Gemini 3, one thing I love is that you mentioned synchronized releases. We have synchronized switches in many products. But for me, seeing it on X, possibly Copilot or Replied or Figma, you know, everyone coming together, they are also synchronously releasing. Yes, that's right. For me, that's scalable innovation, right? Not just us, but other companies in the world. It's amazing to see all of this.

Logan Kilpatrick:

Yeah, it's fantastic. Look at other posts, and the Nano Banana Pro instance, I’m sure you spent quite some time studying this model. People are going crazy over it, it's amazing. I have to continue...

Sundar Pichai:

I can't help but ask, are we enhancing the world's productivity, or merely satisfying entertainment needs? Is this a net progress or not? Those infographics look incredible. I think when we move beyond the entertainment phase, I just saw Ajrim post his core weaving analysis infographic on X. So, you know, this prompts me to study it closely, trying to understand it. One issue that PowerPoint brought years ago was that people kept making more and more slides, and I used to collect a lot of information, which kept growing. Perhaps with Nano Banana Pro, we are back to a tool that can compress information and present it to the world in a more understandable way.

Logan Kilpatrick:

Yes, that's exactly what I was going to say next. Historically, I have personally been skeptical about how useful many generative media models are to the world. Clearly, it's useful from an entertainment perspective, but it feels like Nano Banana Pro has crossed the chasm, especially in infographics and fact-checking in conjunction with Google Search, actually, just like I believe I can clearly see how this can become part of Google's mission—organizing the world's information to make it universally accessible—through those infographics. It shocks me, it's so interesting. I think it's a great reminder that we will see those use cases. I remember when we were creating some content, the Nano Banana team mentioned that they didn't even deliberately try to make the infographics look good; it just happened naturally as the model became very powerful, and the text rendering capabilities greatly improvedSundar Pichai:

This is very interesting. Another thing it shows me is how much potential creativity exists in the world. Another wonderful thing we are witnessing is that I believe people are going to express themselves, and we are providing the tools that enable them to realize the ideas in their minds, right? So I think otherwise, we have always been limited by the tools in front of people. You may not realize it, but we are creating more and more expressive tools, and they are becoming easier to use for more and more people. So seeing all of this, you know, is also incredibly exciting.

Logan Kilpatrick:

I have a question about this later, but one of the… I have to credit Tulsi. Tulsi suggested to me yesterday to ask you this question when I talked to you because she was curious, when you see these releases and these significant moments for Google happening, what parameters do you use to measure the success of these moments? Is it the online response? Is it how the adoption looks on the first day? Or how do you measure whether this really brings a change for Google?

Sundar Pichai:

You see, I am quite active on release day, trying to understand what is working. I am looking for feedback, for example, I am trying to understand on X how ordinary users are experiencing the product. I might respond to people saying, look, this is a valid point, we should address it. So in a sense, I assess this by observing these things. I am aware that internal teams are also using Gemini itself to gather and organize information. We have great dashboards. So I try to synthesize information from various sources. One of them is that I need to experience it firsthand, right? So I receive reports, but I am also out there trying to understand how people are using it, what they are posting, right? And I think that is important. I will go to some people or look at those large screens showing multiple dashboards, checking QPS, understanding usage, worrying about capacity issues. But all of this gives you a real sense of what people are doing and saying. That’s my way. It combines online monitoring, talking to people, walking around, and sitting down to communicate with people. I want to understand, especially on the first day, it really helps me understand what is working and what is not.

Logan Kilpatrick:

I feel like I can sense the excitement in the office now too.

Sundar Pichai:

I see some version of banana wherever I go, some…

Logan Kilpatrick:

I don’t know who did it, but Kudos to the events or facilities team, they somehow managed to bring in a hundred thousand bananas into this building and made it happen. The exciting thing is that this is just the first chapter or page of Gemini 3. We haven’t launched Flash yet. We don’t have any other 3.0 category models yet. We released Gemini 2.5 Pro. In fact, when I look at a bunch of benchmarks, even 2.5 Pro is not leading in all aspectsClearly, competitors have caught up. But even the 2.5 Pro is still the best in its class, with a lot of capabilities, and has taken a step forward with Ro on top of 2.5.

Sundar Pichai:

The 2.5 Pro was showcased at Google I/O. You can feel that this is a huge leap forward. What makes me feel good is that Demis's team and the GDM team are keeping a good pace, right? So we push forward in this area about every six months. And it’s getting harder, right? Because you are... that's right, the 2.5 Pro is a very excellent model. So to make a significant and meaningful leap from that, I think it’s difficult.

But that’s what makes the progress exciting. I know you’re always excited about Flash, which is in development and coming soon. Developers are excited about Flash because it allows you to serve more people. In the pursuit of the cutting edge, this is indeed very important. So I’m excited about 3.0 Flash. I think it will be a very excellent model. It might be our best one yet, right? We will see what our internal pre-training team thinks about the next version.

So, this culture of continuous innovation and release, I think makes this moment special, and it absolutely feels like as we enter 2026, with our comprehensive progress across all layers of the tech stack, there will be many exciting advancements.

Logan Kilpatrick:

Do you have any strange or interesting release day rituals? Or is it just about getting through the day?

Sundar Pichai:

Well, usually it’s like this, my morning routine is a bit sad in that I pick up my phone to see what’s happening in the world. In fact, I don’t even check Google’s emails because my thought is that if something interesting happens regarding Google, it will always be in the news. So I try to step back and absorb the news. That’s my approach. So largely, the ritual becomes using our products. When we appear in the news, I try to understand the questions you mentioned earlier about how it works. So that’s my main habit. On release day, I try to keep the schedule less structured so I can take time to walk around and see the teams dedicated to the product, check in with them, and see how they feel about what they’ve released. This interaction is very important to me. So yes.

Logan Kilpatrick:

I have another question in this regard, it’s about... I have an interesting observation. I think Dennis and others might have talked about this internally, but there’s a mini kitchen in the grading canopy office where a lot of DeepMind activities happen. Every time I’m there, it makes me feel... obviously Google is huge, global, and all these things are happening. But that blue MK makes Google feel small and friendlyI am very curious if there is anything interesting in this, like how you... it feels small and intimate, I am very curious how you...

Sundar Pichai:

Oh, this reminds me of early Google. Obviously, you know, I used to go there often. Probably, you know, Sergey was there, along with people like Min, Jeff, and Sanjay, the Parr program still, they were all making their own espresso. How can you better appreciate the culture than watching people make espresso with precision there? I would never dare to make espresso there. I know a lot about how to make good coffee, but I feel a bit intimidated among that group of people. But, you know, just last week, Demis and Oriol were walking around there, you know, it's talent dense, and people are constantly communicating. Visitors come. The exchange of ideas is very active. So I like that, you know, it reminds me of what the company looked like in its early days. Some of our service teams, like Emma and others, were also there. You know, that's where I mentioned I might go to see QPS, you know, I was wandering in front of these people's screens trying to understand what was happening. So that's definitely part of what I like about how the company operates.

Logan Kilpatrick:

Yes, my Google function request is that we need to somehow recreate something like MK in all PAs. I don't know how to achieve that, but...

Sundar Pichai:

You know, there are other teams that have similar versions of this. I think it really helps to bring people back to the office because when you're there, you realize the value of idea exchange. You can still go back to your place of work and have focused time, but you know, that moment is really helpful. I think, yes.

Logan Kilpatrick:

So far, a lot of the stories you've talked about regarding AI seem like we are making these very long-term investments and laying the foundation for the company's success, like ten years ago. I am curious how you view this now, or obviously... these bets are correct, like Cloud is doing well, Waymo is doing well, and quantum computing hopes to do well too. A bunch of other... quantum computing things have just been announced that are beyond my understanding, but I've been trying to understand through the "atmospheric coding" experience...

Sundar Pichai:

This is one of the ways to test whether Gemini 3 can help understand these topics at a deeper level, which is very appealing.

Logan Kilpatrick:

Yes, it can bring everything to life, and I am satisfied with what we want to express with this slogan. But how do you view the infrastructure development for the next decade? Or have we already realized that AI is key, so now all hopes are pinned on it? I am curious how you view the development prospects for the next decade and what areas we should invest in now to prepare for the next phase of successSundar Pichai:

Oh, you see, I think this has always been important, right? You know, ten years ago, we bet on artificial intelligence and invested deeply and comprehensively. We bet on creating other large new businesses to diversify the company, investing in YouTube, investing in cloud computing. Google is known as a cloud-native company that you can hardly imagine, but at that time we weren't fully offering this service externally. So, that was a deep, large-scale investment in cloud computing. You know, Waymo, these things take time. Waymo is a long-term investment. I think the turning point we see now is far from just that.

There will always be someone predicting the future, right? Quantum computing is a great prediction. I think about five years from now, we will be as excited about quantum computing as we are about artificial intelligence today. But I have been thinking about this timeframe. For example, two weeks ago we announced the "Project Sun Catcher," where we will build data centers in space. Obviously, this is like the moon landing program. It seems that some ideas are indeed crazy. But you know, when you really calm down and think about how much computing resources we will need in the future, everything starts to make sense; it's just a matter of time. So, how do you make progress? You need to work backward, set 27 milestones, and then move forward step by step. So, by 2027, we might be able to deploy some TPUs (Tensor Processing Units) in space. Oh, maybe we will even encounter a Tesla sports car flying in space. It's interesting to think about.

But this is an example of the long-term projects you want to undertake and implement, along with isomorphic projects like AlphaFold and drone delivery with Wing. We are doing exciting work in the field of robotics. So, you know, it's important to think long-term and keep making progress.

Logan Kilpatrick:

When I saw that TPUs were going to space, I contacted Thomas and said we should put Gemini on the lunar rover and let it explore the moon. It would be a great marketing campaign, even if it's not super useful scientifically. So...

Sundar Pichai:

Who knows, maybe the product has already been tested somewhere.

Logan Kilpatrick:

I believe so. This... you mentioned this clue earlier, that the continuous enhancement of capabilities is equivalent to raising the baseline threshold of creativity for everyone. I personally do feel this way. I feel like I am not inherently super creative in an artistic sense. However, I can now handle tasks that historically required creativity to complete, which empowers me. I feel like I have actually become more creative, and even the way I view the world has changed because of these tools, and I no longer worry about not being able to do something.

I think "ambient coding" is a huge example. This is a pivotal moment; this power—one of the most transformative powers in history, the ability to create software and code—is now accessible to more peopleI am very curious about you; clearly, you sometimes engage in "atmospheric coding." I'm curious about how you view that moment when AI builders (not just traditional software engineers) are able to...

Sundar Pichai:

...create things. What excites me now is that, you know, it's almost like the rise of the internet, when blogs suddenly appeared, and more people became writers, right? And then there’s YouTube, where more people became creators. Yes, you can feel this change in the programming field, and even within Google, the number of people submitting their first certifications has increased dramatically, right? And that’s precisely because of these tools; they make everything easier, right? You know, maybe you are a product marketer with an idea. How would you have described it in the past?

Now maybe you’re a bit like you’re "atmospherically coding" it out, showing it to others. So you can really see this in action. I just spoke with someone on the team who doesn’t code but has been trying. He teaches his son Spanish verb conjugations and uses the HML animation page of Gemini 3 to describe the sun. You see, when you hear stories like this, and this person is a member of our communications team, right? So you can see how everyone is starting to engage with it. So this is very promising. In my limited time, I’ve tried it too.

It’s almost like, you know, not just "atmospheric coding," but these IDEs now make coding so enjoyable, right? Of course, I’m not dealing with large codebases where you have to ensure everything is correct and security is in place. So you know, those people should provide feedback. But I do feel that things are becoming more accessible. It’s exciting again. And the amazing thing is, it’s only going to get better.

Now, whenever people talk to me about Waymo, I always like to say: remember, this is the worst time for Waymo's driving technology, right? It’s only going to get better. Similarly, for all these tools we are working on, you know, using Gemini 3 for "atmospheric coding" in AI Studio, you know, seeing it is both amazing and it’s also in its worst state. Yes, that’s right. Both are true. So, in a sense, you will see a lot of progress in the future. So I think this is definitely an exciting time, and I can’t wait to see what people around the world create with it.

Logan Kilpatrick:

Yes, that’s great. I think my last question is, what’s next? What can we expect? There are a lot of cool things in the works, but what comes to mind first?

Sundar Pichai:

I think some people need to take a break. I hope the team, all of us, can take a little break. But, you see, I’m excited about the roadmap for Gemini. I’m excited about how it integrates into all our products. We are also releasing new things, right? I love Flow. I’ve been experiencing Flow notebook alum.You know, it has a passionately growing community, seeing journalists working on it, people using it for doctoral research, really putting all the research in there, it's amazing. So there's more to come in the future.**

Logan Kilpatrick:

I'm looking forward to it. Thank you, Sundar, for the interview. And thank you all for watching. See you next time