Google just lost two unusually important AI people in two days.
On June 18, Noam Shazeer announced he was leaving Google for OpenAI. On June 19, John Jumper announced he was leaving Google DeepMind and, after a break, joining Anthropic.
Any one of these moves could be explained away as normal frontier-AI talent warfare. OpenAI and Anthropic are hiring aggressively. Google is huge. People leave.
But back-to-back, these departures feel different.
Shazeer is not just another senior researcher. He is one of the authors of “Attention Is All You Need,” a Gemini co-lead, and one of the central figures Google paid so much to bring back from Character.AI in 2024.
Jumper is also not just another scientist leaving a lab. He was one of the key people behind AlphaFold, the project that made DeepMind’s “AI can do real science” argument feel concrete. He shared the 2024 Nobel Prize in Chemistry with Demis Hassabis and David Baker. AlphaFold is not a nice demo. It is one of the strongest public examples of AI producing scientific value outside the normal chatbot story.
So if this were only about talent, it would already be notable.
But my bigger concern is not the headcount. My bigger concern is that Google’s AI product momentum also seems to be slowing down.
Google I/O had plenty of announcements: Gemini 3.5 Flash, Gemini Omni, Antigravity 2.0, and a long list of agent and developer-tool updates. On paper, that should have been a strong moment for Google.
In practice, I did not feel much of a shift.
That may sound subjective, but for products like these, subjective momentum matters. Developers can feel when a tool is entering their daily workflow. They can also feel when a product is mostly producing changelog entries.
Antigravity 2.0 is a good example. I have written before that I found the launch hard to get excited about, and my view has not changed much. Since then, there have been updates: a redesigned quota screen, PDF support, a new /btw slash command, and bug fixes. Those are useful improvements. They are not nothing.
But they do not feel like the kind of weekly upgrade that changes how I work.
That contrast matters because the competing developer tools are moving in a way that feels much more alive. With Claude Code and Codex, I often get the feeling that “something changed this week, and I need to try it because it might affect my workflow.” Sometimes it is a new agent pattern. Sometimes it is better observability. Sometimes it is a small capability that suddenly makes a tedious process easier.
With Google, I keep waiting for that feeling.
Logan Kilpatrick has been one of the most important public voices around Gemini and Google AI Studio. When the product cycle is hot, people like him become part of the distribution engine. They explain what changed, show examples, and give developers a reason to reopen the product.
But recently, even that external energy has felt quieter. I do not think this is a personal issue. It is more likely a product issue. If there is no major thing to show, even a strong advocate has less to amplify.
The missing piece is still Gemini 3.5 Pro. People expected it. People discussed it. Then attention moved elsewhere.
Right now, a lot of the AI conversation around developers is not centered on Gemini. It is centered on the next OpenAI cycle, Anthropic’s Fable 5 drama, and even strong Chinese models like GLM-5.2. Whether or not those products are all better is a separate question. The important thing is that they are occupying mindshare.
And in AI, mindshare has a way of turning into default behavior.
This is why the two departures worry me more than they otherwise would. A company can absorb talent losses if the product engine is clearly accelerating. A company can survive a quiet product month if the talent story is strong. But when both signals weaken at the same time, the narrative starts to shift.
The old Google argument was simple: yes, Google may look slow from the outside, but it has the deepest bench, the best research culture, the most infrastructure, Android, Chrome, Search, Workspace, Cloud, DeepMind, TPUs, and decades of distribution.
That argument still has force. Google has not suddenly become irrelevant. It would be absurd to say that.
But the market is no longer waiting patiently for Google to connect all of those pieces. Users are choosing tools now. Developers are building habits now. Teams are deciding which agents to trust now. Once a product becomes part of someone’s daily workflow, switching back is not automatic.
This is the part Google should worry about.
The AI race is not only about who has the best model on a benchmark. It is also about who can make the model feel reliable, available, and worth opening every day. It is about who can create a product surface where updates feel meaningful instead of scattered. It is about who can turn research depth into user habit.
Google has the pieces. I am less sure that it currently has the rhythm.
I still want Google to catch up. That is not a polite ending. I genuinely want more frontier-model choices, not fewer. For developers and ordinary users, a market with three or four strong labs is much healthier than a market where one or two companies become the only practical options.
If Google falls too far behind in daily AI workflows, the loss is not just Google’s. It is a loss for the entire ecosystem.
So my bottom line is simple: the exits of Noam Shazeer and John Jumper are not, by themselves, proof that Google AI is in trouble. But combined with the softer product momentum around Gemini and Antigravity, they make Google look more vulnerable than it should.
Google does not need another giant keynote to fix that.
It needs a product rhythm that developers can actually feel.
Sources: Axios on Noam Shazeer leaving for OpenAI, Business Insider on John Jumper leaving Google DeepMind for Anthropic, Google’s I/O 2026 announcement roundup, Google developer keynote news, and the Google Antigravity changelog.