🧠 OpenAI Co-Founder Joins Anthropic

CTOs of billion-dollar companies keep taking individual contributor jobs at Anthropic. Meta is offering two million dollars a year and still losing people. And this week, Andrej Karpathy joined too. Something is pulling the smartest people in AI toward the same building, and what he's been hired to work on tells you exactly what it is. Let's dive in!

In today’s insights:

LATEST DEVELOPMENTS

Evolving AI: Karpathy joins Anthropic to help Claude speed up the research behind future Claude models.

Karpathy announced the move on X, saying the next few years at the LLM frontier will be "especially formative." He'll join Anthropic's pre-training group, which runs the massive training cycles that shape Claude's core capabilities. His new team is built around a clear bet: using Claude to accelerate the research that builds the next generation of Claude. It caps a winding path. Karpathy was a founding member of OpenAI in 2015, led Tesla Autopilot from 2017 to 2022, briefly returned to OpenAI, then launched Eureka Labs in 2024. He's now back inside a frontier lab.

This is what AI development folding in on itself actually looks like. Anthropic's CFO recently said roughly 90% of code at the company is already written by Claude, and putting Karpathy on pre-training pushes that loop deeper into the research stack. The race used to be about who could buy the most compute. It's quietly becoming about who can turn their own model into the fastest collaborator on the next one.

Evolving AI: 4 Key Insights for Scaling LLM Applications.

LLM workflows can be complex, opaque, and difficult to secure. Get the latest ebook from Datadog for practical strategies to monitor, troubleshoot, and protect your LLM applications in production. You’ll get key insights into how to overcome the challenges of deploying LLMs securely and at scale, from debugging multi-step workflows to detecting prompt injection attacks.

Free Download →

Evolving AI: Google used I/O 26 to turn Gemini into an agent that lives across its products.

Google is framing Spark as a shift from assistant to active partner, a general-purpose agent that reasons across connected apps and acts on your behalf, rolling out first to AI Ultra subscribers with MCP support expanding the connections over the summer. Sitting alongside it, Gemini Omni accepts image, audio, video, and text inputs and outputs editable video grounded in real-world knowledge across the Gemini app, Flow, and YouTube Shorts. Powering the stack is Gemini 3.5 Flash, which Google says beats 3.1 Pro on coding and agentic benchmarks at 4x output speed, and Search now runs on it too with an intelligent query box and info agents that monitor topics in the background.

The I/O 26 announcements show Gemini hardening into an agent layer that spans Google's ecosystem rather than living inside a single app. Search reasons and acts, Workspace executes, Spark orchestrates, and 3.5 Flash makes the economics work at scale. The standalone chatbot era looks like a brief stop on the way to AI woven through the products people already open every day.

Most marketing teams spend Monday morning pulling numbers. Viktor spends it posting them. Cross-platform brief in #growth before the first standup. Spend anomalies flagged before they compound.

Your marketing team stops reporting and starts deciding.

Get Started for Free

Evolving AI: Google revealed its first Gemini-powered audio glasses, launching this fall with Samsung, Warby Parker, and Gentle Monster.

Google announced its intelligent eyewear lineup at I/O 2026, with audio glasses shipping first this fall. Users tap the frame or say "Hey Google" to activate Gemini, which manages directions, calls, texts, photo capture with Nano Banana edits, and real-time translation. Agentic features let it run multi-step tasks across apps like Uber and Doordash in the background. The glasses pair with both Android and iOS, and a second tier with built-in lens displays is set to follow.

Android XR is the OS layer Google built with Samsung and Qualcomm to span the full range of extended-reality hardware, from cheap audio frames up to mixed-reality headsets. So this isn't really one product, it's the consumer entry point into a whole stack. With Meta dominating early and Apple still circling, the next interface battle isn't about a single device, it's about whose assistant gets to live everywhere.

QUICK HITS

🤝 OpenAI partners with Dell to bring Codex on-premises.

✝️ Vatican picks Anthropic's Olah to help launch landmark AI encyclical.

🔍 OpenAI launches public tool to verify AI-generated content.

💾 Intel, Qualcomm explore buying Tenstorrent to chase Nvidia.

 *partner link