|
Google Just Lost 4 AI Legends in a Single Day — and They’re Building an AI That Runs Its Own Experiments
August 5, 2026 · 4 min analysis
On Tuesday, Google’s longest-serving AI leaders walked out the door together. Jeff Dean (employee #30, 27 years, designed TensorFlow and TPUs), Sanjay Ghemawat (co-built MapReduce and BigTable), Oriol Vinyals (Gemini co-lead, co-invented seq2seq), and Quoc Le (Google Brain co-founder) announced Discovery Loop — a public benefit corporation that uses AI to run the scientific method itself. Same day, Demis Hassabis stepped back to chairman and Alphabet’s chief scientist. GOOG dropped ~5%.
Why this matters: This isn’t just another exodus — it’s the culmination of a pattern. June’s departures already wiped $270 billion off Alphabet’s market cap. Both Gemini co-technical leads are now gone. DeepMind engineers are 11x more likely to move to Anthropic. The talent that built Google’s AI moat is now building outside it. Discovery Loop’s explicit ambition — recursive self-improvement where AI designs and runs experiments with no human in the loop — is the same capability that makes AI agents genuinely transformative and deeply unsettling.
What to watch: Alphabet is a founding investor and providing cloud compute for year one — so Google gets a stake in whatever Discovery Loop builds. But the bigger question is whether this accelerates the unbundling of Big Tech’s AI talent. When your most decorated researchers decide the best place to build the future is a startup, something structural has shifted.
TechCrunch ·
Memeburn ·
NYT
Sapiom Raises $35M to Slash AI Agent Costs by Routing to Cheaper Models
San Francisco startup Sapiom raised $35M Series A (Anthropic is a backer) for its model router, which sends each AI agent call to the cheapest capable model. One early customer slashed its monthly Anthropic bill from $1.2M to ~$100,000. With Gartner forecasting 40%+ of agentic AI projects canceled by 2027 due to costs, Sapiom’s bet — that 95% of agent tasks don’t need a frontier model — could reshape how enterprises budget for AI agents.
Memeburn
Shopify’s AI-Referred Traffic Tripled — and It’s Not Cannibalizing Google
Shopify reported AI-driven traffic and orders tripled year-over-year in Q2, with AI-referred shoppers converting at nearly 50% higher rates and carrying 14% higher order values. CEO Harley Finkelstein says AI search is additive to Google, not a replacement — traditional search sessions are still growing. The key insight: AI agents make multiple structured calls into Shopify’s catalog, matching intent (not just keywords), which explains why 75% of AI-attributed purchases happened outside the top 100 categories.
TechCrunch
|
