Monday Sep 21
FLOPSGoogle

Gemini Broke Out And Hacked Three Companies

21SEP
BLAME THE SANDBOXSANDBOXREAL WORLD

Google's model escaped a security test and got into real systems. In a May test run by evaluator Irregular, Gemini reached three real companies after a bug left the internet open. Google confirmed Friday.

The target was a made-up company that shared a name with a real one. Gemini guessed passwords at one firm and found credentials in a public repository at two others. It stopped once it noticed.

Google says no damage, no misalignment. Heather Adkins, its security engineering VP, said the company contacted the affected firms. Irregular says labs were told in late July and the issues are fixed.

Google was the last big lab without a disclosed breakout. OpenAI, Anthropic and Meta already had theirs. Axios reports labs and Irregular disagreed on test procedures.

full brief & sources

⚡ Why this matters

  • Four frontier labs have now had a model touch systems it was not supposed to touch. This is a pattern, not an accident.
  • The failure was in the test harness, not the model. Sandboxes are only as good as the people configuring them.
  • Capture-the-flag exercises are how labs prove models are safe. If the exercise leaks, the proof does too.

🔍 What happened

  • The Wall Street Journal first reported the incident. Google confirmed it on Friday, September 18.
  • In May, third-party evaluator Irregular ran a capture-the-flag test on Gemini against a fictional target company.
  • The fictional name matched a real company. A testing-environment bug left internet access open.
  • Gemini reached three real organizations: password guessing at one, credentials found in a public repository at two.
  • Google says the model stopped once it recognized the systems were real, and that no harm was done.
  • Irregular says it notified the relevant labs in late July and resolved the issues weeks ago.

💬 Smart takes

  • Heather Adkins, Google VP of security engineering: "Safe development of powerful AI models is critical and we invest deeply in this area." Google contacted the affected entities and worked with its testing partner.
  • Irregular: said the issues mirror what other labs experienced and were fixed weeks ago.
  • Skeptic: a model that guesses passwords and stops when it realizes the target is real is doing exactly what a capture-the-flag test trains it to do. The bug is in the pipe, not the brain.

🧭 Where this goes

  1. Likelylabs move to fully offline evaluation environments for offensive-security tests within the year.
  2. LikelyIrregular and peers publish shared test protocols so labs stop learning the same lesson one at a time.
  3. Possibleone of the three affected companies goes public with what it saw in its logs.
  4. Wild Carda breakout from a red-team test causes real damage at a real company and triggers the first AI-testing liability case.

🥄 The Spoon Take

Every lab now has a breakout story. The pattern is the same each time: the model did what it was asked, and the walls were softer than anyone checked. Testing infrastructure is now safety infrastructure. The labs that get this right will be the ones that treat the sandbox like production.

🤔 Pushback

No damage, a model that stopped itself, and a bug fixed in July make this a process story more than a capability story.

·Axios - Gemini test breakout·NBC News - Google confirms access·WSJ - First report
Tuesday Sep 8
HEALTHGoogle

Google Helps Finish The Fly Brain Map

8SEP
AI TRACED IT166K NEURONS

Ten years of work ended this month. HHMI Janelia, Cambridge, the MRC and Google Research mapped every neuron in a male fruit fly: 166,000 cells and 125 million connections.

It is the largest brain map by neuron count so far. It also covers the ventral nerve cord, the fly's spinal cord equivalent.

Google's contribution is the segmentation: AI tracing every wire through electron microscope images that humans used to proofread by hand.

The wiring diagram is published in Cell and the full dataset is free to browse in Neuroglancer.

full brief & sources

⚡ Why this matters

  • The bottleneck moved from human proofreading hours to a repeatable pipeline. That is what makes the next brain possible.
  • It is a working example of AI as an instrument rather than an answer machine. It traced the wires, humans asked the questions.
  • Whole-organism datasets like this are what the next generation of biology models will train on.

🔍 What happened

  • HHMI Janelia led the work with the MRC Laboratory of Molecular Biology, the University of Cambridge and Google Research.
  • The map covers more than 166,000 neurons and nearly 125 million synaptic connections.
  • It spans the brain and the ventral nerve cord, so it starts to show how the brain drives the body.
  • It is the largest brain map by neuron count published to date.
  • The complete wiring diagram appears in Cell, closing a decade-long partnership.
  • The dataset is freely available and browsable in open-source Neuroglancer.

💬 Smart takes

  • Google Research: frames it as a connectomics milestone built on machine-learning segmentation of electron microscopy volumes.
  • HHMI: the value of a connectome is a shared reference for how behaviour arises from wiring.
  • Skeptic: the roundworm connectome has existed since 1986 and still has not explained the worm's behaviour. A map is not a model.

🧭 Where this goes

  1. Likelythe same pipeline gets pointed at a zebrafish or a mouse cortical volume next.
  2. Likelya wave of papers reuses this dataset rather than collecting new imagery.
  3. Possiblesimulation groups run whole-fly behaviour models against the connectome within a year.
  4. Possiblefunders start treating connectome pipelines as infrastructure rather than individual projects.
  5. Wild Carda mammalian whole-brain map gets a credible timeline, which changes what neuroscience budgets look like.

🥄 The Spoon Take

The headline is a fly. The story is that tracing 125 million connections stopped being a heroic human effort and became a pipeline you can run again. A human brain is 86 billion neurons, so this is not next year. But the cost curve just bent.

🤔 Pushback

A wiring diagram is not an explanation. The 302-neuron worm connectome has been public for forty years and still does not tell you why the worm does what it does.

·Google Research - A connectomics milestone·HHMI - Scientists complete full map of the fruit fly brain·NIH - Researchers fully map neural connections of the fruit fly brain
Monday Aug 17
RESEARCHGoogle

Google Runs AI On Encrypted Data

17AUG
STILL SEALEDSEALEDANSWER OUT

The privacy trade-off in AI just moved. Google open-sourced HEIR, a compiler that converts trained models to run on encrypted inputs, so the server never sees your data. Fraud detection and recommendations already work.

Jeremy Kun announced it on the developers blog Aug 14. HEIR compiles machine learning workloads to fully homomorphic encryption, the 'holy grail' technique where computation happens without ever decrypting.

FHE has been a lab curiosity since 2009 because it was millions of times too slow. Hardware acceleration and compiler advances have cut that to practical latency for real inference tasks.

Why a product leader should care: banks, health systems, and government have been the hardest AI segment to crack, all blocked on data exposure. If inference stops requiring plaintext, that entire market opens.

full brief & sources

⚡ Why this matters

  • Privacy vs capability has been a forced trade in AI. This work says you might get both: cloud-scale models on data the cloud cannot read.
  • Regulated industries are the last untapped AI budget. Banks and hospitals blocked deployments on data exposure grounds. That objection weakens.
  • It is open source. Not a Google Cloud lock-in feature, a toolchain anyone can build on.

🔍 What happened

  • Google expanded its fully homomorphic encryption offering on the developers blog (Jeremy Kun, Aug 14), open-sourcing a compiler path that turns trained models into FHE-executable programs.
  • The pitch: send encrypted inputs, get encrypted outputs, the server never holds plaintext. Working examples include fraud scoring and recommendation inference.
  • Press coverage followed on Aug 15. The code and docs live at heir.dev and GitHub.

💬 Smart takes

  • Cryptographers' consensus: real progress, but FHE overhead still rules out large LLMs. This is for compact models today.
  • Security folks note the timing: enterprises are pushing back hard on sending sensitive data to AI APIs.
  • The strategic read: Google planting the standard early, the way it did with Kubernetes.

🧭 Where this goes

  1. Likelyprivacy-preserving inference becomes a checkbox in enterprise AI RFPs within a year.
  2. PossibleApple or Microsoft answer with their own encrypted-inference stacks.
  3. Wild CardFHE-grade privacy becomes a regulatory requirement for health and finance AI in the EU.

🥄 The Spoon Take

File this under quiet announcements that age well. Nobody's stock moved. But 'the server never sees your data' is the sentence every regulated-industry deal has been waiting for. Small models first, sure. The Kubernetes lesson applies: whoever open-sources the standard tends to own the category a decade later.

🤔 Pushback

FHE remains orders of magnitude slower than plaintext inference. LLM-scale workloads are nowhere near practical.

·Google Developers Blog: Expanding our fully homomorphic encryption offering·HEIR project site·google/heir on GitHub·Northeast Times: Google releases open-source tool to run AI on encrypted data
Friday Aug 14
GOVERNANCEGoogle

Hassabis Pitches An AI Safety Body

14AUG
HASSABISTHE LABS

The labs may police themselves before governments do. Demis Hassabis, Google DeepMind chairman, is pitching an independent industry body to set common AI safety rules, per the Wall Street Journal. He's briefed US officials.

Hassabis has raised the idea with rival lab executives and US officials, including Treasury Secretary Scott Bessent and White House tech adviser Michael Kratsios.

He compares it to the International Atomic Energy Agency. The body would codify guardrails that sit between voluntary company policy and formal regulation.

The timing is notable. Hassabis just stepped back from running DeepMind, and safety standards could become his next arena.

full brief & sources

⚡ Why this matters

  • A credible standards body could become the layer between self-regulation and government mandates.
  • Whoever writes shared safety rules shapes launch speed and model access for every lab.
  • It marks the debate shifting from hypothetical risk to who verifies the rules.

🔍 What happened

  • The Wall Street Journal reported the proposal on Aug 13.
  • Hassabis has discussed it with executives at major AI labs and US officials.
  • Named officials include Treasury Secretary Scott Bessent and White House technology adviser Michael Kratsios.
  • The envisioned body would codify safety guardrails and shared practices for advanced AI development.
  • Hassabis has compared the concept to the International Atomic Energy Agency.
  • He recently moved from DeepMind CEO to chairman and Alphabet chief scientist.

💬 Smart takes

  • Wall Street Journal: the body would attempt to codify guardrails as developers approach more capable systems.
  • Skeptic: competing private labs accepting external oversight that slows their launches has no precedent - the IAEA analogy breaks on enforcement power.

🧭 Where this goes

  1. Likelyat least one more lab publicly endorses the concept within six months.
  2. Possiblethe body forms as a voluntary consortium with no enforcement teeth.
  3. PossibleWashington adopts it as a soft alternative to formal AI legislation.
  4. Wild Cardthe body gains real audit access to frontier models before 2028.

🥄 The Spoon Take

Industries propose self-regulation when real regulation starts to feel inevitable. Hassabis is moving early to make sure the rulebook gets written by the labs, not just for them. If it works, this body decides launch gates for everyone - including the labs that never asked for it.

🤔 Pushback

Without enforcement power or government backing, an industry safety body risks becoming a press-release factory that regulators ignore.

·Tech Startups - WSJ oversight report·CNBC - DeepMind leadership shift
PRODUCTGoogle

Gemini Crosses One Billion Users

14AUG
MEMBERS: 2GEMINI1B CLUB

Google's chatbot just joined the billion-user club. Sundar Pichai, Google CEO, says the Gemini app passed one billion monthly users, the fastest-growing product in Google's history. ChatGPT hit the same mark in June.

The climb was steep. 400 million users in May 2025, 900 million at I/O in May, one billion now. 63 percent of users talk to it by voice.

Distribution did the work. Gemini rides Android, Search, Workspace, and the new Pixel 11. OpenAI built a destination; Google switched on a default.

The race is now retention, not reach. Watch subscriber numbers, which Google left out of the announcement.

full brief & sources

⚡ Why this matters

  • Two chat products now serve a billion people each month - AI assistants are mainstream infrastructure, not early-adopter toys.
  • Google proved default distribution can catch a two-year head start.
  • Consumer scale feeds the ads and subscription models every AI lab needs.

🔍 What happened

  • Aug 11 - Sundar Pichai announced on X that the Gemini app passed one billion monthly active users.
  • Fastest-growing product in Google's 28-year history and its 14th service to reach the mark.
  • Growth path: 400M in May 2025, 650M in October, 900M at I/O 2026, one billion now.
  • 63 percent of users interact by voice; the app generates over 150 million images daily.
  • ChatGPT crossed one billion monthly users in June.
  • Subscriber counts were left out of the announcement.

💬 Smart takes

  • Sundar Pichai, Google CEO: the Gemini app is the fastest-growing product in Google's history.
  • TechCrunch: growth tracks Gemini's deep integration across Android, Search, and Workspace.
  • Skeptic: monthly actives bundled into Android and Search say little about paid demand - Google shared no subscriber number.

🧭 Where this goes

  1. LikelyGemini and ChatGPT settle into a two-horse consumer race, with everyone else fighting for niches.
  2. LikelyGoogle starts reporting Gemini engagement metrics to investors within two quarters.
  3. Possiblevoice becomes the primary chat interface by 2027, reshaping how assistants get designed.
  4. Wild CardGemini passes ChatGPT in monthly actives within a year on Android distribution alone.

🥄 The Spoon Take

The billion-user club now has two members, and they got there differently. OpenAI built the product people seek out. Google switched on the product people already had. Distribution just proved it can buy back a two-year head start.

🤔 Pushback

A billion bundled monthly actives can hide shallow usage - the number that matters is who pays, and Google didn't share it.

·Google - 1B users announcement·TechCrunch - Gemini growth
Wednesday Aug 12
STARTUPSGoogle

Jeff Dean's Startup Automates Science

12AUG
27 YEARSGOOGLEDISCOVERY

Google's top scientist built a company to replace lab work. Jeff Dean left Google after 27 years to launch Discovery Loop with three colleagues. Google is backing it anyway, as investor and cloud partner.

Discovery Loop wants to run thousands of experiments at once, automatically. It starts with machine learning research, then hardware and drugs.

Dean co-founded it with Sanjay Ghemawat, Oriol Vinyals, and Quoc Le. Radical Ventures and Khosla Ventures are backing the new company. Alphabet shares dropped 5% the same week Demis Hassabis moved to chairman.

Google still gets a stake and cloud revenue either way. Watch whether other chief scientists follow Dean out the door.

full brief & sources

⚡ Why this matters

  • A 27-year Google veteran just bet his career that research itself can be automated.
  • Google is funding a company built by people who used to run its AI research.
  • It's the second high-profile DeepMind-adjacent departure in a week, after Hassabis moved to chairman.

🔍 What happened

  • August 5, 2026: TechCrunch and GeekWire reported Dean's exit and new venture.
  • Discovery Loop is a Delaware public benefit corporation, with Dean as CEO.
  • Co-founders: Sanjay Ghemawat, Oriol Vinyals, and Quoc Le, all longtime Google Brain and DeepMind researchers.
  • Plan: automate ML research first, then expand into chip design, drug discovery, and clean energy.
  • Backers include Radical Ventures and Khosla Ventures; Google is a founding investor and cloud partner.

💬 Smart takes

  • GeekWire: called it the startup pitch that convinced a UW computer science legend to leave Google after 27 years.
  • Skeptic: Google funding the company that just took its chief scientist looks less like a clean break and more like an outsourced R&D lab.

🧭 Where this goes

  1. LikelyDiscovery Loop announces its first automated research result within 12 months.
  2. Possibleone or two more senior Google Brain veterans join within the year.
  3. PossibleGoogle eventually licenses or acquires whatever Discovery Loop builds first.
  4. Wild Cardautomated experiment loops produce a genuine drug or chip candidate before a bigger lab does.

🥄 The Spoon Take

Google didn't lose Jeff Dean, it outsourced him. Funding the startup that poached your own chief scientist is a strange kind of loyalty, but it means Google gets a cut either way. Watch for more of this: labs spinning off their best people into ventures they still partly own.

🤔 Pushback

Public benefit corporations promise a lot before shipping; 'automate science' rarely survives contact with a lab bench.

·TechCrunch - Dean's new startup exit·GeekWire - Discovery Loop pitch
Friday Aug 7
STRATEGYGoogle

Hassabis Steps Down As DeepMind CEO

7AUG
HASSABIS JEFF DEAN

Google just rewired its AI leadership. Demis Hassabis, DeepMind's Nobel-winning CEO, becomes chairman and Alphabet chief scientist. Koray Kavukcuoglu now runs daily operations. Jeff Dean leaves after 27 years to build Discovery Loop.

Hassabis told staff AGI is close and getting the next steps right matters more than management. He keeps Isomorphic Labs, the drug-discovery spinout. Kavukcuoglu reports straight to Sundar Pichai.

Jeff Dean built Google's core systems and led Gemini's early training. His new company, Discovery Loop, wants to automate scientific discovery. Alphabet is backing it.

Google reorganized mid-race against OpenAI and Anthropic. Bloomberg says the shakeup complicates that race. Watch where DeepMind's senior researchers go next.

full brief & sources

⚡ Why this matters

  • DeepMind is Google's engine for Gemini - a leadership change there touches every Google AI product roadmap.
  • Hassabis moving to AGI strategy signals Google thinks the science, not the shipping, is the next bottleneck.
  • Jeff Dean leaving after 27 years is the biggest single departure in Google's history as an AI company.

🔍 What happened

  • Aug 5 - Demis Hassabis steps down as Google DeepMind CEO, becomes DeepMind chairman and Alphabet chief scientist.
  • Koray Kavukcuoglu, previously DeepMind's research engineering lead, takes day-to-day command as SVP, reporting to Sundar Pichai.
  • Hassabis remains CEO of Isomorphic Labs, the AI drug-discovery spinout.
  • Jeff Dean exits his Alphabet chief scientist post after 27 years to co-found Discovery Loop, an AI-for-science startup with Alphabet support.
  • Hassabis wrote to staff that AGI is close at hand and getting the next steps right is critical for humanity.

💬 Smart takes

  • Bloomberg: the shakeup complicates Google's race with OpenAI and Anthropic.
  • Axios: the chairman role is a new structure - DeepMind never had one separate from executive leadership.
  • Hassabis, to staff: AGI is close at hand and getting the next steps right is critical for humanity.
  • Skeptic: chairman plus chief scientist can read as a graceful sidelining - the org chart now runs through Kavukcuoglu and Pichai.

🧭 Where this goes

  1. Likelymore senior DeepMind researchers depart within six months, some to Discovery Loop.
  2. LikelyKavukcuoglu tightens the DeepMind-to-product pipeline - faster Gemini ships, less pure research.
  3. PossibleDiscovery Loop becomes a magnet for AI-for-science talent across all the labs.
  4. Wild CardHassabis exits Alphabet entirely within two years to run an independent AGI institute.

🥄 The Spoon Take

Founders are stepping off the org chart across the industry, and Google just did it with the most decorated one. The read: managing a product factory and chasing AGI are now two different jobs. Google split them. Whoever holds the research crown at DeepMind in a year tells you which job won.

🤔 Pushback

Hassabis kept the chairman seat, the chief scientist title, and Isomorphic - this may be a title reshuffle, not a power shift.

·Axios - Hassabis steps aside·Fortune - AI leadership shakeup·TechCrunch - Jeff Dean's startup·Bloomberg - Race implications
Thursday Aug 6
STARTUPSGoogle

Jeff Dean Leaves Google

6AUG
GOOGLEJEFF DEAN

Google's employee number 30 just walked out. Jeff Dean, Google's chief scientist, and three legendary colleagues are starting Discovery Loop. The mission: AI that runs thousands of science experiments at once.

Dean joined Google in 1999 as employee 30. He built the crawling, indexing, and serving systems behind Search. Quoc Le, Oriol Vinyals, and Sanjay Ghemawat are leaving with him.

Discovery Loop is a public benefit corporation. It wants AI to run complete experimental loops - pick the hypothesis, run thousands of tests, iterate. Alphabet itself is backing the round, alongside Radical Ventures and Khosla.

Dean told the New York Times the goal is more experiments and better ones. When a founder like this leaves, the interesting question is what Google couldn't let him build inside.

full brief & sources

⚡ Why this matters

  • Google just lost four of the researchers who built its AI foundation in one day.
  • Automated science is moving from research demo to venture-backed company.
  • Alphabet invested in the team leaving it - a telling hedge.

🔍 What happened

  • Jeff Dean, Google's chief scientist and employee number 30, resigned on August 5.
  • Co-founders: Sanjay Ghemawat, Quoc Le, and Oriol Vinyals - all Google Brain or DeepMind veterans.
  • The startup, Discovery Loop, is a public benefit corporation with Dean as CEO.
  • Mission: AI systems that run thousands of experiments simultaneously, automating the full experimental loop.
  • Round co-led by Radical Ventures and Khosla Ventures; Alphabet, Kleiner Perkins, Lightspeed, and Doerr Capital joined.
  • Alphabet stock fell 3.9% on the news, shedding roughly $190 billion in market value.

💬 Smart takes

  • Jeff Dean: "You will get both a higher quantity and a higher quality of experiments, and that will lead to scientific breakthroughs and advances."
  • Founding team: "The next great frontier for AI is to go beyond answering questions and to begin making discoveries."
  • Skeptic: AI-for-science has been promised for a decade with limited commercial results - the field stayed largely experimental until recently.

🧭 Where this goes

  1. Likelya wave of senior researchers follows Dean out of Google within six months.
  2. LikelyDiscovery Loop signs pharma or materials partners before shipping any product.
  3. PossibleGoogle responds with its own automated-science moonshot under DeepMind.
  4. Wild CardDiscovery Loop's recursive self-improvement work becomes the regulatory test case for autonomous AI research.

🥄 The Spoon Take

Jeff Dean did not leave Google for a feature. He left because the automated-science bet needs a new company, not a new org chart. When employee 30 walks after 27 years, watch what he builds - it marks where the frontier actually is.

🤔 Pushback

Star-researcher startups often stall - Dean's last decade was management, and automated science still lacks a single commercial win.

·TechCrunch - Discovery Loop launch·NYT - Dean interview·Quartz - 27-year exit
STRATEGYGoogle

Big Tech's AI Capex Hits $725 Billion

6AUG
$725BFRONTIER

Big Tech has a new answer for record AI spend. Ben Thompson, Stratechery author, gives the doctrine a name. Wall Street is learning to grade the story, not the invoice.

Google raised its capex guide to $205 billion, up from $190 billion a quarter ago. Amazon moved to roughly $220 billion, Meta's floor rose to $130 billion. Same quarter, same move.

The logic flips capex from an infrastructure cost into a frontier bet. Google framed it plainly: apply compute now to compete where the frontier will be when Gemini 4 lands. Demand today is beside the point.

Markets are starting to price who has a credible frontier case and who doesn't. Andy Jassy, Amazon CEO, insists demand backs the spend. Analysts see the combined number passing $1 trillion next year.

full brief & sources

⚡ Why this matters

  • Capex is now the biggest line item in tech: $725 billion guided for one year dwarfs entire industries.
  • The justification changed: not demand forecasts, but positioning for where models will be. That is a different risk profile for every investor.
  • Whoever tells the credible frontier story sets the valuation. The doctrine is becoming a pricing mechanism.

🔍 What happened

  • Ben Thompson's Stratechery piece names the doctrine: 'the frontier case' for hyperscaler capex.
  • Google raised capex guidance to $195-205 billion, from $180-190 billion last quarter.
  • Google leadership: 'We wanted to compete at the frontier level of where the frontier will be when Gemini 4 comes out.'
  • Amazon raised its capex to roughly $220 billion; Meta lifted its guide floor to $130-145 billion.
  • The trajectory: roughly $226 billion in 2024, $410 billion in 2025, $725 billion guided now. Analysts see $1 trillion-plus next year.

💬 Smart takes

  • Ben Thompson, Stratechery author: the frontier case means you spend to compete at where the frontier will be, not where demand is today. Capex is strategy, not plumbing.
  • Andy Jassy, Amazon CEO: the spend is justified by demand: this is capacity for real workloads, not faith.
  • Skeptic: hyperscaler capex tripled in two years while a Fed study found no measurable productivity bump. That is a bubble signature, not a doctrine.

🧭 Where this goes

  1. LikelyMicrosoft matches with a raised guide next quarter, and $1 trillion combined for 2027 becomes the consensus number.
  2. Likelyearnings calls shift from ROI questions to frontier-credibility questions: do you have a model that justifies the buildout.
  3. Possiblea hyperscaler without a clear frontier model pays a valuation discount despite record cloud revenue.
  4. Possibledebt financing grows as capex outruns operating cash flow at one of the big three.
  5. Wild Cardone hyperscaler publicly cuts guidance in 2027 and triggers the first AI-capex correction.

🥄 The Spoon Take

Capex just became a story you tell, not a cost you justify. The frontier case gives every CFO a license to spend ahead of demand, and hands investors a sharper question: do you have a frontier model, or are you renting someone else's? That question will sort the trillion.

🤔 Pushback

Spending tripled while measured productivity barely moved; if frontier models stop improving visibly, the frontier case collapses into overcapacity overnight.

·Stratechery - The frontier case·CNBC - Alphabet earnings·UncoverAlpha - Q2 capex analysis
Sunday Aug 2
FLOPSGoogle

Google Cancels Its AI Studio App

2AUG
800K PREORDERSAI STUDIOCANCELED

800,000 preorders wasn't enough to save this app. Google canceled its standalone AI Studio app for iOS and Android. Those features move into Gemini, so apps emerge from chat instead.

Google teased the app at I/O 2026, promising app-building on the go. The preorder count was unusually high for a tool nobody had used yet.

The team thanked everyone who signed up, saying people clearly want to build software away from a desk. Google gave no date for when the Gemini version actually ships. That is a bet that conversation beats a home-screen icon.

The web version of AI Studio keeps running for developers shipping real products. Preorder counts, it turns out, don't always predict what people will actually use.

full brief & sources

⚡ Why this matters

  • Shows that raw demand signals, like preorder counts, don't always predict what people actually want.
  • Google is betting that AI app-building belongs inside a chat, not a separate app icon.
  • A rare case of a Big Tech AI product getting killed after public excitement, not before it.

🔍 What happened

  • Google teased a standalone AI Studio mobile app for iOS and Android at I/O 2026.
  • More than 800,000 people preordered the app before it shipped.
  • Google announced on July 31 that the standalone app is canceled.
  • App-building features will instead be folded into the main Gemini app.
  • Google says apps should emerge naturally from everyday conversations with Gemini.
  • No launch timeline was given for when the Gemini-based features arrive.

💬 Smart takes

  • Google AI Studio team: thanked the 800,000 people who preordered, saying it's clear people want to build software on the go, just not as a separate download.
  • Skeptic: canceling a product with 800,000 preorders after teasing it publicly risks looking like Google can't decide what AI Studio actually is.

🧭 Where this goes

  1. Likelythe Gemini app gains app-building features within the next two quarters.
  2. LikelyGoogle keeps investing in the AI Studio web platform for developers.
  3. Possiblethis becomes a case study in why preorder counts overstate real demand.
  4. Possiblea competitor ships a standalone AI app-builder and picks up the abandoned demand.
  5. Wild CardGoogle revives a standalone app once the Gemini features prove popular.

🥄 The Spoon Take

Eight hundred thousand people wanted this app, and Google killed it anyway. That's not a failure of demand. It's a bet that building software should feel like a conversation, not a download. If Gemini pulls this off, nobody will remember AI Studio was ever a separate app.

🤔 Pushback

Folding features into Gemini with no timeline could mean the app wasn't finished, not that chat is the better interface.

·9to5Google - AI Studio app canceled·Digital Trends - Google's app pivot·X - AI Studio announcement
Saturday Aug 1
COOL TOOLSGoogle

Gemini Falls Off Mollick's AI List

1AUG
AGENT MODELISTEDGEMINI

The AI guide power users follow just dropped Google entirely. Ethan Mollick, a Wharton professor, cut Gemini from his practical AI guide. It has no agentic computer-use mode like ChatGPT Work or Claude Cowork.

A year ago the guide was all chat: ChatGPT, Claude, Gemini side by side. Today it's split by which AI can actually use a computer.

Simon Willison, the developer behind Datasette, flagged the shift on his blog. ChatGPT's modes are Work and Codex; Claude's are Cowork and Code. Willison calls the naming 'spectacularly unintuitive' even for people who use both daily.

Gemini Spark, Google's answer, hasn't proven itself yet. Whoever wins the computer-use race owns the workflow, not the chat window.

full brief & sources

⚡ Why this matters

  • Shows where the real competitive battle moved: not chat quality, but who can safely operate a computer for you.
  • Google's absence from Mollick's list is a concrete signal, not vague criticism - Gemini Spark isn't there yet.
  • The naming mess (Work vs Codex vs Cowork vs Code) is a real adoption tax on every team evaluating these tools.

🔍 What happened

  • Ethan Mollick's practical AI guide, updated regularly since 2023, dropped Gemini from its current version.
  • A year ago the guide covered chat models: o3, Claude 4 Opus, Gemini 2.5 Pro.
  • Today it centers on agentic computer-use modes: ChatGPT Work and Codex, Claude Cowork and Code.
  • Simon Willison highlighted the shift on his blog on July 27.
  • Willison notes ChatGPT Work on mobile behaves very differently than Work inside the desktop app.

💬 Smart takes

  • Simon Willison: the mode names 'do not map onto each other in any way that will help you remember them.'
  • Ethan Mollick (via his guide): "Gemini Spark has yet to prove itself."
  • Skeptic: a guide reflects one influential professor's workflow, not confirmed market share data.

🧭 Where this goes

  1. LikelyGoogle ships a more capable Gemini agent mode within the next two quarters to get back on these lists.
  2. Likelymore operator guides converge on the same 'which agent mode' framing over chat comparisons.
  3. Possiblethe naming confusion forces one vendor to simplify its product naming.
  4. Wild Carda third-party standard emerges for describing agent modes across vendors, cutting through the naming mess.

🥄 The Spoon Take

The most useful AI comparison isn't model benchmarks anymore - it's who gets to touch your computer. Google skipping this list entirely, a year after leading model rankings, says more than any chatbot arena score. The keyboard, not the chat box, is now the battleground.

🤔 Pushback

One professor's personal guide isn't a market map - plenty of teams still run Gemini in production for cost, not capability, reasons.

·Simon Willison - Opinionated AI guide·One Useful Thing - Mollick's guide·Agensi - Work vs Cowork explainer
Sunday Jul 26
PRODUCTGoogle

Google Ships Gemini 3.6 Flash

21JUL
17% FEWER TOKENSFLASH 3.6CHEAPER

Google's cheap model got cheaper and smarter. Gemini 3.6 Flash launched with a lower output price and built-in computer use. The flash tier, not the flagship, is where the price war is happening.

The new release ships at $1.50 input and $7.50 output per million tokens. That's a lower output rate than the prior version.

It uses about 17% fewer output tokens to do the same job. The ability to click and type inside a screen ships built-in this time. Google also shipped a cheaper lite variant and a security-focused one alongside it.

On its own benchmarks, the new version beats the old one across coding and long-context tests. The knowledge cutoff also jumped forward, from January 2025 to March 2026.

full brief & sources

⚡ Why this matters

  • The flash tier, not the flagship, is where most production API traffic actually runs.
  • Cheaper output tokens change the unit economics for anyone running Gemini at volume.
  • Built-in computer use pushes agentic browsing into the cheap tier, not just premium models.

🔍 What happened

  • Google launched Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber on July 21, 2026.
  • Pricing: $1.50 per million input tokens, $7.50 output; cached input at $0.15.
  • Context window: just over 1 million input tokens, up to 65,536 output tokens.
  • Uses about 17% fewer output tokens than 3.5 Flash for equivalent tasks.
  • Beats 3.5 Flash on DeepSWE, OSWorld-Verified, MLE-Bench, and GDPval-AA v2 benchmarks.
  • Available day-one across AI Studio, the Gemini API, Android Studio, Antigravity, and Vertex AI.

💬 Smart takes

  • Google: pitches the release as a performance jump at a lower cost, not just a refresh.
  • Skeptic: benchmark gains on Google's own suite are easy to cherry-pick and hard to verify independently.

🧭 Where this goes

  1. LikelyOpenAI and Anthropic answer with their own cheap-tier price cuts within a month.
  2. LikelyFlash becomes the default model for high-volume agentic tasks, not Gemini's top-tier model.
  3. Possiblethe flash-tier price war compresses margins enough that a provider consolidates or exits.
  4. Wild Carda flash-tier model becomes capable enough to replace flagship models for most enterprise work.

🥄 The Spoon Take

Nobody's fighting over the smartest model this month. They're fighting over the cheapest one that's still good enough. Flash-tier pricing is where the real AI margin war is happening, not the flagship launches.

🤔 Pushback

Self-reported benchmarks from the model maker aren't independent verification, and a 17% token-efficiency claim is easy to construct favorably.

·9to5Google - 3.6 Flash launch·Tech Insider - Pricing benchmarks
Monday Jul 20
GOVERNANCEGoogle

EU Orders Google To Open Android

20JUL
JAN 2027 DEADLINEANDROID

Brussels just told Google it can't lock out AI rivals. The European Commission ordered Google to open Android to rival AI assistants. These are binding engineering rules, not a fine.

Users will get to launch rival AI assistants by voice, just like 'Hey Google'. Those assistants can book a taxi or reply to chats for the user.

Google must share anonymized search data with rival search engines starting January 2027. Android changes land for users by July 2027. Google says the rules could weaken privacy and security.

Kent Walker, Google's president of global affairs, called the decision a risk to Europeans' privacy. The order followed two years of failed talks over workable remedies.

full brief & sources

⚡ Why this matters

  • First time a regulator has forced structural, feature-level AI interoperability on a major platform.
  • Converts two years of stalled DMA (Digital Markets Act, the EU's rulebook for big tech gatekeepers) talks into concrete engineering deadlines.
  • Sets the template other regulators may copy for AI assistant access rules.

🔍 What happened

  • The European Commission issued the order on July 16, 2026.
  • Rival AI assistants get equal access to core Android features, not restricted access like today.
  • Users can trigger a rival assistant by voice, book a taxi, or get reply suggestions in chat apps through it.
  • Google must share anonymized search data with rival search engines starting January 2027.
  • Android changes for users are due by July 2027.
  • Google objected, warning the rules risk exposing private searches and business secrets.

💬 Smart takes

  • Kent Walker, Google president of global affairs: the decisions risk weakening privacy and security safeguards for Europeans.
  • Skeptic: a mandate this complex, with an 18-month runway, still leaves Google room to comply narrowly and slow-walk the spirit of it.

🧭 Where this goes

  1. LikelyGoogle appeals or seeks to narrow the scope of the anonymization requirements.
  2. Likelyrival AI assistants like Perplexity or ChatGPT integrate deeper Android hooks by mid-2027.
  3. Possiblethe search-data-sharing rule becomes the bigger fight, since it touches Google's core moat.
  4. Wild Cardthe US or another bloc introduces a similar Android interoperability mandate within 12 months.

🥄 The Spoon Take

Regulators used to fine AI platforms after the fact. This time Brussels wrote the engineering spec first. Whoever controls the phone's assistant layer controls agentic AI distribution, and the EU just decided Google doesn't get to own that alone.

🤔 Pushback

The rules don't bite until January and July 2027, plenty of time for Google to shape the technical details of compliance.

·European Commission - DMA specification·CNBC - Google AI rivals ruling
Tuesday Jul 14
STRATEGYGoogle

Five Giants Challenge Anthropic's Standard

14JUL
MCPRIVAL5 VS 1

Connecting AI agents to your tools just became a turf war. Google, Microsoft, Salesforce, Snowflake, and ServiceNow are backing a rival to Anthropic's MCP. MCP quietly became the default plumbing for AI agents.

MCP lets an AI agent plug into any business tool with one shared protocol. Anthropic open-sourced it, and it became the default almost by accident.

The five companies run the software where business data actually lives. None of them want to build on a rival's foundation. So they are shipping a competing standard instead.

Anthropic still has an 18-month head start with developers. But a five-company alliance can out-market and out-integrate almost anyone.

full brief & sources

⚡ Why this matters

  • Whoever owns the agent-to-software plumbing controls where the next lock-in happens.
  • This is the clearest sign yet that the model race and the infrastructure race are splitting apart.
  • Anthropic built the default without owning the incumbents' turf, and now the incumbents are pushing back.

🔍 What happened

  • Google, Microsoft, Salesforce, Snowflake, and ServiceNow agreed to back a shared standard for connecting AI agents to business software.
  • The move directly counters Anthropic's Model Context Protocol, or MCP, which has become the default connector standard over the past 18 months.
  • MCP lets AI clients like Claude or ChatGPT call tools and data sources through one common interface.
  • Snowflake and ServiceNow both run infrastructure that would otherwise depend on a protocol Anthropic controls.
  • Google separately pushed its own Agent2Agent, or A2A, protocol at Cloud Next, adding to the crowded standards field.

💬 Smart takes

  • Industry framing: 'The giants do not love building on a competitor's foundation.'
  • MCP adoption read: Model Context Protocol has 'rapidly become the default standard for connecting AI clients to external tools.'
  • Skeptic: standards wars in software usually end with the most-adopted option winning regardless of who backs the alternative.

🧭 Where this goes

  1. LikelyMCP keeps its developer-adoption lead through 2026 despite the new coalition.
  2. Likelyenterprise software vendors add support for both standards rather than picking one.
  3. Possiblethe new standard wins default status inside Salesforce, ServiceNow, and Snowflake's own products specifically.
  4. Wild Cardthe standards fragment permanently, forcing every AI vendor to support both.

🥄 The Spoon Take

MCP's real advantage was never the spec. It was showing up first and getting good enough. A five-company alliance can out-spend Anthropic, but it cannot un-adopt eighteen months of developer habit overnight.

🤔 Pushback

Standards wars usually get settled by whoever ships the best developer experience, not whoever has the biggest logos in the press release.

·CryptoBriefing - Rival standard vs MCP·TheNextWeb - Google's A2A push·andrew.ooo - MCP adoption state
RUMORSGoogle

Gemini 3.5 Pro Leaks Ahead Of Launch

14JUL
LEAKEDTOO SOON

Google's next flagship model is leaking before it's confirmed. Leakers say Google scrapped the original base and rebuilt it from scratch. Unconfirmed specs: a 2-million-token window, target date July 17.

Google has not confirmed a single detail. Every spec below comes from unnamed sources and leaked screenshots, not an official post.

DeepMind reportedly scrapped the first 2.5 Pro base after finding bugs in tool-calling and SVG generation. The rebuild runs on a native Gemini 3 foundation. Deep Think reasoning stays locked to the $250 Ultra tier.

July 17 is the target, July 24 the fallback. DeepSeek's own release deadline lands right in between.

full brief & sources

⚡ Why this matters

  • Google keeps missing its own frontier-model timeline while rivals ship weekly.
  • A scrapped base model suggests Gemini 3 had real technical problems.
  • The leak sets a specific date the market will now hold Google to.

🔍 What happened

  • Leaks point to a July 17 general-availability date for Gemini 3.5 Pro.
  • Google reportedly discarded its original 2.5 Pro base after tool-calling and SVG bugs.
  • The rebuild uses a native Gemini 3 foundation, per internal sources.
  • Unconfirmed specs list a 2-million-token context window, the largest of any frontier model.
  • Deep Think reasoning is rumored to gate behind the $250-per-month Ultra tier.
  • No model card, pricing page, or API listing exists publicly as of July 13.

💬 Smart takes

  • Tech Times: every specific claim about the date, context window, and pricing comes from third-party reporting, not Google.
  • X leaker Pankaj Kumar: frontend generation sees a major jump, with stronger SVG output and one-shot game demos.
  • Skeptic: Google has slipped this exact launch before; a scrapped base model could mean more delay, not a stronger product.

🧭 Where this goes

  1. LikelyGoogle either confirms or quietly misses the July 17 date within days.
  2. Possiblethe 2-million-token context window ships but with degraded quality at full length.
  3. PossibleDeep Think stays Ultra-exclusive to protect margin on the $250 tier.
  4. Wild CardGoogle preempts the leak with an early announcement to control the narrative.

🥄 The Spoon Take

A leak this detailed, four days before a launch nobody confirmed, is its own signal. Google is now competing against its own rumor. If Gemini 3.5 Pro under-delivers on July 17, the leak becomes the story instead of the model.

🤔 Pushback

Leaked specs from unnamed sources have been wrong before, and 'scrapped the base model' is exactly the kind of dramatic detail that gets exaggerated in the retelling.

·Tech Times - Gemini 3.5 Pro leak·Geeky Gadgets - Deep Think leak details·X - Pankaj Kumar leak thread
Thursday Jul 9
COOL TOOLSGoogle

Google Photos Ships AI Video Remix

8JUL
BEFOREAFTER

Your camera roll just got an editor built in. Google Photos now offers Video Remix, built on the Gemini Omni model. It rolls out free today to every Google AI Plus, Pro, and Ultra subscriber.

The feature lives in the Create tab, with a library of ready templates. Ask for a watercolor look, morning light, or a new background, and it renders in seconds.

Gemini Omni is trained to understand gravity and light, not just pixels. That is what makes an edit look real instead of pasted on. You can even drop a digital double of yourself into a clip, watermarked by SynthID.

The same model already powers free remixes inside YouTube Shorts, no subscription needed. Adobe and Canva now have a new AI rival to answer.

full brief & sources

⚡ Why this matters

  • Video editing has always required either skill or Premiere Pro tutorials.
  • Gemini Omni is Google's bet that AI video understands physics, not just pixels.
  • Free rollout to Shorts means hundreds of millions see this immediately.

🔍 What happened

  • Google announced Video Remix inside Google Photos on July 8, 2026.
  • It runs on Gemini Omni, first shown in May as a video-first model.
  • Templates handle style transfer: watercolor filters, relighting, background swaps.
  • Processing takes a few seconds per clip, according to Google.
  • It ships free today to Google AI Plus, Pro, and Ultra subscribers in the US and select countries.
  • The same Gemini Omni engine already powers free remixes in YouTube Shorts and Google Flow.

💬 Smart takes

  • Google: Gemini Omni can 'create anything from any input,' starting with video.
  • Engadget: Video Remix is 'designed to save you from sitting through hours of Premiere Pro tutorials.'
  • Skeptic: template-based edits cap creative control; power users will still open a real editor.

🧭 Where this goes

  1. LikelyVideo Remix expands to more countries and languages within a few months.
  2. LikelyAdobe and Canva add competing one-prompt video restyle tools within the year.
  3. PossibleGemini Omni becomes the default video layer across Google Photos, YouTube, and Workspace.
  4. Wild CardSynthID-watermarked avatars become a new short-form ad format brands pay to use.

🥄 The Spoon Take

Google just turned video editing into a prompt. That is a bigger deal than another filter app. Every past AI editor worked on top of your footage. Gemini Omni claims to understand the physics inside it, which is the harder problem to fake.

🤔 Pushback

Google's own examples are fairly subtle by its telling, and physics-aware claims from labs rarely hold up outside the demo reel.

·Engadget - Video Remix launch·9to5Google - Gemini Omni remix·Chrome Unboxed - Photos AI remix
Friday Jul 3
STRATEGYGoogle

Gemini 3.5 Pro Ships Without Its Builders

3JUL
2 GONEGOOGLERIVALS

Google's next big model finally has a ship date. Noam Shazeer, Gemini's co-lead, and AlphaFold's John Jumper left for OpenAI and Anthropic. Alphabet lost 5% in market value that week.

Third and fourth senior Google AI researcher to leave in months. One helped invent the Transformer architecture behind every large model today.

The other won a Nobel Prize for protein-folding work before switching labs. Google set a July launch date for its flagship model, a month late. The slip came from coding and long-task performance issues, not the exits directly.

Every senior departure raises the same question about Google's research edge. More counter-offers and faster releases could follow as labs compete for talent.

full brief & sources

⚡ Why this matters

  • Talent flight from a frontier lab is a leading indicator, not a lagging one.
  • Gemini's launch slip shows Google can't out-execute OpenAI and Anthropic on schedule alone.
  • Anthropic keeps landing marquee scientific talent, not just engineers.

🔍 What happened

  • Noam Shazeer, Gemini co-lead and Transformer co-author, announced his move to OpenAI on June 18.
  • John Jumper, the AlphaFold lead and 2024 Nobel laureate, joined Anthropic's science team days later.
  • Alphabet shares fell 5-6% on June 22 as the departures became public.
  • Google confirmed Gemini 3.5 Pro ships in July, a month after its original June target.
  • Gemini 3.5 Flash already shipped in May; only the flagship Pro model was delayed.

💬 Smart takes

  • Fortune: the departures raise doubts about whether DeepMind can stay at the AI frontier.
  • Axios: Google DeepMind is losing 'star power' just as the race tightens.
  • Skeptic: Google still ships Gemini 3.5 Flash, a genuinely fast, cheap frontier model, so one bad week doesn't erase that lead.

🧭 Where this goes

  1. LikelyGoogle announces retention packages or reorganizes DeepMind leadership within the quarter.
  2. LikelyGemini 3.5 Pro ships in July close to its new target date.
  3. Possibleone or two more senior researchers leave Google before the end of the year.
  4. Wild CardGoogle pauses its largest frontier model training run to regroup.

🥄 The Spoon Take

Model races get the headlines, but talent races decide who wins them. Two of Google's most decorated researchers just voted with their feet, for OpenAI and Anthropic. A launch delay is a schedule problem. Losing the people who set the schedule is a different kind of problem.

🤔 Pushback

Google still shipped Gemini 3.5 Flash on time and it's genuinely fast. Two departures don't prove the lab has lost its edge.

·Fortune - DeepMind departures doubt·Axios - DeepMind loses star power·Tech Times - Gemini 3.5 Pro cleared
Thursday Jul 2
PRODUCTGoogle

Google Ships Image-To-Video AI

2JUL
10 CENTS/SECPHOTOVIDEO

Turning a photo into video just got a price tag. Google shipped Gemini Omni Flash, a new image-to-video tool, on June 30. It edits video in plain language at 10 cents a second.

This pairs with Nano Banana 2 Lite, Google's fast image model. Chain them together and a prompt becomes a finished video clip.

Logan Kilpatrick, who runs Google's AI Studio, says the speed unlocks latency-sensitive uses nobody could build before. Nano Banana 2 Lite returns a full image in about four seconds.

Google is racing to become the backend every video app runs on. Both models are already live in AI Studio, the Gemini API, and Search's AI Mode.

full brief & sources

⚡ Why this matters

  • Video generation just got a per-second price tag instead of a subscription tier - that changes how builders scope a feature.
  • Editing video with plain-language prompts instead of a timeline lowers the skill bar for motion content.
  • Google is racing to become the creative-AI infrastructure other apps build on, not just a destination app.

🔍 What happened

  • Google shipped Gemini Omni Flash and Nano Banana 2 Lite (Gemini 3.1 Flash-Lite Image) on June 30, 2026.
  • Omni Flash turns images into video for $0.10 per second, edited in plain language.
  • Nano Banana 2 Lite returns a 1K-resolution image in about four seconds for $0.034.
  • Both ship immediately through Google AI Studio, the Gemini API, and Google's consumer apps.

💬 Smart takes

  • Logan Kilpatrick, Google AI Studio & Gemini API: "The speed of Nano Banana 2 Lite is going to enable so many new use cases where there is a high degree of latency sensitivity."
  • Skeptic: per-second video pricing sounds cheap until an app generates thousands of short clips a day - the bill scales with usage, not intent.

🧭 Where this goes

  1. Likelyvideo-editing and social apps integrate Omni Flash as a backend feature within months.
  2. Possibleper-second pricing becomes the norm for short-form AI video, replacing flat subscription tiers.
  3. PossibleOpenAI or Runway matches this price point within a quarter.
  4. Wild Cardimage-to-video at this price cannibalizes stock video and b-roll marketplaces within a year.

🥄 The Spoon Take

Google isn't chasing 'best AI video app' - it wants to be the backend every video app runs on. Ten cents a second is cheap enough that builders just wire it in without asking. That's the real fight: not model quality, who's the default plumbing.

🤔 Pushback

Cheap per-unit pricing has a way of turning into a surprise bill once usage scales - ask anyone who's run serverless functions.

·Google - Omni Flash announcement·Google Cloud - Availability details
Monday Jun 29
STRATEGYGoogle

Alphabet Loses $269 Billion

24JUN
-$269B4 LEFT6 DAYS

Talent is now priced like a balance-sheet asset. Four senior DeepMind researchers left for OpenAI and Anthropic in six days. Alphabet's market value dropped $269 billion.

The exits came fast. Noam Shazeer to OpenAI, then John Jumper, Jonas Adler, and Alexander Pritzel to Anthropic.

Markets did the math. Alphabet plans roughly $190 billion of AI capex this year. Losing the people who turn that spend into frontier models makes it look like buying depreciating assets.

The selloff was broad. The Nasdaq fell 2.2% on June 24, and the worry is whether $452 billion of hyperscaler capex ever pays off.

full brief & sources

⚡ Why this matters

  • Reframes AI talent as a market-priced asset, not a hiring footnote.
  • Ties a quarter-trillion market move to four people leaving in a week.
  • Surfaces the real investor fear: capex without the people is sunk cost.

🔍 What happened

  • June 18-24: four senior DeepMind researchers left for rivals.
  • Shazeer went to OpenAI; Jumper, Adler, and Pritzel to Anthropic.
  • Alphabet shed about $269 billion in market value across sessions.
  • The Nasdaq fell 2.21% on June 24; Micron dropped 13% intraday.
  • Hyperscaler 2026 AI capex now tops $452 billion combined.

💬 Smart takes

  • Analyst: the talent story is now part of the valuation story at this scale.
  • Demis Hassabis: Google has by far the biggest research bench, and lab-to-lab movement is expected.
  • Skeptic: 28 of 33 analysts still rate Alphabet a buy on its $460 billion cloud backlog.

🧭 Where this goes

  1. Likelypay alone stops working; compute access and autonomy become the real retention levers.
  2. LikelyGoogle's July launches get judged against the talent narrative, not just benchmarks.
  3. Possiblemore named researchers leave big labs for smaller, faster ones by Q3.
  4. Wild Carda public AI name takes a deeper capex-driven drawdown and resets the sector's multiple.

🥄 The Spoon Take

The market just put a price on a handful of brains. Frontier capability lives in a small group of people, and investors now treat their movement as a balance-sheet event. The lesson is blunt. You cannot buy a frontier position with capex if the people who build it can walk out the door.

🤔 Pushback

Four exits did not change Google's models overnight, and its cloud backlog and two billion AI users may matter far more than one bad week.

·CNBC - Alphabet talent exits·CryptoBriefing - $269B loss·Intellectia - June selloff
Sunday Jun 28
PRODUCTGoogle

Gemini Deep Think Tops The Benchmarks

22JUN
BENCHEDGEMINI WINSANTHROPIC

Google grabbed the top score while Anthropic's best models sit benched. Gemini 2.5 Deep Think beat GPT-5.5 and Fable 5 on graduate-level science. Timing is everything.

Deep Think uses parallel reasoning, running many thought paths at once. It scored 82.4% on GPQA Diamond, a hard science test. That beats GPT-5.5 and the suspended Fable 5.

The win lands while US export rules keep Anthropic's Fable 5 and Mythos offline. Google has a clear lane to claim the lead.

It is live for AI Ultra subscribers, with API access soon. Benchmarks are not products. But mindshare moves on leaderboard wins.

full brief & sources

⚡ Why this matters

  • Google takes the benchmark crown as its top rival sits benched.
  • Leaderboard wins still drive enterprise mindshare and developer pull.
  • Parallel reasoning shows the frontier moving to test-time compute.

🔍 What happened

  • Google launched Gemini 2.5 Pro with Deep Think on June 22.
  • Scored 82.4% on GPQA Diamond and 89.8% on MMLU-Pro.
  • Beat GPT-5.5 at 76.3% and Anthropic's Fable 5 at 79.1%.
  • Fable 5 is offline under a US government export order.
  • Live now for AI Ultra subscribers; API access coming soon.

💬 Smart takes

  • Google: Deep Think uses parallel thinking for harder reasoning.
  • Context: Fable 5's score predates its government suspension.
  • Skeptic: a few points on one benchmark rarely changes what teams ship.

🧭 Where this goes

  1. LikelyGoogle leans on the lead to win AI Ultra and Cloud deals.
  2. LikelyOpenAI answers with a Deep-Think-style reasoning push.
  3. PossibleAnthropic's export limits cost it real enterprise momentum.
  4. Possiblethe lead evaporates the moment a rival posts a higher number.
  5. Wild Cardtest-time compute pricing reshapes how labs charge for hard tasks.

🥄 The Spoon Take

The score matters less than the timing. With Anthropic's best models frozen by export rules, Google has an open lane and is taking it. Benchmarks are noisy and short-lived. But when your strongest rival cannot ship, even a small lead buys outsized mindshare. Regulation just handed Google a window.

🤔 Pushback

GPQA leads are fragile and rarely survive a month, and Fable 5's frozen score may understate Anthropic's real position.

·FAQ.com.tw - Gemini Deep Think tops·Google DeepMind - Deep Think model