Episode Summary
TL;DR and Show Notes Hosts and Guests Alex Volkov - AI Evangelist at Weights & Biases / CoreWeave, @altryne Co-hosts: @WolframRvnwlf , @nisten , @ldjconfirmed Guest: Logan Kilpatrick, MTS at Google DeepMind / AI Studio, @OfficialLoganK Google I/O 2026 Google went all-in on agents across Search, Gemini, Antigravity, Workspace, Android, Cloud and YouTube ( I/O site , Alex thread ) Antigravity 2.0 became the central agentic coding harness across Google ( Sundar , Google OS demo ) Gemini 3.5 Flash launched as a fast, determined workhorse model for agentic loops ( Logan , Noam Shazeer , Jeff Dean ) Gemini 3.5 Flash is rolling out across the Gemini app, Search AI Mode, Gemini API, Google AI Studio, Antigravity and Gemini Enterprise Agent Platform ( Koray Kavukcuoglu ) Google Search is getting new Gemini 3.5 Flash-powered agentic capabilities, including a new AI-powered Search box and backgroun
In This Episode
- ⚡ Introduction & Co-host Banter
- 🏢 Google I/O Overview & Gemini User Growth
- 🏢 Antigravity & Agentic Coding at Google
- 🏢 Google Spark & Agentic Agents
- 🏢 OpenAI Codex Mobile & Agent Workflows
- 📰 TLDR - Weekly AI News Roundup
- 🏢 Google Gemini Omni - Multimodal AI
- 🎨 Jeff Dean on Omni Architecture
- 🤖 Cursor Composer 2.5
- 🏢 Anthropic & xAI / Colossus Infrastructure
- ⚡ Data Center Water Usage Debate
- ⚡ This Week's Buzz - Weights & Biases
- 🏢 Logan Kilpatrick Interview - Google AI Studio
- 🏢 Google's Product-AI Integration
- 🏢 Antigravity as Google's Agent Harness
- 🤖 Managed Agents & Interactions API
- 🤖 Flash 3.5 - Built for the Agentic Era
- 🏢 Gemini 3.5 Flash Discussion
- 🏢 OpenAI Solves Erdős Math Problem
- ⚡ Closing - Foothills of the Singularity
Hosts & Guests
By The Numbers
⚡ Introduction & Co-host Banter
Alex Volkov: Hello and welcome to ThursdAI for May 21st. I'm an AI Evangelist with Weights & Biases from CoreWeave.
- I'm super excited to tell about all about the Google I/O week, Google heavy week to, this week.
- I went to Google I/O for the third time in a row.
🏢 Google I/O Overview & Gemini User Growth
I went to Google I/O for the third time in a row to, to cover the G- Google I/O conference. shout out to the DeepMind team for the invitation.
- we also went live, from there, and I can't wait to tell you all about this.
- with me to cover the show is LDJ and Wolfram, so hopefully we'll bring them back up.
- LDJ: I'm doing great Alex Volkov: What, what is the standout kind of thing that, that you saw on, from the Google experience?
🏢 Antigravity & Agentic Coding at Google
Alex Volkov: we have to talk about Antigravity. Folks, we've been tracking Antigravity since the acquisition of Windsurf.
- The rocky acquisition, to remind you, do you guys remember there was a whole, half of the Antigravity, half the w- of the Windsurf team, went to Google with the two billion plus so
- and then some of the rest folks were got picked up by Cognition, and now Cognition Windsurf, is part of Cognition.
- Antigravity became kind of like, like an internal tool to Google.
🏢 Google Spark & Agentic Agents
Alex Volkov: the thing that we must absolutely talk about, I unfortunately don't have access to, to test yet, is Google Spark. Google Spark is a new OpenClaw competitor, I think from Google.
- LDJ, have you seen anything on the Google Spark, thing?
- have you looked into that release as well?
- Wolfram, what do you think about Google Spark?
🏢 OpenAI Codex Mobile & Agent Workflows
Alex Volkov: we switched from OpenClaw to Hermes, at least for now. I'm, I'm not saying this is the, the, the end all, but at least for now.
- Last week, Codex released Codex Mobile.
- We chatted about this on the show a little bit, but I think it came, like, a little bit of the end of the show if I, if I remember correctly.
- Alex Volkov: But Codex released Codex Mobile.
📰 TLDR - Weekly AI News Roundup
Alex Volkov: welcome to the TLDR, the section in ThursdAI where we talk about kind everything that happened major in the world of AI news this week. And we used to say everything, and recently we're saying this is a cur-curated list of news of what we deem most important because saying everything is impossible to cover.
- Google IO had a lot of launches and, n-not all of them are actually relevant to the folks at ThursdAI, but the most important ones.
- So, you-- today, your host and guest, Alex Volkov, AI Evangelist with Weights & Biases and CoreWeave, and then we also have, Wolfram Raven, also AI Evangelist at, Weights & Biases
- Yam is not with us today, but Nisten, LDJ, and we're gonna have a friend of the pod, Logan Kilpatrick, the m-member of technical staff on the Google AI Studio team, join us in arou
🏢 Google Gemini Omni - Multimodal AI
Alex Volkov: I wanna start with Omni. I think Omni was the biggest, the biggest, like crazy, crazy unlock from, from the conference.
- I wanted to show you something that I created with Omni, I think.
- So Google launched this, Google Omni.
- Google's first conversational video editor.
🎨 Jeff Dean on Omni Architecture
Jeff Dean: in the same way that Nano Banana suddenly brought a Gemini model to image generation and let you interactively edit that image with- Yeah ... you know, the full capability of a multimodal Gemini model, you could, you know, say, "Please make the sky purple instead of blue," but everything else will remain the same.
- Alex Volkov: just each other Jeff Dean: in the center.
- Jeff Dean: Then you also input images and quickly add this dog.
- And have a picture of your dog or whatever.
🤖 Cursor Composer 2.5
Alex Volkov: Yes, uh, Composer 2.5. Thank you, Nisten.folks, Cursor is back.
- Many people like d- dismissed Cursor, with the launch and whatever, but Cursor is absolutely back, and they have a great, great, great pre-training and post-training team.
- Composer 2.5 from the teams at Cursor , is a continued training on top of Kimi K 2.5, a Chinese model, with, with permission.
- And now it's training, partly on Colossus.
🏢 Anthropic & xAI / Colossus Infrastructure
Alex Volkov: Anthropic, was revealed to be paying a, a billion, and a quarter, $1.25 billion per month for AI compute to run inference on the Memphis, like, Colossus facility as well. this is a bombastic deal that allows Anthropic to, like, serve more, AI inference at scale.
- which is, which is, which is crazy.
- and now Anthropic is actually feeling, like, less restrained.
- Nisten, any comments on the, on the Anthropic deal or SpaceX AI, IPO- On the- S1?
⚡ Data Center Water Usage Debate
Alex Volkov: Can we talk about the water thing real quick? I really want to talk about the water thing, Nisten Tahiraj: yeah, Alex Volkov: because- Nisten Tahiraj: It's all hysteria around it.
- Alex Volkov: many people who don't like AI now have a new talking point, that data centers are using a lot of water, instead of giving this water to people.
- This is a very similar talking point to crypto from before.
- Data centers, crypto, GPUs as well.
⚡ This Week's Buzz - Weights & Biases
Alex Volkov: this week's Buzz Corner, where we talk about everything that happens in, in the world of Weights & Biases and CoreWeave. And, to help us guide us through this, Wolfram, I think you have a few items for us that is very interesting for us to discuss, starting with a few emails from Wolf Venture, I think Wolfram Ravenwolf: Yeah, do you want to look at the Flash events right now or just talk about the, CoreWeave news?
- Let's do, let's talk about CoreWeave news.
- out Wolfram Ravenwolf: So basically, I shared it, in nine days and 20 hours, the premier conference on robotics and automation, the ICRA, will happen in Vienna, and CoreWeave is a
- So if anyone is interested in automation or robotics, that is a great conference.
🏢 Logan Kilpatrick Interview - Google AI Studio
Alex Volkov: let's say hi to our friend, Logan Kilpatrick. Logan, MTS at Google- Yeah, I've got Kilpatrick ...
- hey, Logan, welcome back to Thursd AI.
- time in forever, I think, you guys didn't launch on a Thursday.
- You launched a little bit before, so it gave us time to prepare.
🏢 Google's Product-AI Integration
Logan Kilpatrick: I've thought a lot about, like, why is Google so well-positioned? And actually not even-- I think well-positioned is the wrong framing of this.
- People always talk about like, "Oh, it's so-- it must be so nice to be Google.
- You can make, you can make AI and, you know, bring it to people." I frame it as more like it's an obligation and, like, a requirement for us to do this.
- And I think the reason why that's the case is because, you look at the products that we actually offer, and you look at sort of where customers want those product experiences to go
🏢 Antigravity as Google's Agent Harness
Logan Kilpatrick: I mentioned like sort of Google's infrastructure had historically been this through line, and then we sort of had Gemini as this through line, and now actually the Antigravity Agent Harness is the through line, and sort of as all of those products, again, like very naturally, we don't have to like make this stuff up, it's just like the obvious direction the product should be going, become agentic. It, it's actually an accelerant for a lot of those product te
- Like we have a single agent harness, it powers those experiences internally.
- and IO was fun actually because we get to sort of bring the Agent Harness to developers as well, and so if you're sort of however you build and you wanna build, if it's in a hosted
- I had a bunch of conversations with folks at IO and, a lot of fun conversations sort of talking about this like model product harness symbiosis, and sort of that's, I think, the di
🤖 Managed Agents & Interactions API
Logan Kilpatrick: we built this next generation API called the Interactions API. The goal for the Interactions API was sort of looking around the corner and sort of, sort of how, what, how is the usage of AI systems evolving?
- Historically, you were just interacting with a model.
- You would send in tokens, you would get out tokens.
- and then all this sort of agentic tool use stuff started to happen.
🤖 Flash 3.5 - Built for the Agentic Era
Logan Kilpatrick: I'll make one quick comment about Flash. I was having this conversation with folks yesterday, and my, my colleague, Tulsi Doshi, actually made a, an interesting comment about Flash- Alex Volkov: Also somebody who appeared on the pod before.
- We need to go pester Tulsi to make sure she tweets more, 'cause she's, like, very close to the action and, and, and sort of dialed in.
- But the comment that she made which resonated with me was, like, I think a lot of the comments about Flash is sort of comparing sort of previous iterations of Flash, and the way th
- And I don't know, we probably didn't land this point clear enough.
🏢 Gemini 3.5 Flash Discussion
Alex Volkov: LDJ, we started this show with a little bit of the kind of the controversy about, like, the pricing and the Flash not being Flash. After hearing from Logan, what, what, what's your take on, where, where, like, this Gemini model lands?
- LDJ: Yeah, I mean, I feel like to really get a true picture, we ultimately have to go use it, right?
- LDJ: I think it can definitely at least be, competitive for certain tasks and for certain enterprises and for certain price ranges and task difficulty ranges, yeah.
- Nisten Tahiraj: I'm using it right now, and I definitely see the determinism in it.
🏢 OpenAI Solves Erdős Math Problem
Alex Volkov: we have to talk about the math thing. We have to talk about the Erdos math thing.
- I think it's the, the most important thing to happen outside of Google IO.
- LDJ, would you want to comment about the, the, the, the Erdos, Yes ...
- Alex Volkov: Do you want to present this first, like what happened and, like, the, the-- give us the title, and then we'll, we'll, we'll talk about this?
⚡ Closing - Foothills of the Singularity
Alex Volkov: Demis Hassabis, the founder of DeepMind, or the co-founder of DeepMind, the Nobel laureate, the, you know, the creator of AlphaGo and AlphaZero and AlphaFold, basically stood on the stage and said, "Hey, it now feels like we're standing on the foothills of the singularity." Isomorphic Labs is part of DeepMind as well. I think when a person like Demis Hassabis is talking about something like that, I think it's very, very important, and I think, he's not a doomer i
- He's the most, like, AGI-build positive person in the world, and he talks about also responsibility because many people are, are afraid of what's coming.
- And also, you cannot just, like, go entirely and be accelerationist w- w- with disregard for safety 'cause people are getting to be impacted by some of these things.
- And, to finish on a very positive note, this is why we're here.
Hosts and Guests
Alex Volkov - AI Evangelist at Weights & Biases / CoreWeave, @altryne
Co-hosts: @WolframRvnwlf, @nisten, @ldjconfirmed
Guest: Logan Kilpatrick, MTS at Google DeepMind / AI Studio, @OfficialLoganK
Google I/O 2026
Google went all-in on agents across Search, Gemini, Antigravity, Workspace, Android, Cloud and YouTube (I/O site, Alex thread)
Antigravity 2.0 became the central agentic coding harness across Google (Sundar, Google OS demo)
Gemini 3.5 Flash launched as a fast, determined workhorse model for agentic loops (Logan, Noam Shazeer, Jeff Dean)
Gemini 3.5 Flash is rolling out across the Gemini app, Search AI Mode, Gemini API, Google AI Studio, Antigravity and Gemini Enterprise Agent Platform (Koray Kavukcuoglu)
Google Search is getting new Gemini 3.5 Flash-powered agentic capabilities, including a new AI-powered Search box and background information agents (Sundar)
Gemini Spark was announced as a 24/7 personal AI agent that can proactively work across Google surfaces (News from Google)
Google teased Gemini-powered Android XR smart glasses with eyewear partners Gentle Monster and Warby Parker (Google, Alex live reaction)
Google AI Studio and the Gemini API got major agentic developer updates, including Managed Agents (Google AI Developers)
Vision & Video
Google DeepMind launched Gemini Omni, a “create anything from anything” multimodal model starting with conversational video editing (DeepMind, Google DeepMind on X)
Omni is available in the Gemini app, Google Flow and YouTube, with API support coming soon (Logan, Gemini App, Sundar)
Key distinction: Omni is not just text-to-video, it is an iterative multi-turn video editing model that combines Gemini intelligence, world knowledge, multimodal inputs and generative media (Google)
Big CO LLMs + APIs
OpenAI announced a general-purpose reasoning model made progress on the Erdős planar unit distance problem, challenging an 80-year-old mathematical belief (OpenAI, X)
Cursor launched Composer 2.5, built on Kimi K2.5, with Opus-class coding performance at much lower cost (Cursor blog, X)
Alibaba released Qwen 3.7-Max, an agentic frontier model with long autonomous runs and robotics demos (Qwen blog, X, robot demo)
Andrej Karpathy joined Anthropic to work on frontier LLM R&D (X)
SpaceX IPO filing revealed Anthropic is paying $1.25B/month for AI compute at the Memphis Colossus facility (Axios, Sawyer Merritt)
The jury in Musk v. Altman found Musk’s OpenAI claims barred by statute of limitations, with Musk saying he will appeal (Elon Musk, Sawyer Merritt, Max Zeff)
Open Source LLMs
Tools & Agentic Engineering
Google launched Managed Agents in the Gemini API, letting developers spin up hosted Antigravity agents with Linux sandboxes and persistent state (Docs, X)
xAI launched Grok Build, an agentic CLI coding tool in beta for SuperGrok Heavy users (xAI CLI, X)
Hermes and OpenClaw can now use X subscription auth for semantic search and Grok tooling (Alex)
OpenAI Codex Mobile is now available in the ChatGPT mobile apps for remote agent workflows (OpenAI)
Anthropic doubled Claude usage outside peak hours for a limited period, including Claude Code and other Claude surfaces (Claude)
This Week’s Buzz - W&B / CoreWeave
Weights & Biases by CoreWeave is at ICRA 2026 in Vienna, with robotics and automation taking center stage (ICRA, W&B event page)
NVIDIA heads to ICRA 2026 with robotics work around generalist humanoids, physical AI and sim-to-real systems (NVIDIA Robotics, NVIDIA ICRA)
Wolfram is speaking about WolfBench at the AI Developer event in Cologne before heading to ICRA in Vienna (Wolfram)
Other Topics
Data center water usage discourse came up again, including why comparisons need real scale and context rather than viral fear math
The broader theme of the week: coding agents are becoming general agents, and the major labs are now competing on the full stack of model, harness, tools, context and compute