Bloomberg reported Wednesday that Jonas Adler and Alexander Pritzel are leaving Google for Anthropic. Adler was leading Google’s AI coding effort, which is the workstream behind Gemini’s recent coding-benchmark gains. Pritzel was on the training side of Gemini, which is the workstream behind the model existing at all. Internally both were flagged as people you do not want to lose if you are trying to keep parity with Claude. They are now leaving to work on Claude.
This is the fourth and fifth senior departure from Google in six days. On June 20, John Jumper, the DeepMind director who shared the 2024 Nobel in Chemistry for AlphaFold, announced he was also moving to Anthropic. The week before, Noam Shazeer, a Gemini lead and one of the original Transformer paper authors, left Google to join OpenAI. Shazeer had previously left Google to found Character.AI, then returned in the 2024 acquihire that was Google’s way of getting him back without buying his startup at the price his startup was actually worth. He has now left again. The Character.AI money is presumably fully vested.
The unsubtle through-line is the IPO calendar. Anthropic is reportedly preparing a public filing and OpenAI is targeting a September IPO at a $730 billion private-market valuation. Both companies can credibly dangle pre-IPO equity at people who already have Google equity, which by 2026 has been a flat-to-modestly-positive line on most senior researchers’ net worth statements for about three years. The arbitrage is straightforward. Stay at Google for incremental RSU vesting against a mature stock, or take a meaningful equity grant at a company whose private valuation has roughly tripled in eighteen months and is about to be repriced by a public market that is currently happy to pay any number you put in front of it. For people with strong outside options, this is not a hard math problem.
The institutional question for Google is harder. DeepMind and Google Brain were reorganized into a single AI org in 2023 specifically to keep the talent that built modern deep learning inside one company. Three years later that talent is leaving in clusters, and the clusters are concentrated in the parts of the org that produce Gemini and the parts of DeepMind that produce the actual research. Sundar Pichai’s standing answer in earnings calls is that Google has the deepest research bench in the industry. That is still true. The question that matters is whether the second-deepest bench is now the one shipping the frontier model, and whether the gap closes or widens once Anthropic’s Korea-fueled enterprise pipeline starts producing revenue that justifies the kind of compensation packages Google’s compensation committee was built to avoid approving.