Good morning {{first_name| Artificial intelligence lovers}}. Google had a lot to show at I/O, but the theme was clearer than the list of releases: Gemini was transformed into the proxy engine behind every Google product.
New launches, including Omni, Gemini 3.5 Flash, Spark, Antigravity 2.0, and a search overhaul, point to the same strategy – making Gemini capable, fast, and cheap enough to live wherever users already are, with an agent doing the work.
-
Gemini Agents’ Busy Day at Google I/O
-
Anthropic Lands OpenAI co-founder Andrei Karpathy
-
Create automated business reports using AI
-
Google Smart Glasses compete with “smart glasses”
-
4 new tools for AI, community workflow, and more
Google I/O
Rundown: Google only foot A flurry of new Gemini-powered releases and features at the main I/O event, including the new Omni model, 3.5 Flash, Spark Agent, Antigravity 2.0, and new agent search upgrades.
-
Gemini omni can role Inserting text, images, audio, or video into the video output, with Google describing the model as the “Nano Banana for video.”
-
Gemini Family 3.5 He takes off With Flash, which comes close to competitors like Opus 4.7 and GPT-5.5 across a variety of standards at 4x the speed and at half the cost.
-
Gemini Spark is a new 24/7 character factor That runs on Google Cloud virtual machines to take proxy actions across Workspace, Chrome, email, and chat.
-
Google touted the search update as its biggest redesign in a generation, addition Multimedia inputs, 24/7 information agents, and a generator user interface for custom layouts.
Why it matters: The bigger theme across these upgrades: integrating more efficient, multi-media Gemini across the Google suite in a big way. Its Flash 3.5 benchmarks don’t beat the competition, but combining fast, cheap capabilities close to the border with green tools that millions of users already live with is powerful.
Together with DATADOG
Rundown: Datadog LLM’s monitoring capability gives teams end-to-end visibility into their LLM applications, from claims and tool calls to infrastructure and cost performance, so that production AI systems remain reliable, secure, and worth the investment.
-
Correct multi-step AI workflow errors before they turn into incidents
-
Quality control and monitoring of regressions across model versions
-
Secure LLM pipelines against spot injection and data risks
Anthropic
Image source: Images 2.0 / The Rundown / @Karpathy on
Rundown: OpenAI co-founder and famous AI researcher Andrei Karpathy only Announce He has joined Anthropic, and is reportedly working on creating an internal group focused on automating the AI training pipeline with Claude.
-
Karpathy helped found OpenAI in 2015, led Tesla’s Autopilot program until 2022, returned briefly to OAI, then left in 2024 to start an AI education startup.
-
Karpathy joins the pre-training team under Nick Joseph and will also lead the new internal effort to implement CLOUD to Anthropic’s training pipeline.
-
“The next few years as LLMs will be particularly formative,” Karpathy wrote on X, adding that he plans to resume his teaching work “in due course.”
Why it matters: This is a huge get for Anthropic. Karpathy is one of the most respected AI researchers in the world, and it’s notable that he was lured back to the frontier after leaving OAI (again). What’s also noteworthy is what he’s doing there, moving forward with the same self-build model we’ve seen across the industry this year.
Artificial intelligence training
Rundown: In this guide you’ll learn how to connect Codex (or Claude Code) to Google Analytics using a free tool called Composio. You’ll use simple prompts to explore your data and create a Markdown report that you can schedule to run weekly.
-
In Codex, request the following: “Use the Composio CLI to examine our Google Analytics setup. Select the available accounts and properties, and find the property that tracks [SITE_OR_APP]And let me know what dimensions and metrics you will use for this useful traffic report. Don’t generate the report yet”
-
Once reviewed, request the following: “Use Composio+GA to identify the three most important trends over the last 90 days. Return a markdown report with an executive summary, tables, source notes, and recommended actions.”
-
Have Codex visualize the report as a PDF/website, and finally convert the steps into reusable skills, automating these skills daily/weekly/monthly
Pro Tip: Have Codex audit the Composio toolkit for useful connections. Use the exploration, analysis, and visualization framework on other connectors like YouTube or Shopify.
Submitted by Fiddler
Rundown: External LLM Evaluation charges you a fee per API call and increases as you scale. If you try to control spending by undersampling, you risk incidents creeping in undetected. Fiddler Centor models perform assessments in your infrastructure with no costs per call, reducing assessment total cost of ownership as you scale.
Enter your numbers to see:
-
Annual evaluation cost via GPT, Claude and Gemini models depending on your tracking volume
-
How each resident you add increases your spending per tracking
-
How your evaluation cost curves change as agent traffic grows
Rundown: Google just introduced a humor From its Intelligent Eyewear Gemini smart glasses, to partnering with Warby Parker and Gentle Monster on the first audio AI frames that will ship this fall, the display-equipped Project Aura will follow.
-
Voice Tires pair with Android or iOS phones, and power Gemini’s audio, navigation, messaging, and photo system with Nano Banana editing and live translation.
-
The glasses will have cameras, microphones, over-the-ear speakers and the ability to access Gemini by “Hey Google” or tapping on the frame for quick help.
-
This collection is the first batch of glasses from Google since the folding of Google Glass, with Samsung using the hardware and frames to handle Gentle Monster and Warby Parker.
-
Android XR from XRealHala ProjectThe viewing glasses are further apart, which will allow users to watch videos or content for interactive experiences.
Why it matters: Google’s Meta Ray-Ban competitor is officially coming, and its first iteration is a lightweight version – with the first audio frames coming in a step behind the competitor’s already screen-equipped options. But the appeal of the Google and Gemini ecosystem could certainly be the more difficult choice for consumers to connect their AI-driven lives together.
-
📱 metabind – Create native MCP applications using an interactive user interface. One click to ChatGPT, Claude or your iOS/Android AI Assistant*
-
🤖 Gemini Spark – Your personal Google Agent running 24/7 on Cloud VMs
New reports open Google Deepmind’s Demis Hassabis was an early angel investor in Anthropic, as Google has also separately invested billions in the AI competitor.
Anthropic Rolled Sandboxes and MCP tunnels to Claude Managed Agents, allowing teams to run the tool’s implementation and access internal servers without publicly disclosing it.
meter published First Frontier Risk Report Find Agents from top laboratories can independently complete engineering work that takes several weeks, but struggle with tasks that are difficult to verify.
OpenAI He is Adoption Google’s SynthID watermark for ChatGPT images, along with a public validator that detects if an image came from their templates.
Google and Blackstone formed AI joint cloud venture to lease TPU chips from Google The PE giant’s $5 billion investment represents a major push into AI infrastructure.
In each newsletter we show how the reader is using AI to work smarter, save time or make life easier.
Today’s workflow comes from the reader unknown:
“Occasionally I see a ‘sponsored’ ad on Facebook that makes me curious to look at it. After reading pages and pages urging extreme values for the item and other pages of testimonials, the price eventually comes up if I half believe the pomp and circumstance.
I had the wild idea of feeding the link to my AI, which was preceded by the phrase “Is this a scam?” And I get a detailed yes, no, maybe with details and background, and it’s usually “avoid this and go to ‘another website/store’ and check out the same item”.
How do you use artificial intelligence? Tell us here.
That’s all for today!
Before you go, we’d love to know what you thought of today’s newsletter to help us improve The Rundown experience for you.
Rowan, Joey, Zack, Shubham and Jennifer – the humans behind The Rundown
