Friday, August 7, 2026 · 10 curated articles · Special topic: Google's top AI brains leave
Editor's Picks
The most important AI story of the week is not a new chatbot. It is a personnel map.
On August 5, Sundar Pichai told Google staff that Demis Hassabis would become Chair of Google DeepMind and Chief Scientist of Alphabet, that Koray Kavukcuoglu would take day-to-day control of DeepMind as SVP, and that Jeff Dean was leaving after 27 years. The sentence that detonated social media was the last one. Dean and Sanjay Ghemawat are founding an independent public benefit company. Alphabet will stay involved as a founding investor and Cloud partner.
WIRED later filled in the missing names. The company is called Discovery Loop. The cofounders are not two veterans taking a sabbatical. They are four of Google's hardest-to-replace system builders: Jeff Dean, Sanjay Ghemawat, Oriol Vinyals, and Quoc Le. Their stated goal is blunt. They want AI systems that can automatically close discovery loops in machine learning, science, and engineering.
That is why the exit matters more than a farewell post. Google is splitting its AI organization into an execution engine and an AGI-science strategy layer at the same moment its best infrastructure people decide the fastest path to automated discovery is outside the company. The industry is no longer arguing only about who has the best model. It is arguing about which organizational form can invent the next research process.
These ten stories follow one rupture: Google's AI talent is being reorganized inside Alphabet and rebuilt outside it.
The Exit
Jeff Dean Leaves Google After 27 Years
after an incredible 27-year run, Jeff Dean is at a moment where he wants to try something new
Jeff and Google Senior Fellow Sanjay Ghemawat are launching an independent public benefit corporation to accelerate discoveries in ML, science, and engineering
In a message to Google staff, Sundar Pichai confirmed that Jeff Dean is leaving the company after 27 years. Dean helped build Google's search infrastructure, co-created Google Brain, and later became a technical colead on Gemini and chief scientist of Google DeepMind. Pichai framed the departure as personal choice rather than conflict, and said Alphabet would keep working with Dean and Sanjay Ghemawat as a founding investor and Cloud partner. The wording is careful. The signal is not. One of the people most associated with turning Google research into running systems is no longer building those systems from the inside.
Source: Google The Keyword
Four Google AI Stars Found Discovery Loop
4 of Google’s Top AI Brains Are Leaving—and Launching Their Own AI Startup
Our mission is straightforward: we are building AI solutions that can automatically solve important problems in machine learning, science, and engineering
WIRED reported that Dean is not leaving alone. He is cofounding Discovery Loop with Sanjay Ghemawat, DeepMind research VP Oriol Vinyals, and Google Brain cofounder Quoc Le. Dean told WIRED the idea came together only a few weeks earlier. The four are longtime colleagues and friends. Discovery Loop is structured as a public benefit corporation, and its public mission is to automate scientific and engineering discovery rather than ship another consumer chat product. This is less a lifestyle startup than a bet that AI should become the researcher, not only the research assistant.
Source: WIRED
Discovery Loop Wants Closed Experiment Loops, Not Another Chatbot
By advancing the pace at which we conduct engineering and scientific discovery, we can bring the benefits of science and technology to the world much faster
Discovery Loop's own site is short and specific. The company is not pitching a general chatbot. It is pitching automated loops: systems that can propose, test, revise, and retain progress across machine learning, science, and engineering problems. WIRED says the founders even plan to make Discovery Loop its own first customer, using the system to accelerate their own experimental work before selling those loops outward. That framing matters because it puts process, not parameter count, at the center of the next AI race.
Source: Discovery Loop
The Reorganization Inside Google
Demis Hassabis Becomes Chair of DeepMind and Alphabet Chief Scientist
Demis will become the Chair of GDM and Chief Scientist of Alphabet, while continuing to lead Isomorphic Labs
I’ve decided that now is the right time for me to hand over my day-to-day operational responsibilities at GDM
In the same staff note, Pichai said Demis Hassabis would step back from day-to-day DeepMind operations and take a more strategic role as Chair of Google DeepMind and Chief Scientist of Alphabet. Hassabis will keep leading Isomorphic Labs and says he wants more room to shape AGI and scientific applications, including health. The company is separating the person who defines the AGI narrative from the person who ships Gemini every week.
Source: Google The Keyword
Koray Kavukcuoglu Takes Day-to-Day Control of DeepMind
Koray, the current Chief Technology Officer of GDM and our Chief AI Architect, will step up as SVP of Google DeepMind, reporting to me
He will oversee Gemini model development, Frontier AI research, and the Gemini app and developer teams
Koray Kavukcuoglu becomes Senior Vice President of Google DeepMind and reports directly to Pichai. His portfolio covers Gemini model development, frontier research, the Gemini app, and developer teams. Pichai highlighted product momentum in the same memo: the Gemini app has more than 950 million monthly users, and Gemma has more than 900 million downloads. The organizational message is clear. Google wants one executive accountable for the product engine while Hassabis works the longer AGI and science horizon.
Source: Google The Keyword
Alphabet Tries to Keep a Tether to the Leavers
We’ll continue to work with them as a founding investor and Cloud partner, and collaborate on a research framework for ML systems and related infrastructure advances
Google did not treat the exit as a clean break. Pichai said Alphabet would remain a founding investor and Cloud partner to Discovery Loop and would collaborate on ML systems and infrastructure research. According to WIRED, Pichai tried more than once to keep the group inside Google before accepting the departure. The compromise is familiar in this industry: if you cannot retain the people, try to retain optionality through capital and compute.
Source: Google The Keyword
Why They Left
Large-Lab Inertia Is the Real Villain in the Founders' Account
In a large organization there is always a lot of inertia you have to overcome to make any radical changes
We want to build something different
WIRED's interview is useful because it is not written as a morality play. Vinyals says large organizations always carry inertia, and radical change has to fight that inertia first. Dean says the idea was recent, not a years-long secret campaign. The four already worked on different problems inside Google but converged on the same conclusion: AI can automate scientific and engineering loops, and that work needs a different structure. This is less "Google is finished" than "the next research process may not fit the current lab form."
Source: WIRED
The Industry Reads the Shakeup as a Structural Split
In another big change, veteran Googler Jeff Dean — the chief scientist of Google DeepMind and Google’s 30th employee — is leaving with Google Fellow Sanjay Ghemawat to form an AI-focused public benefit corporation called Discovery Loop
The Verge's account treats the DeepMind reshuffle and Dean's exit as one package. Hassabis moves up and out of daily operations. Kavukcuoglu takes the product and model engine. Dean leaves with Ghemawat to start Discovery Loop, with Google as founding investor. Read that way, the week is not a single resignation. It is a redrawing of where Google keeps execution, where it parks AGI ambition, and where its most systems-heavy talent chooses to invent next.
Source: The Verge
The Wider Field
AMD Buys Taalas to Etch Models Closer to Silicon
AMD Acquires Taalas to Advance Compute Solutions for Rapidly Growing AI Inference Market
Early tech demos show model-specific integrated circuits churning out up to 17,000 tokens a second
While Google reorganizes its people, the hardware layer keeps moving. AMD said on August 6 that it would acquire Taalas to strengthen inference performance and efficiency, planning to integrate the technology with AMD Instinct GPUs. The Register reported early demos of model-specific circuits producing up to 17,000 tokens a second. Discovery Loop is a bet on automated research process. Taalas is a bet on making inference itself more specialized. Both are attempts to escape the limits of the current stack.
Source: AMD
Humans Still Miss One in Three Dangerous Agent Commands
The average player missed 1 in 3 threats (mean accuracy 66.3%)
7% approved every single prompt
As AI labs race to give systems more autonomy, a Scale X analysis of more than 40,000 game runs found that human approvers missed one in three threat commands under time pressure. Only 35.2% of players caught every threat, and 7% approved every prompt. The study is a game, not a production audit, but the implication is practical. If Google's best engineers are leaving to automate more discovery loops, the rest of the industry still has not solved the human-in-the-loop problem for ordinary agents. More autonomy without better controls is not free progress.
Source: Scale X
This special report is curated by WindFlash AI from public company, research, and product sources about Google's AI reorganization and the Discovery Loop exit.