There's a certain irony in building the world's smartest machines and then watching your smartest humans stroll out the door. Google is currently living that irony in glorious high definition, and its exit interviews must be absolutely brutal to sit through.
DeepMind's Revolving Door Spins Again
According to TechCrunch, Google is poised to lose two more high-profile Gemini contributors — Jonas Adler and Alexander Pritzel — to Anthropic. Adler worked on Google's AI coding effort, Pritzel on the training that made Gemini tick, and together they join a departure list that increasingly reads like an AI hall of fame roll call.
The marquee names sting the most. John Jumper, the DeepMind director who shared the 2024 Nobel Prize in Chemistry for AlphaFold's protein-structure breakthroughs, is also headed to Anthropic. Meanwhile Noam Shazeer — a foundational figure in modern AI who spent much of the last 26 years at Google — has decamped for OpenAI. When your Nobel laureates start quietly updating their LinkedIn, something is up.
Equity Is the New Signing Bonus
Why the sudden stampede? Because the rivals are sprinting toward IPOs, and pre-public equity is the shiniest carrot in tech right now. As TechCrunch framed it, it's "a great time for the companies to recruit top AI talent with a promise of equity" — the kind of stock that could turn into generational wealth if the listings stick the landing.
Google's real problem is that money alone hasn't done the trick. Shazeer only returned in 2023 via a reported $2.7 billion Character.AI acquihire, and even that eye-watering homecoming couldn't hold him for long. When a nine-figure welcome-back package has a shelf life of about three years, you don't have a compensation problem — you have a gravity problem, and the gravity is pulling toward Anthropic and OpenAI.
The pattern matters beyond bruised egos in Mountain View. Frontier models are built by disturbingly small clusters of specific people, so each departure isn't just a headcount dip; it's institutional knowledge walking across the street to a competitor.
Google still builds phenomenal models. The trouble is that the people who build them keep discovering the door is unlocked — and the neighbors are very obviously hiring.
Source: TechCrunch