At Google I/O, the company made significant strides in AI product development, emphasizing efficiency and ecosystem integration. Gemini 3.5 Flash was highlighted as a cost-saving solution for businesses, with claims of substantial token spend reductions—pointing to Google’s internal focus on AI operational efficiency. A major new announcement was Gemini Spark, a 24/7 personal AI agent that integrates with Google’s tools and the upcoming MCP standard, offering parallel task execution (like scheduling, emailing, and generating reports) through voice commands. This reflects Google’s ambition to dominate the personal AI assistant space, competing directly with OpenAI’s OpenClaw. The announcement comes after Google’s early lead in personal agent innovation, predating OpenAI’s acquisition of OpenClaw’s founder. Google also showed strong product commitment through product leadership like Drop Woodward, who demonstrated real-world use cases. Internal dynamics were noted, such as Demis Hassabis closing the keynote, signaling rising influence in Google’s cloud and AI strategy. Additionally, the move of Andre Carpathi to Anthropic to build models using Claude in pre-training reflects a broader industry trend of self-improvement and AI recursion. Google’s hardware ambitions were briefly showcased through audio-enabled Google Glasses partnerships with Gentle Monster and Warby Parker. Despite limited details on Gemini for science, the event signaled Google’s strategic focus on AI-driven scientific discovery. With no major new cloud or open-source model releases from other labs mentioned, Google positioned itself as a leading force in accessible, integrated, and efficient AI products.
I'm here at Google I/O, but for sure on it, the theater in Mountain Dew, I go to I/O I/O I/O to your going pendant on my own with Sorceress, and have a few interesting conversations lined up today, some on the grid, some not, and just out of the keynote that are some whole Google crew. It took notes and a figure would go through, but that stood out to me, just as someone who has been following Google for a really long time, reporting on it, has been to I/O a bunch, and is, you know, closely trapped in. What Google is doing in it. I thought there's a lot of interesting stuff there. And also, if you have questions, unless you watch the keynote and that blog, probably not seeing everything that was announced, so it would be a question. There are, I think, in the chat here. And if not, when the video goes out after, you can spot it to the video, to the minute. And I'll see your question, you know, to try to answer the question. I'll have a full issue and source, say, breaking down everything from Google. Today, and some of the interviews and conversations I've been on it, including a question or at a press Q&A yet, which I'll get through it a second. I just wanted to go through things that stood out to me, keynote, and again, do you have any questions about I/O or things you want me to ask? The executive here, a lot of me are speaking with a lot of them. Again, I'm not, but still questions, but I'm here all day. Yeah, it's really interesting. So, they're happy to have around how big Google it. Every time I come here and they do this like state, the, you know, this under does it to begin a, mind the world, the scale that they're operating at. So they're now 13 products with over 1 billion users, each five with more than 3 billion, which is the most users of any company on earth as far as I know. Some of us is definitely bragging about catbacks this year, who's talking about bumping catbacks in AI spend, which I think is something you do when, you know, your stock price has gone up as high as Google's has. And they're kind of through that trough of disillusionment period they had, I would say about a year ago, wherever I was wondering if search was going to be totally destroyed by chat GP tier, if Gemini was going to be able to pull to the frontier against open air and then they won the antitrust lawsuit, like to see everything kind of reflected in their stock price. So there was a bit of like bravado to the end. Notice a lot of TPU talk, I think an unusual amount of TPU talk out of star at the beginning of the keynote for I was a very consumer developer conference. So they've talked about TPUs in the past, but I think to me the something posted that is, you know, there's a big focus on TPU, expanding TPU outside of even dual zone properties which you're seeing within traffic and others. And really, you know, it's, I'm reminded of the old Bezos quote, you know, your margin is my opportunity. And I think Google sees Nvidia's margin and sees that as an opportunity that they remain Google remains compute constrained with TPUs, you know, people internally are still fighting like hell over TPU allocation. And they're, they're still going to sell them externally because they think it's a smart business for you. And it probably is, but just interesting to see it place to prominent the TPU piece to promising at IO this year. And yeah, everyone is joining. Oh, hey Jordan. Does it feel like this is an endless game of frog is from Jordan where whichever company it is announced most recently has the most advanced stuff or as Google or anyone doing something truly innovative. Good question. I mean, I had a scoop in the newsletter last week that Gem night 3.5, which they announced today was, was going to put them back in kind of benchmark area of GPT 5.5 and just 4.7 that it wasn't going to necessarily push them ahead of the frontier. And that that is the K, you know, sources were good as always. Yeah, it is this constant aim of like the rat race, I think, with AI. Although today, and that's again something you do when we're not pushing the frontiers, Google didn't really talk much about benchmarks at all. And they talked about the age use, they talked about, you know, improvements in coding. But I think where they see their edge is just kind of connecting everything together. And I was going to get to next the thing they have is that they are playing to win on the consumer front with Gem night. It's very clear. You see it in the product direction and Gem night is at 900 million monthly active users, which is putting it in the same kind of par as as chat GPT. So it's really a two horse race for a consumer AI right now between chat GPT and Gem night. And you're seeing a lot of consumer out to property velocity on the Gem night stuff. Drop Woodward, who is the head of product for for Gem night and the Gem night app is really a rising sparring site, Google. He had a lot of time on stage today going through different Gem night use cases. And then else, I would say the biggest headline besides Zoom model is they announced Gem night Spark, which is basically their answer to OpenClaw, OpenClaw for Nummies, which I was waiting to see which big tech company would do this first and give them the cadence of how conferences are. I guess it made sense that Google would be first. And it's interesting. No, it remains to be seen how powerful it is. You know, people need to get access to it. But what's impressive is they haven't looked up to Google Cloud. So it's a 24/7, you know, personal GCP instance to power one of these things. So I'm so great at mid joke about, you know, you don't have to open your laptops. So solving a lot of the complexity in OpenClaw and Hermes and all these other, you know, agents that people have skunk working together have done. And it's kind of interesting that Google got to that, you know, OpenClaw for Normandy's thing before OpenAI, which acquired the founder of OpenClaw, the claw father, Peter Steinberg. So I assume OpenAI will have something on this front very soon. But yeah, the demo today with with Gemini Spark was, you know, this personal agent taking action, 24/7 integrated integration with Google tools, but also I think really importantly integrated in via MCP that's coming. It's anything that can connect an MCP. So I think Google will verify and fully MCP can first. That's important because it signals that they know Gemini is and the thing that they can just date fully to Google's properties. They want this to really be this universal assistant. And they were stressing that it's, you know, it works on iOS, it works on it, Android, they've got a Mac app for Gemini, which actually looks pretty good. I'm excited to try it. And obviously all the Android integrations that go much deeper. But I think Google is really not only, yeah, playing to win on the assistant side on the Cheshire PT side, but on this OpenClaw kind of personal agent side. And, you know, the demo Josh word showed was, you know, firing off multiple tasks in parallel via voice, having it change a calendar or send an email to someone, create a digest and a deep research type report all doing that in parallel. And I think they're gating spark to $100 a month ultra plan, or I may not be calling it ultra. Google remains the most confusing company in the world of how it names things. And I don't think that's going to change anytime soon. But yeah, I would say Gemini is definitely going to stand out in terms of they're just going full tilt on the product side. So I know the questions here. Eli, what's up? Was the model day announced specialized to any use, thinking how mythos has a specific strength that made it noteworthy? Okay, good segue into my conversation also with Sundari yesterday, which I'll have in the newsletter. Yes, right. Not really a conversation. It was a remake that asked the question. But, um, yeah, they're really focusing on tool use. So like the gym to use the ability, which is what all the labs are doing, you know, the ability to like control a computer. There's a demo where the new, the Gemini Mac app was like controlling the Mac, which like Apple doesn't have anything like that. And it's kind of interesting to Google staring out of a Mac in it at one of their conferences. They have a cyber thing that they talked about, but they're still, they're still behind mythos, I think, on general capability in terms of what they've released. Now, it was interesting. I asked Sundari about this during this press Q and I yesterday and I haven't published his answer to me yet. I was saving it just for you guys on the livestream here. I asked him how mythos, if at all, as reset Google's expectation of what it means to be the frontier. And he basically said that he thinks Google has the capability of mythos internally as well, which I thought was interesting, which to me suggests, I mean, and this is evident in some way that anthropic is gating access to mythos, which, you know, those say the security concern, which I do think that's true, but I also think they're compute constrained. And you can see it in the cost of mythos, it's incredibly expensive to rock. And I think maybe, you know, everything that Google was doing with its models today was about, you know, efficiency. It was the flash version. It was like bragging about how much token spend savings companies could get if they switched to Gemini Flash. And yeah, I don't know. It was interesting to see Sundar's response to that because I don't feel like they had anything at IO today that actually they were messaging with St. Paul Park as a myth.
us. Now, I understand that Google has this kind of data problem internally with coding, which all the labs are struggling with. Other side, it seems like an anthropic, but there's this data flywheel thing that's under I was saying in response to my question that they're working on where they're trying to combine anti-gravity, which is Google's, you know, kind of cloud code, cloud code competitor, codex competitor. It is obviously like a third horse in that rate. If he had a pork fest, I think you got a cursor ahead of him probably, too. But they're trying to, they're basically finding it using that internally with their own engineers is helping them hopefully catch up on the coding data fly. I still think it's an issue. I still think they're not anywhere near up in AI and try back on code. At least it doesn't seem like I think they would have been beating their chest more on the keynote today. But that said, I mean, 3.5. I was interested to see Gemini 3.5, that it was, oh, and by the way, if everyone's new here, people knew, ask a question and an audience heard, I'm just asking people's questions and ranting about what I saw today at Google I/O. I just got to, you know, Gemini 3.5 Flash. There was a moment where Sundar was talking about, and it was one of these kind of sake made up, you know, marketing numbers, but the token spend that companies are standing on some of these more expensive cloth model, you know, for other frontier models, it could just thematically cut if they switched to flashy would give them the same kind of results. So Google, you're push really hard on efficiency. And you know, the competition I've been having a lot lately is that a lot of people inside companies that are spending a ton of the tokens know that a huge amount of the spend is incredibly wasteful, right? There's token maxing. There's literally contest just to build things and have engineers mats out there. They're token budgets and people know that this is a bit of a bubble in terms of token spend at least at these prices. I think Google's probably recognizes that they probably feel that internally if I had to guess. And they're going hard at it on the Gemini 3.5 Flash sign. So that was a standout too in terms of like differentiation with the model approaches this year. Okay, question from Kate. Very few details from Demis about Gemini for science. Do you know more? Yeah, that was kind of an interesting one. That was the closer was Demis came back out also a sign about the power dynamic inside Google right now. Maybe I'm reading too much into it, but you know, Sundar would usually come out and do the closing, you know, thank you for coming. They closed with Demis today. And I think that just shows his cloud continues to rise inside Google. Not that it needed to rise anymore. But yeah, he came out to close with this kind of this Gemini for science tease and had some kind of, you know, big quote about how we're at the footsteps of the singularity or whatever. I'm not really sure to be honest. Yeah, there were very few details. I'll definitely be asking the Google executives. I'm seeing today about that. I can report back on that. You know, science is something that Demis has been very focused on. He's been long talking about AI's potential to accelerate scientific discoveries. He's got ice, perfect labs. You know, AI, health science, spinoffs company that Google's also funding. So I think TBD, but it just seems like that's a strong area that he wants to double down up. But I'll definitely ask more about that. Okay, another question in terms of time. When do you think the other labs release their own cloud, open AI and meta? I mean, open clogs this. You can use your CHSBTA account to access it. Is codex an open-cloth competitor? You could argue it kind of is. It's definitely easier to use than open-cloth. I don't know. I still haven't seen it. Open-cloth, I say for normies, and that's not meant to be derogatory. But like, you don't need to understand coding at all. You only need to have a concept of these frameworks that we're talking about outside code and codex. I haven't seen that from a company as large as Google until today. Right, with what? Look, Gemini Spark is. I'm sure others will fast follow. And I know meta is working on it. I bet that'll probably not be till September to connect time for them. Open AI has their conference in September, but they move fast. I would imagine they'll have more kind of open-cloth side before them. But TBD, I mean, Google came out strong today with Gemini Spark. Another one from Jordan. Yeah, everyone. Hi, I'm just taking a few questions briefly here at Google I/O. I just got out of the keynote. And throughout the kit, the little sanded I/O with Andre Carpathi announcing that he's joined today. Oh, actually, I don't see that. He said to Google, I mean, if he had, like, been a co-founder of Google, yeah, it's more of a kicking sanded open AI, for sure. You know, you can't have A/L apps not announcing things on I/O day. Okay. I remember when I was here. It was two years ago. Ilya left on open AI officially, and we were all at the press happy hour, and we all had to, you know, set up the exam and go file. So there's always going to be some kind of AI drama on the day by you. That is just like, that's a sure as the sun rises. But yeah, Andre, I'll have that in the newsletter later today too. Really interesting move to have him go to Anthropic. I'll put him on X earlier for those who didn't see it. He's going to work on the pre-training team. He's going to build a team that is using Claude itself to build their models, the pre-training. So this recursive self-improvement concept that's really taking gold and a lot of the labs. Andre, it's going to have Anthropic to work on. I know I'm not a sports guy, but I'm using a lot of, you know, Katie to the Warriors, you know, memes about that move today. Okay. I feel like I've got enough unless there's any more questions. This was an experiment. I haven't done one of these before. I'm thinking, as I said at the top for those who missed it, I go to a lot of these events in person, these big tech conferences and, you know, do the written newsletter usually after. But, you know, Sussex has this live thing. Why not try it? Why not take questions? You all help me kind of inform what I'm asking out and talking to sources about. So this is an experiment. We'll see if I do it again. I just wanted to pop on here and say hello and talk a little bit about Google. But, yeah, I'm going to, I'm going to jump back into things. Oh, one last thing. Google Glasses. They had, they showed off the audio on the glasses with Gentle Monster and Warby Parker towards the end. I'm going to be interviewing Rick Astero, Google's STP and Devices and Platform and Android and all that in like 20 minutes. So I will be asking him all about that. So stay tuned for, stay tuned for more on Google's AI hardware, Android, etc. Chrome strategy in the newsletter. That would be all. Thank you, everyone. All right. Peace. you
Podcast Summary
Key Points:
Google emphasized Gemini 3.5 and its efficiency-focused "Flash" version, highlighting significant token spend savings for businesses, signaling a strategic push toward cost-effective AI models.
Gemini Spark was unveiled as Google’s response to OpenAI’s OpenClaw, offering a 24/7 personal AI agent with deep integration into Google tools and MCP (Multi-Client Platform), marking a strong move in the personal assistant and agent space.
Google reinforced its ecosystem dominance by showcasing Gemini’s cross-platform availability (iOS, Android, Mac) and internal tool control, while also indicating a focus on connecting AI across services, positioning it as a universal assistant rather than just a chatbot.
Summary:
At Google I/O, the company made significant strides in AI product development, emphasizing efficiency and ecosystem integration. 5 Flash was highlighted as a cost-saving solution for businesses, with claims of substantial token spend reductions—pointing to Google’s internal focus on AI operational efficiency. A major new announcement was Gemini Spark, a 24/7 personal AI agent that integrates with Google’s tools and the upcoming MCP standard, offering parallel task execution (like scheduling, emailing, and generating reports) through voice commands.
This reflects Google’s ambition to dominate the personal AI assistant space, competing directly with OpenAI’s OpenClaw. The announcement comes after Google’s early lead in personal agent innovation, predating OpenAI’s acquisition of OpenClaw’s founder. Google also showed strong product commitment through product leadership like Drop Woodward, who demonstrated real-world use cases.
Internal dynamics were noted, such as Demis Hassabis closing the keynote, signaling rising influence in Google’s cloud and AI strategy. Additionally, the move of Andre Carpathi to Anthropic to build models using Claude in pre-training reflects a broader industry trend of self-improvement and AI recursion. Google’s hardware ambitions were briefly showcased through audio-enabled Google Glasses partnerships with Gentle Monster and Warby Parker.
Despite limited details on Gemini for science, the event signaled Google’s strategic focus on AI-driven scientific discovery. With no major new cloud or open-source model releases from other labs mentioned, Google positioned itself as a leading force in accessible, integrated, and efficient AI products.
FAQs
Gemini Spark is Google's personal AI agent designed to work 24/7, integrated with Google tools and MCP (Model Control Protocol). It offers automated task handling like calendar updates and email creation, similar to OpenClaw, but is built as a personal assistant with deep Android and iOS integration and access to Google Cloud.
Google highlights that switching to Gemini Flash can significantly reduce token spending for companies using expensive frontier models. This emphasizes efficiency and cost savings, addressing concerns about wasteful token usage in AI development.
Google has positioned Gemini as a major consumer AI competitor with 900 million monthly active users, matching ChatGPT. Its focus on tool use, cross-platform integration (iOS, Android, Mac), and real-time task execution distinguishes it in the market.
TPUs are receiving increased attention as Google expands their use beyond internal data centers, signaling a strategic move to maintain compute leadership. This reflects Google's effort to counter Nvidia's market dominance by offering cost-effective and efficient AI compute power.
It shows Google's push to build a deeply integrated, personal AI assistant that works across devices and Google services. The product signals a shift toward universal AI agents, emphasizing connectivity and automation within the Google ecosystem.
Google believes it has internal capabilities comparable to Anthropic's Mytho, though Anthropic gates access due to security and compute constraints. Google emphasizes efficiency and integration over raw frontier capability, focusing on practical, scalable AI solutions.
Chat with AI
Loading...
Pro features
Go deeper with this episode
Unlock creator-grade tools that turn any transcript into show notes and subtitle files.