[01]
Anthropic hires OpenAI co-founder Andrej Karpathy, former Tesla AI lead
OpenAI co-founder Andrej Karpathy joined Anthropic to lead a new pre-training research team using Claude itself to accelerate training.
// daily-digest · wed 20 may 2026 · morning refresh · last 72h focus
Karpathy joined Anthropic's pre-training team and Anthropic closed the Stainless acquisition on the same day; Google's I/O developer keynote introduced WebMCP and Antigravity 2.0; Anthropic's annualized revenue hit $30 billion and topped the CNBC Disruptor 50; Ben Thompson mapped the economics of data center opposition; Sean Goedecke argued the "just-say-no engineer" was a ZIRP artifact; Fortune counted 1,200 U.S. AI bills with no common testing standard; and Penn physicists replaced electrons with hybrid light-matter particles to cut AI computing energy.
— refreshed for you, in 10 items.
OpenAI co-founder Andrej Karpathy joined Anthropic to lead a new pre-training research team using Claude itself to accelerate training.
Anthropic acquired Stainless, the SDK-generation startup used by OpenAI and Google, for over $300 million.
Anthropic topped the CNBC Disruptor 50, with annualized revenue surpassing $30 billion and a valuation now above OpenAI's.
Google unveiled Gemini Spark, a 24/7 personal AI agent that can email itself tasks and act autonomously on long-horizon goals.
Google announced WebMCP, Antigravity 2.0, Gemini 3.5 Flash, and AI-powered Android Studio code migration at the I/O developer keynote.
Thompson examines community opposition to data centers and argues the only durable fix is direct financial compensation to affected communities.
Goedecke argues the senior engineer archetype whose role is to block features was a product of the zero-interest-rate era, not a permanent fixture.
Google redesigned the Gemini app and cut AI Ultra to $100 per month, directly targeting ChatGPT and Claude subscription tiers.
Fortune surveys 1,200 state and federal AI-related bills and finds no unified standard for evaluating model safety or capability claims.
Penn physicists created hybrid light-matter particles called exciton-polaritons that can perform AI computations at a fraction of the energy electrons require.