![\"\"\/](\"https:\/\/www.europesays.com\/ai\/wp-content\/uploads\/2026\/08\/5f41e64a30be8ee5_1786124585_inline_1.jpg\")
<\/p>\nNine days into August 2026, Alphabet lost $190 billion in market capitalization in a single trading session. The proximate cause: four of Google’s most decorated AI researchers \u2014 including Jeff Dean, the company’s 30th employee and the architect of its AI strategy for 15 years \u2014 announced they were leaving on the same day to start a venture called Discovery Loop. Simultaneously, Demis Hassabis, the Nobel Prize-winning co-founder of DeepMind, declared he was stepping back from running Google’s AI division to become Chief Scientist at Alphabet and chair of DeepMind, handing operational control to Koray Kavukcuoglu, a 13-year veteran who will report directly to Sundar Pichai.<\/p>\n
The market read it as chaos. According to Nilay Patel, The Verge’s editor-in-chief, and co-host David Pierce, speaking on The Vergecast days after the announcement, the real story is more specific \u2014 and more revealing about where Google thinks AI is headed. The reorganization is not a retreat. It is an admission that the person who built Google’s AI organization does not believe the company’s products are what AI is for.<\/p>\n
The AGI Believer Walks Away From the Product Fight<\/p>\n
Demis Hassabis has never been subtle about where his attention lies. His public obsession \u2014 the protein-folding breakthrough that won him a Nobel Prize, his repeated claim that humanity is “in the foothills of the singularity,” a phrase Pichai has since adopted in company communications \u2014 points toward someone who finds the most interesting use of AI to be curing disease, not searching your Gmail.<\/p>\n
“If you’re Demis Hassabis, you’re like, no, the product is AGI,” Nilay Patel noted. That conviction now has structural consequences. Hassabis is leaving behind the job of allocating GPUs across search, Workspace, cloud, and Android to become what Patel characterized as essentially an advisor \u2014 an honorific that lets him pursue frontier research while someone else fights the product wars.<\/p>\n
The someone else is Koray Kavukcuoglu, previously DeepMind’s CTO, now senior vice president of Google DeepMind reporting directly to Pichai. He will not hold a standalone CEO title. DeepMind is being subsumed into Google proper: a research lab becomes a business unit, expected to feed Gemini into every product Google ships.<\/p>\n
The problem, as Patel frames it, is that an org chart always ships. And what Google’s org chart has been shipping is a mess.<\/p>\n
<\/p>\n
One Company, a Dozen Different AIs<\/p>\n
Nilay Patel’s own experience with Google’s AI products tells the story. The Gemini in his GM car is a different model with different capabilities than the Gemini on his phone \u2014 and it sometimes silently reverts to plain Google Assistant. His Google Home cannot read his email to extract an order code, even though he has connected the account. Gmail, Workspace, and home devices all run recognizably different AI.<\/p>\n
“Everyone else has decided it’s the same AI,” Patel said. “Google, for a variety of cost reasons, has made the AI on Google Home a different thing than the AI that shows up in AI overviews or Gemini, which shows up in all these other places.”<\/p>\n
Companies like Anthropic and Apple treat their AI as one consistent service \u2014 you encounter the same system everywhere. Google appears to run dozens of cost-optimized variants, each tuned to a specific surface. The result is not an ecosystem. It is a product line with no product.<\/p>\n
David Pierce offered the definitive characterization of Google’s instinctual response to this problem: “The classic Google fashion would be for them to almost work together, but never actually quite work together. That’s the true Google way \u2014 we all acknowledge that these other products exist, we just don’t really do anything about it.”<\/p>\n
If Hassabis believed the product was AGI, he had zero incentive to spend time making Gemini coherent across a car dashboard and a kitchen speaker. Now Kavukcuoglu inherits that job \u2014 and he inherits it at a moment when the quality of Google’s consumer AI is under a microscope.<\/p>\n
The Poisoned Search Economy<\/p>\n
The most vivid evidence of what happens when AI quality degrades arrived the same week, in the form of a screenshot posted by Bill Gruskin on Bluesky \u2014 verified by The Verge as a real Google AI Overview. A search for “what does buying the dip mean” returned this answer:<\/p>\n
“Buying the dip means purchasing an investment like Dean’s French onion dip or stocks after a price dip” \u2014 with supporting bullets including “lower costs,” “long-term hope,” and “snack option \u2014 purchasing dairy or bean dips at the grocery store.”<\/p>\n
The AI had somehow conflated a financial strategy with a refrigerated snack. But the failure is not just comedic. It is structural.<\/p>\n
Google’s AI search features now cite Reddit more than any other domain on the internet, according to SEMrush data from May 2026 reported by The Verge’s Mia Sato. Google pays Reddit roughly $60 million a year for a licensing deal Reddit publicly considers undervalued. Reddit’s most recent earnings \u2014 $805 million in revenue, daily active users up 18% year over year \u2014 show why Google keeps paying.<\/p>\n
The problem is that both platforms are being flooded by AI-generated content designed to exploit exactly that loop. Brands post disguised product recommendations on Reddit \u2014 fake “what’s your favorite note-taking app” threads where the fourth of five suggestions is an astroturfed plant \u2014 for the sole purpose of getting scraped and cited by Google’s and Perplexity’s AI answers. Reddit responds by spending on AI moderation and complaining the Google deal is too cheap. Google responds by surfacing Reddit more heavily. David Pierce described where this ends:<\/p>\n
“The more important your platform becomes, the more AI stuff gets pointed at it, which makes it worse for actual humans. So the humans leave, so the AI becomes more powerful. It all just kind of death spirals as soon as you add AI creation into the process.”<\/p>\n
Pierce reported his own experience from productivity forums: every app-recommendation thread now contains disguised promotions, and he no longer trusts much of anything he sees on Reddit. The platforms are measurably worse, and the mechanism making them worse is the one both companies are doubling down on.<\/p>\n
Patel raised the logical endgame: if major publishers de-index from Google over AI-overview traffic losses, and Reddit’s economics deteriorate under AI-moderation costs, Google could simply buy Reddit outright and treat it the way it treats YouTube \u2014 a first-party content island indexed only by Google. Pierce’s counter was sharp: the “forget Google owns YouTube” dynamic would shatter, the gaming problem would become unmistakable, and an antitrust regime already scrutinizing Google’s search behavior would have a field day.<\/p>\n
![\"\"\/](\"https:\/\/www.europesays.com\/ai\/wp-content\/uploads\/2026\/08\/5f41e64a30be8ee5_1786124669_inline_3.jpg\")
<\/p>\n
Can Google Win While Being Third-Best?<\/p>\n
The question hanging over every part of this story is whether Google is actually behind in AI at all \u2014 and the two hosts openly disagreed on the answer.<\/p>\n
David Pierce made the contrarian case: “Google can probably win without having the best model.”<\/p>\n
His logic is that Google has distribution no competitor can match, vertical integration, an enterprise sales force, and global availability. The company could plausibly be third in frontier benchmarks and still win the market. The cautionary note he raised was Google Plus \u2014 the last time Google decided a new technology was the future and smushed it into every product, the result cratered and damaged the company for years.<\/p>\n
The case that Google is behindThe case that Google can win anywayGemini 3.5 was expected in June 2026 and remains unreleased in August; reporting suggests it missed internal quality barsGoogle disclosed 950 million monthly Gemini users at end of July \u2014 roughly comparable to OpenAI’s claimed “billion people using our models”At the benchmark frontier, Google is widely perceived as trailing OpenAI and AnthropicGoogle has distribution, a sales force, and global infrastructure no pure AI lab can replicateConsumer-facing AI Overviews continue to produce widely screenshot failuresGoogle posted a net cash operating loss because it chose to spend on data centers \u2014 and could stop tomorrow and resume printing moneyMultiple top researchers have departed for competitors in recent monthsGoogle could be third-best and still win \u2014 “I kind of think Google can be third best and still win,” Pierce said<\/p>\n
Nilay Patel was less sanguine. His warning was blunt: “If the products aren’t good along the way, all of that starts to teeter.”<\/p>\n
The bet Google is making \u2014 spending between $195 billion and $205 billion this year alone on AI infrastructure \u2014 only works if the demand materializes and the products justify it. If the consumer experience remains fragmented and the search results keep recommending French onion dip to investors, the infrastructure bet collapses under its own weight.<\/p>\n
The Talent Exodus: Four of the Best Walk Out the Same Door<\/p>\n
The people leaving are not marginal. Jeff Dean joined Google in 1999 as its 30th employee, co-founded Google Brain in 2011, and led development of the Tensor Processing Unit that powers Google’s AI infrastructure. Sanjay Ghemawat was his programming partner across MapReduce and Bigtable, the systems that let Google index the web at planetary scale \u2014 jointly honored by the Association for Computing Machinery in 2012. Oriol Vinyals co-led the Gemini model design and was a core AlphaFold contributor. Quoc Le co-authored the Seq2Seq paper that laid groundwork for modern natural language processing.<\/p>\n
All four left on the same day to form Discovery Loop, a public benefit corporation that aims to automate the scientific method itself \u2014 propose an experiment, run it, evaluate results, repeat \u2014 using AI. Alphabet is a founding investor and cloud partner. Radical Ventures and Khosla Ventures co-led the seed round.<\/p>\n
“We wanted to build something differently than how things are built at Google right now,” Ghemawat explained, noting that Google’s infrastructure is optimized for consumer apps and advertising, not for the kind of nimble research environment they envision. Dean reportedly began planning the venture just five weeks before the announcement.<\/p>\n
Their departures extend a pattern. In June, Noam Shazeer \u2014 a co-author of the 2017 “Attention Is All You Need” paper that introduced the transformer architecture \u2014 left for OpenAI, less than two years after Google paid billions to acquire his startup Character.AI and bring him back in-house. Days later, John Jumper, who shared the 2024 Nobel Prize in Chemistry with Hassabis for AlphaFold, joined Anthropic. Senior DeepMind researchers Jonas Adler and Alexander Pritzel followed him.<\/p>\n
Key DepartureGoogle RoleDestinationJeff DeanChief Scientist, Google DeepMind; employee No. 30Discovery Loop (co-founder)Sanjay GhemawatGoogle Senior Fellow; MapReduce\/Bigtable co-creatorDiscovery Loop (co-founder)Oriol VinyalsVP Research, Gemini co-lead; AlphaFold contributorDiscovery Loop (co-founder)Quoc LeGoogle Brain co-founder; Seq2Seq inventorDiscovery Loop (co-founder)Noam ShazeerGemini co-lead; transformer paper lead authorOpenAIJohn Jumper2024 Nobel Chemistry co-laureate; AlphaFold creatorAnthropic<\/p>\n
The test for Google is now explicit: Gemini 4 will be the first flagship model the company ships without the researcher who shaped its AI strategy since 2011. Kavukcuoglu, who led early DeepMind breakthroughs like WaveNet and DQN, owns the outcome. Google Cloud executives have reportedly welcomed the appointment, viewing Kavukcuoglu as more product-focused than Hassabis \u2014 exactly the orientation the job now demands.<\/p>\n
The Bigger Picture: Every Platform Is Under Siege<\/p>\n
Google’s AI identity crisis is not happening in isolation. Across the same news week, a series of stories converged on the same theme: platforms built on human content are being squeezed by AI-generated content, and the responses look more like symptoms than cures.<\/p>\n
Disney’s everything-app delusion. Disney disclosed plans to turn Disney Plus into the single app through which users manage cruise plans, theme park rides, toy purchases, Hulu content, and ESPN sports. The company also acknowledged exploring a free ad-supported tier and launched a program giving TikTok creators official access to Marvel, Star Wars, and Pixar characters \u2014 with the resulting TikToks licensed back into the Disney Plus app. Patel read the strategy as a succession-style impulse: the temptation of a one-interface-to-the-empire approach that cannot work because “users download apps for a purpose, not an empire.” The deeper driver, both hosts argued, is that the industry has replaced engagement with time-spent as its master metric, and Disney is structurally incapable of winning a volume fight against platforms that do not pay for content at all.<\/p>\n
David Ellison’s ticking clock. Paramount CEO David Ellison is trying to acquire Warner Bros. Discovery, and a court ruling has stopped it dead: a full antitrust trial begins March 2027. The deal agreement includes a ticking fee that starts accruing if the deal has not closed by September 30, 2026 \u2014 well before trial. If Paramount closes in June 2027, the hosts estimate the fees alone will run between $1.5 and $2 billion. Ellison published a New York Times op-ed arguing a combined Paramount-Warner would account for less than 20% of television watch time \u2014 which Pierce noted are “enormous numbers to say out loud and call small.” Patel pointed out that Ellison omitted TikTok from his market-share comparison, and that the Ellison family footprint is far larger than the op-ed suggests: Oracle \u2014 run by his father Larry \u2014 owns 15% of TikTok US and bankrolled Ellison’s acquisition of Paramount. Oracle stock is down 26% year to date. If the AI trade keeps deteriorating, the money to pay ticking fees and restructure two media giants may simply not be there.<\/p>\n
The FCC dismantles its own guardrails. The Federal Communications Commission voted to eliminate the 39% ownership cap on broadcast stations, replacing it with case-by-case review. Chair Brendan Carr argued the move would strengthen local news against national programmers like Disney and Comcast. Patel dismantled the claim: the stations being bought are purchased by private equity and national broadcasters like Sinclair and Tegna, which have repeatedly forced local anchors to read administration-favorable coverage. Consolidation has been killing local news for decades, not funding it. The FCC’s own general counsel was confronted on Capitol Hill by Senator John Kennedy \u2014 as hard-right a Republican as exists \u2014 who warned: “Sometimes the FCC scares me… you’re getting into the foothills of violating the First Amendment.” Patel’s reaction: “The foothills. A lot of bad foothills going around right now.”<\/p>\n
Snapchat draws a line. Snapchat \u2014 a terrible business and a massively relevant cultural product with 971 million monthly active users \u2014 announced it will stop recommending “wholly AI generated” videos in its Spotlight feed, declaring “there’s enduring value in rewarding original perspectives, personal storytelling, and the moments that people choose to create and share themselves.” The catch: detecting AI videos requires metadata or watermarking approaches everyone agrees on, and none reliably work at scale. Google has SynthID, but it depends on cameras, editors, and platforms all supporting it. The revolt against AI slop is real. The technical mechanism to enforce it is not.<\/p>\n
BMW learns that nobody asked for in-car ads. BMW pushed a Spider-Man promotion into the infotainment systems of millions of vehicles \u2014 swipe down, a Spider-Man logo appears, tap it, the car plays an animation. Public reaction was overwhelmingly negative. The distinction, both hosts noted, is agency: an optional Easter egg is a feature; something that just appears on your dashboard is an ad. The industry’s consensus future \u2014 autonomous cars as a “shopping mall of screens” \u2014 is colliding with the obvious fact that passengers will have their phones anyway.<\/p>\n
SpaceX is a data-center company now. In its first earnings since going public, SpaceX reported spending $15.8 billion on AI infrastructure versus barely over $1 billion on space and connectivity. Leasing data-center capacity made more money than rockets. The only reliably profitable segment is Starlink \u2014 which is a telecom, not a space business. xAI’s data-center commitments leave the parent underwater. “It’s pure Elon Musk,” Pierce said. “He wanted Tesla thought of as anything but a car company.” Starlink is a telecom no matter how many rockets launch it, and competing with AT&T, T-Mobile, and Verizon means taking on businesses with extraordinarily difficult economics.<\/p>\n
The week also brought OpenAI’s response to Apple’s trade-secrets lawsuit \u2014 essentially “we didn’t do it,” with the remarkable detail that the ex-Apple engineer at the center of the case allegedly kept his Apple off-boarding documents specifically to make sure the employees he recruited followed proper procedure. Both hosts expect the fight to run for years.<\/p>\n
Google’s reorganization is the purest expression of the forces squeezing every institution in this story. The person who believed the product was AGI has left the building, and product AI is now an operating unit accountable to quarterly results. Whether that produces an actually coherent product \u2014 or merely more orderly fragmentation \u2014 is the question that will define Google’s AI era. The earliest answer arrives whenever Gemini 3.5 ships. Between now and then, every screenshot of an AI Overview recommending snack dip to an investor is a data point in the case that the coast is not clear. The midterm elections in November could shift the antitrust ground under the Paramount-Warner deal and the FCC’s deregulatory push. And if major publishers accelerate their exodus from Google’s index, the Reddit dependency deepens, the death spiral tightens, and the speculation about Google buying Reddit stops being speculation.<\/p>\n","protected":false},"excerpt":{"rendered":"Nine days into August 2026, Alphabet lost $190 billion in market capitalization in a single trading session. The…\n","protected":false},"author":2,"featured_media":133371,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[4],"tags":[6744,1483,3013,56672,33755,7718,65298,66296,2408,7543,16884,28175,26271,8796,2588,658,141,43224],"class_list":["post-133370","post","type-post","status-publish","format-standard","has-post-thumbnail","category-agi","tag-agi","tag-alphabet","tag-artificial-general-intelligence","tag-bmw","tag-david-ellison","tag-demis-hassabis","tag-discovery-loop","tag-fcc","tag-gemini","tag-google-deepmind","tag-jeff-dean","tag-koray-kavukcuoglu","tag-paramount","tag-reddit","tag-snapchat","tag-spacex","tag-sundar-pichai","tag-warner-bros-discovery"],"_links":{"self":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts\/133370","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/comments?post=133370"}],"version-history":[{"count":0,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts\/133370\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/media\/133371"}],"wp:attachment":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/media?parent=133370"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/categories?post=133370"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/tags?post=133370"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}