Google's Mind Is Now Less Deep

Google just had one of the strangest days in its recent history, and the strangest part isn't any single departure. It's the clustering. Jeff Dean, Google employee number 30 and one of only two people ever named Senior Fellow, announced he's leaving after nearly 27 years. His fellow Senior Fellow, Sanjay Ghemawat, who's been building Google's technical backbone alongside Dean for decades, is walking out the same door. Both are reportedly heading to a startup focused on recursive self-improvement, which is quickly becoming the industry's next obsession now that everyone's bored of just saying AGI.

They're not walking alone. Oriol Vinyals, DeepMind's VP of research and a technical lead on Gemini, and Quoc Le, who co-founded Google Brain, are also gone. Both co-authored a landmark 2014 paper on scaling data for pretraining models, a paper that also happened to be co-written by a guy named Ilya Sutskever. Google's official goodbye note mentioned Dean and Ghemawat by name and conveniently skipped Vinyals and Le, which either says something about internal politics or is just an oversight nobody has explained yet.

Layered on top of all that, Demis Hassabis is stepping back from running DeepMind day-to-day to become Alphabet's Chief Scientist, a title that sounds important and means almost nothing operationally. He'll keep running Isomorphic Labs. DeepMind's actual day-to-day work has apparently already shifted to Koray Kavukcuoglu, who was already Google's Chief AI Architect and already reported straight to Sundar Pichai. Notably, nobody is getting the title of DeepMind CEO. That's not an accident; it reads like Google folding DeepMind more tightly into the main AI operation instead of letting it run as its own semi-independent kingdom.

Zoom out and this isn't an isolated week. Noam Shazeer left for OpenAI a couple months back. John Jumper, a Nobel laureate who worked alongside Hassabis, jumped to Anthropic. Peter Norvig, Google's head of research for a quarter-century, is now off working on superintelligence somewhere else entirely. Add today's names to that list and you get something that looks less like turnover and more like an exodus, the kind of talent bleed companies usually try very hard to hide, not announce all at once.

And maybe that's the point. Bundling the bad news together, rather than letting it trickle out over weeks, at least gives Google a shot at controlling the narrative instead of getting picked apart headline by headline. There's a version of this story where Google trades research glamour for shipping discipline, and actually comes out ahead, especially if the real problem was too many teams building too many overlapping AI products with nobody able to force alignment. But that's a generous read on a day that, on its face, looks like several of the smartest people in the building deciding it was time to go build something else.