2026-08-11 03:55:39
I'm old enough to remember the last time Mark Zuckerberg was embracing "open" as the future of AI. Which is to say, I'm at least two years old.
A lot has changed since the last time Meta released an open weight model – namely, seemingly the entire team building models for Meta. Famously, "Llama" was put out to pasture as Alexandr Wang and his Scale team was brought in to fix what ailed those previous models. That included not only spending billions on fresh AI talent and data centers, but also setting aside that notion of "open" in favor of a closed approach to AI development. Why? Because it sure seemed like that model – and those models – had won the day just over a year ago.
To hear Meta tell it – as I'm sure we're about to endlessly – the company never left "open" behind, they simply did what they needed to do in order to reset their efforts and catch up, fast. And there's undoubtedly some truth to that, but also some gaslighting. If OpenAI and Anthropic were still running away with the AI game – or if Meta was in the lead – we probably wouldn't be hearing too much about "open" right now. But again, a lot has changed in the past year.
Most notably, OpenAI slipped and fell behind their chief rival Anthropic. And Anthropic, in turn, has kept trying to commit seppuku to sacrifice themselves at the hands of an angry Trump administration. This, mixed with their new frontrunner status has led to an inevitable backlash against Anthropic, even though they mostly maintain their lead in models and in the all-important coding category.
But the real key was China seemingly catching up to that frontier of AI. Not quite, but perhaps close enough – closer than DeepSeek 18 months ago, and with AI in a more mature state, being "close enough" perhaps matters more now. And that's especially true given the cost and consumption concerns that a lot of corporate America is feeling at the moment with regard to AI. If those Chinese AI models are "close enough" at a fraction of the cost of the frontier models...
And while it may sound crazy to think that American companies would willingly use Chinese AI models in a time when the current administration has made it clear that China is the US's techno enemy #1, China left a door open to blur such lines with the focus on yes, open weight models.
If anyone can download these models and host them elsewhere – including on the clouds of the Big Tech players in the US – what's the problem? I mean, there are potential problems in that no one knows exactly what went into these models to train them as they're not actually "open source", as that's not the same thing as "open weight" despite what op-eds and New York Times headlines (see below) would have you believe. Still, much as was the case with the "DeepSeek Moment", I'm not sure how much the Chinese models will end up mattering in the end – as they're already tweaking their terms and models in showcasing why "open" here very much deserves the quotes – versus what the rise of such models point to.
That is: open weight models may now be a viable alternative to the closed variety and, relatedly, distillation of such models may actually spur innovation in the sector.
That's why we saw hundreds of US companies sign on to a letter to help ensure that the US government didn't destroy this "open" path in their attempts to slow down China. Obviously, the initial batch of companies that signed on, from NVIDIA on down, were conflicted with such ideals. But it created the intended groundswell to seemingly both get the government to heel but also pressure current "closed" AI leads like OpenAI and Google to sign on.
Yes, yes, both offer "open" models too, but only far less powerful varieties, released months after their frontier models. That is not what China is doing. And none of this is what Anthropic wants to become the norm. (They still refuse to sign the letter despite the pressure because they say while they're generally okay with open models to some extent, they're worried about the safety and security implications and ramifications if the frontier goes open weight. Which many read as the company protecting their moat. But others note that this is likely what the company actually believes.)
Anyway, that's a long-winded way – though about 1/10th as long as Mark Zuckerberg's new "essay" on the topic – of saying that Meta is back in the "open" game. Again, a game they'll say they never left, but they really did. And that actually may have been a mistake, as I noted a few weeks back. But it's also probably not too late to correct as the US "open" model race still feels wide open. And without Anthropic, OpenAI, or even Google fully committing to it (again, at the frontier), Meta may have an opening here.
And they're taking it.
To be clear, 'Muse Glimmer' is not at the frontier. Hell, it's not even at the frontier of Meta's own current model offerings – meaning, Muse Spark 1.2, which is by most accounts a very good, but not frontier-level model. They note that Muse Spark 1.2 will also get its weights released at some point soon, but again, Glimmer is not that. Which suggests that, as expected, Meta is going to follow the new general "open" playbook, at least in the US. That is, release a model, and once it's out there in the wild for a while, then you can "open" it up.
Why they didn't just wait to do this with Muse Spark if it really is that close to being "open" too, I don't know. But it seems like they aim for Glimmer to be smaller, to the point where it can run locally on some machines.
The real key will be what happens when Meta releases their actual frontier model, the one codenamed 'Watermelon', which many expect to be both coming soon and also competitive with said frontier. Will that too be "opened" up at some point shortly after launch? Meta isn't saying, but also, to be fair, they're not saying anything about 'Watermelon' beyond nebulously alluding to it. Because it's not actually out there yet. And talking about models before they're ready has gotten Meta in trouble before – see: Llama 'Behemoth' which no one ever did end up actually seeing in the wild.
My guess would be that if 'Watermelon' really is as good as the current frontier, the weights won't be released but it will instead be used to distill some other Muse variant that Meta will release as an "open" model... We'll see!
Regardless, it's clear that Meta, and Zuck in particular, are slamming on the gas. That's why you get a 7,000-word essay just a couple weeks after writing a decidedly more parseable version as an op-ed in The Wall Street Journal. This memo is verbatim to that at many points. So this is like the "Director's Cut" I guess. Boy does it need an editor...
Meta clearly sees "open" as a new opening here, which, yes, is ironic given that it was their initial path. But they also see openings in consumer/personal AI, especially with OpenAI seemingly taking their eye off that ball in their fight with Anthropic. As well as an opening on the business model front, where margins are always opportunities. Still unclear: can Meta make any of this pay off this time? They seemingly have more options now, but this is all completely unproven for them...
Zuck also clearly sees an opening around the messaging of AI. He wants Meta to hold the "positive AI" position. That's going to be easier said than done, quite literally. But yes, "open" is a part of that messaging. So "open" is back! No llamas this time though, sadly.
2026-08-10 04:05:29
You know the saying "less is more"? Well, someone needs to tell Paramount that the corollary to that is "more is less" because their strategy right now is "more is more" and it's fucking stupid. In fact, it might sink Hollywood. Or at least, the theatrical business:
Paramount Skydance has agreed to sign contracts with major theater chains guaranteeing that it will release 30 movies a year in cinemas if it acquires Warner Bros. Discovery according to people familiar with the matter.
Paramount has offered three-year agreements to AMC Entertainment and Cineworld Group’s Regal Cinemas, the world’s two largest theater chains, requiring it to release the films exclusively in theaters for at least 45 days, according to the people, who asked to not be identified because the agreements are private. The films would also not be available to stream online for at least 90 days.
On average over the past decade, it looks like Paramount has released about 10 movies a year. Warner Bros is closer to 12-14. Fingers-and-toes math suggests to me this number combined doesn't add up to 30. Yet somehow they're going to make it add up to 30. And the somehow isn't rocket science, it's a simple lowering of the bar to what should actually be released in theaters.
Yes, those averages are somewhat skewed by the pandemic.1 But you could also argue that the pandemic forced a (over) correction on the industry because too many films were being released in theaters when you consider the world of streaming we were about to enter. That is to say, the market has changed, and it actually wasn't the pandemic that changed it. Like so many other facets of business, it simply accelerated trends that were already happening.
The reality in 2026 is this: not every movie should have full-on theatrical distribution. In fact, most should not. It's not about "deserving" it from an artistic perspective, it's about it making sense from a business perspective. Does it make any sense to release a movie into wide theatrical distribution if it's going to flop? No. Not for the studio and not for the filmmaker. Would it be better to put that movie on a streaming service to find an audience? For many movies, yes.
And it's not black and white. Maybe you do a limited release for some movies. Maybe others only get a two-to-three week run before going to streaming. Maybe some go to streaming first, find that audience, then go to theaters (a la K Pop Demon Hunters). Maybe show finales do as well (a la Stranger Things). There are just so many options when you consider theaters as a piece of the pie, not the entire pie.
But that's not what Paramount Skydance is proposing, of course. They're promising at least 30 movies in theaters for at least 45 days with at least 90 days before they hit streaming. That's just silly in 2026.
And it's so stupid because it's the exact opposite of what they should be doing at the exact wrong time to be doing it. Yes, the box office is booming right now, but it's also a (predictable) pent-up fluke of major releases. Most years are not like this and newsflash: most years going forward will not be like this. And even this, relatively speaking, isn't that impressive. These are "record" numbers largely because of inflation, not because Hollywood has suddenly cracked some code.
So Paramount promising to release 30 movies a year in theaters is just going to lead to... more movies that shouldn't be released in theaters getting released in theaters. And that's going to exacerbate the problems theaters have been facing over the past, I don't know, 50 years or so.
I'm not going to say "shitty" movies because that's not really even the problem here. It's simply that increasingly, there are types of movies that do well in theaters – yes, big budget blockbusters, but also not always, and also not exclusively – but there are also many types of movies, even excellent movies, that simply don't work in theaters any longer. Sure, you can say that sucks. But it's also reality.
And so not just forcing them into theaters but forcing them there for 45 days is not just silly, it's stupid. It's going to lead not to better box office results but more empty theaters. Obviously.
It's true that Netflix had also committed to the 45 day window to try to win Warner Bros. But the key here is quantity. You could certainly make the case that the best movies should have such an exclusive window in theaters. But you cannot make the case that more movies than ever should. You just can't.
Again, this is so fucking stupid. We all know why Paramount is doing it – to try to seal the deal! But it's so obviously going to backfire that it's almost farcical. Get ready for Dear Santa 2 and 3 and 4. In theaters. For 45 days. Even if no one shows up to watch.
Update: Rich Greenfield of LightShed Partners pinged me on my tweet on the matter and points to their report that actually, Paramount and Warner Bros, per the current theatrical scheduling, were already likely above releasing 30 movies next year. Meaning, of course, that David Ellison is making a promise that actually may be easy to keep, at least next year; it looks like sandbagging.
That's interesting, but my only pushback – well, putting aside the notion that the overall company will have to cut costs when/if they actually merge to try to service the massive amount of debt being raised for the deal – there would be that Paramount is in the middle of ramping production post-Skydance deal while Warner is ramping post-killer box office returns last year (which, by the way, they're not currently replicating this year). Which is to say, I suspect their 2027 slates could be a fluke too – in volume, not necessarily in success – and wouldn't expect 30+ movies to the "natural" slate going forward. (See: above.) But it now will be if they merge! Because it's in the contracts! What could go wrong?
1 Also by the notion of a true "wide" release versus the smaller releases they each do via other sub-brand studio imprints. ↩
2026-08-07 23:46:08
I should have known. At this point, I've spent a handful of posts trying to figure out what OpenAI's first true device may look like. While I think I nailed the general idea early on – a voice-first, non-wearable smart speaker without a screen – the actual design for such a device has eluded me. Good thing Mark Gurman has started getting OpenAI leaks to go along with the Apple ones:1
A highly anticipated new device from OpenAI will have a unique look, complete with moving parts that help give it personality, and likely cost more than $300, according to people familiar with the matter.
The product — essentially a smart speaker without a display — will be shaped like a doughnut that’s roughly the size of a hockey puck, said the people, who asked not to be identified because the work is confidential. The idea is to make the device easy to carry around the home with one hand.
A doughnut. You know what else is shaped like a doughnut? An 'O'.
I'm probably reading too much into this, in fact I'm sure I am, but what the hell, it's a Friday and I'm getting ready to leave on holiday. What if, as silly as it may seem, the design of OpenAI's device is derived from OpenAI's name itself?
This isn't totally out of left field, back in 2024 I wrote about the notion that OpenAI was said to be undergoing a big rebrand all centered around the 'O' in their name. Obviously their logo is already a bit O-like, but there was some chatter that they could go full circle, quite literally. Perhaps some internal backlash killed it, or perhaps it was never a serious design sprint. Either way, the letter 'O' was clearly top of mind for them.
A few months before that report, the company introduced the GPT-4o model to much fanfare. There, the 'o' stood for 'omni' – as in that was their first model that fully ventured outside of text. Notably, including voice capabilities. Voice, you say...
A few months later, we got 'o1' the first reasoning model. I still wish OpenAI would have gone with the codename for that model – 'Strawberry' – but again, they were clearly all about the 'o'. And now perhaps they are once again, only with hardware.
The day after I wrote about the potential new 'O' logo, news broke that Jony Ive's LoveFrom design studio was working with OpenAI on some sort of newfangled device for the AI age. About eight months later, 'io' – there's that pesky 'o' again – the team specifically working on that project which spun out from LoveFrom, merged with OpenAI. Adding yet another 'o' to the mix.
Anyway, a doughnut-shaped device the size of a hockey puck sounds a bit strange unless you consider that it's supposed to be portable. Not necessarily to take outside (though maybe there will be some accessories for that?) but around the house. Longtime Apple watchers will know that the company has long been obsessed with handles. While it famously started with the Macintosh and continued with the Apple IIc, Ive seemed to double down on them with the iMac and iBook. The PowerMac G4 Cube had a pop-up handle. And the Mac Pro got them too.
So when I hear "doughnut shaped device" my mind immediately thinks that it's also a way to handle the device. Because that shape is a natural handle.
My mind was already drifting towards a handle when the reports last month had a few details about the smart speaker elements of the device. But I didn't consider that the entire thing could be a handle of sorts. That could make sense! And would certainly differentiate it from the other "smart speakers" out there.2
Of course, as I wrote last month, I'm also of the mindset that this device won't be framed as a smart speaker at all, but rather as a newfangled computer for the AI age. Or perhaps a companion – as in, a robot. John Gruber wrote about the same notion today at Daring Fireball on this latest report:
This doesn’t sound like a “smart speaker” with lights, moving parts, cameras, and other sensors. It sounds like a robot companion, and a robot companion obviously needs a speaker, amongst its other essential features. It’s like calling R2-D2 “a smart speaker” — it wouldn’t be wrong but it’s missing the point. (You could technically call the iPhone “a smart speaker”.) What the io device seems to be missing is autonomous movement. So you need to carry it around like a mogwai, rather than it following you around or going places on its own. I couldn’t give two shits about an OpenAI “smart speaker” if the point is just to play music and podcasts, but a handheld cross between R2-D2 and C-3PO, that could be fucking cool.
It’s a pet, it sounds like. A revolutionary pet, if they pull it off. And people love their pets.
People hear "robot' and immediately think of a humanoid robot – a sort of C-3PO. But the first actual robots we're going to get in our homes, at least at any scale, aren't going to be Tesla's Optimus, they're going to be super simple and perhaps even stationary – well, without an assist from you – robots. Yes, yes, we've long had Roombas,3 but I mean robots you can actually talk to and interact with in a meaningful way. Yes, the first batch of smart speakers are now getting there thanks to their LLM upgrades, but they don't evoke any real emotional response yet in usage. It sure sounds like OpenAI's device will be aiming for that:
OpenAI expects users to rely on the smart speaker throughout the day. It will work similarly to the company’s ChatGPT voice mode on smartphone apps, but with more advanced models for humanlike interactivity. The device is designed to learn more about a user over time, letting it tailor conversations and act more like a real person.
The circle-shaped device will include parts that move on their own, according to the people. That will help show when it’s responding and interacting with the user. The goal is to make the object feel more alive than today’s stationary speaker products.
The moving parts element was also included in the reports last month about the device. And that's why, when Apple's OpenAI lawsuit hit, I immediately thought that it might not actually be about current devices but future ones. Because by far the most compelling future device I've seen out of Apple's research department was this sort of weird, anthropomorphized lamp. Yes, like the Pixar lamp, Luxo Jr.
That type of device – which, to be clear, is still just a prototype in a lab somewhere within Apple – reminds me much more of what OpenAI may be going for here versus, say, a HomePod. Or even the forthcoming 'HomePad' thing.
So yeah, a doughnut/handle/'O' shape makes some sense. I don't know where you put the camera on such a device – perhaps it's meant to "stand"? – or where the moving pieces come into play? I keep envisioning little arms/nubs that wiggle to convey something, but who knows. But I feel like this device may be more interesting than it sounds on paper. Because I feel like it may be the first real robot in our homes.
An actual, physical chatbot, as it were. Bringing the 'O' to home.
One more thing: $300 to $400? That sounds pretty premium. Also very Jony Ive.
1 These leaks seem especially interesting when you consider that both note that OpenAI isn't worried about crossing any of Apple's patents with this device. It sure feels like OpenAI would very much like that notion out there. But again, what about the unreleased products Apple is working on? Such projects sure seem to be a focal point of the messages shared (on both sides) for the lawsuit... ↩
2 Lawsuit aside, presumably OpenAI wants nothing to do with being compared to Alexa/HomePod/etc with this device. While those sold well – at least Amazon's Echo devices – they also seem comically outdated and rudimentary in our current AI world. ↩
3 Sorry Lina Khan! ↩
2026-08-07 00:20:51
The most interesting element in Meta's newly announced 'Muse Code' product isn't the product itself, it's the model. Not the AI model, which is an updated variant of their 'Muse Spark' (non-flagship), but the business model. Per The Wall Street Journal:
Meta’s Muse Code agent has two price tiers: one that’s the same as its general Muse Spark model and another that is less than one-10th the cost. To access the less-expensive tier, which costs 20 cents per million output tokens, users must agree to provide feedback to help improve the agent. Tokens are the basic unit of artificial-intelligence computing.
Yes, you read that correctly. If you opt-in to having your data used to improve the models, you get more than a 10x discount on using those models. Actually, depending on the token type, quite a bit more! If my math is right, it's a 75x discount on cached input tokens. That is... sort of wild.
And also sort of brilliant. As everyone knows, Meta is coming from behind in AI after having to restart their efforts. And the fruits of such labor have been pretty good to date, but also still not frontier-level. And there's some concern that they won't be able to get to frontier level because that line keeps moving, pushed by the incumbents, Anthropic and OpenAI. I mean, not only has xAI struggled to catch up even after billions spent, even Google is struggling to keep up.
Part of the issue is obviously that the leaders simply have so much more usage that they've reached a kind of virtuous cycle, not unlike Google did back in the day with Search. Microsoft and Yahoo poured billions into trying to compete, but they simply could never catch up (let alone make Google "dance").
What's one way to spur usage and try to break customers away from the leaders? Well, a better product can work. But beyond not being so simple, that often takes time, enough time that if it starts working, the incumbents are likely to copy you. So what's a better way? Price.
Mark Zuckerberg has made no secret of the fact that he plans to undercut those competitors to get back into the race. And given the pressure both Anthropic and OpenAI are under to show improvements in their economics as they angle to go public means that a full-on price war is going to be a problem for them. Meta, as an already public and profitable – well, at least before all that AI spend came along – company can afford to undercut, quite literally. And so their margins are Meta's opportunity.
But this is actually even more interesting than that relatively simple and straightforward playbook. Currently, many businesses are having the AI cost come-to-Jesus moment. This includes both small businesses and even Big Tech. Many are learning the hard way just how fast AI costs can spiral out of control.
As such, we're seeing a pivot in the messaging around AI for businesses from maximizing usage to cost controls. Microsoft is leading the charge here, but Meta is right there too. Sure, this is easier to do when you don't have an actual frontier model to sell, but that doesn't mean it's not a good angle. And while Microsoft is focused on being a router between any and all models (well, aside from maybe Google's) so customers can make up their own mind on costs, Meta is here with a new model – again, a new business model.
This also taps into yet one more element being talked up right now against the 'Big AI' incumbents: customer data. Palantir's Alex Karp is leading this charge, but Microsoft's Satya Nadella is right there with him. The argument is essentially: you'd be crazy to give your data over to the Big AI players. You're giving them free rein to use that data to train their models and you're paying them for the privilege!
Never mind that this is fairly overblown given that there are some data protections in place around training and data security and what not, but the high-level point remains. And it does lead to the flip-side question, the one Meta is now trying to answer: if you were paid, would you be open to letting one of the AI model makers use your data?
By "paid" I of course mean, given that massive discount on token usage. Still, it's an interesting trade off. It's one that many big businesses can't make for security reasons. But individuals and perhaps small businesses can probably live with such a choice. At least, that's what Meta is trying to find out.
And actually this also plays into yet another trend at the moment: the push to keep the Chinese "open" models in play in the US market. Why? Well aside from the whole open weight debate, they're simply so much cheaper to use at the moment. Granted, there are already signs this may be shifting. But probably not so far so as to be close to what the incumbents are charging for their frontier models. Per WSJ:
Claude Code and Codex come bundled in pricing plans that cost roughly $20 a month, with more expensive plans for increased usage. Exceeding the usage cap switches the user to pay-as-you-go rates. Rates per million output tokens range from $12 for GPT-5.6 Terra and $30 for Sol, with Claude Sonnet 5 at $10 and Opus 5 at $25.
Meta’s new coding agent is priced roughly on par with models from China, such as DeepSeek, that cost as little as 18 cents per million output tokens. OpenAI has also slashed prices on older models, such as GPT-5.6 Luna, which dropped from $6 to $1.20 per million tokens.
In other words, without the data opt-in Meta's Muse Spark price, at $1.25/million (input) and $4.25/million (output), is priced fairly in line with the American competition (again, for the non-"flagship" models), winning in some cases, losing in others. But where things get really interesting is with that data opt-in. Because now we're talking about $0.10/million (input) and $0.20/million (output). Yes, 10 and 20 cents. Again, that's roughly inline with DeepSeek (which is apparently in the process of raising their prices).
Granted, this is all to use Meta's new Muse Code product (and API), but there's no reason such prices – and the business model – couldn't translate to their broader Meta AI suite as well. The company is in the process of trying to figure out how best to monetize that element of the business. And if it does, will it pressure OpenAI and Anthropic to offer the same kind of deal?
That will be a painful pill to swallow as they're currently getting such data for "free" (but yes, there are ways to turn such rights off or restrict them). And yes, OpenAI does grant higher token limits if you remain opted-in to letting your data help train their models. But if Meta's token discount trade-off idea takes off...
You can't help but be reminded of the famous line around advertising-based business models, "If you're not paying for the product, you are the product." Here, the equivalent is sort of, "if you're not paying for the tokens, you are the tokens" – meaning, the trade-off to get those tokens for "free" (or insanely cheap) is that you're giving up your inputs to help train those models.
Of course, with many (but certainly not all) advertising-based models, "free" really is free. With AI, at least to date, free is free up until a certain point and/or capability, at which point you have to pay. If you squint, you can see a path forward here, where the data trade-off perhaps keeps and expands free usage, but also makes the paid tiers (or pay-as-you-go token usage) cheaper.
The issue is that whereas with advertising-based businesses, the advertisers are paying the companies, with AI data sharing, there is no actual money coming in from that alone. You can certainly argue there's still value in that training data, but just how much and if it will always be constant is a question.
And, of course, that alone wouldn't be enough to help AI companies actually pay for all of this. Again, there's no actual money coming in from those data rights, simply (potentially) less going out. So the model would have to be some sort of hybrid of data-supported free tier, data-supported paid tier, non-data-supported higher paid tier and perhaps even advertising to augment all of them.
I've been skeptical about advertising working well alongside at least our current AI products. And certainly that it could ever work as well as it does with Google Search and/or Facebook/Instagram. But what if it simply needs to augment that data "payment" and/or that actual payment to keep the whole system working in a sustainable manner?
You can see a path to such business models supporting the training and usage of such AI models. There would be trade-offs, for sure, but to truly scale AI, I'm not sure it's the worst idea. Let's see where it gets Meta.
2026-08-06 06:16:05
To me, at least until we have more actual reporting on the matter, the most interesting element of the Google bombshells today is simply the timing.
Just Jeff Dean leaving Google would obviously be a massive story by itself. I mean, he's been there almost 27 years. He was employee #30. As everything you read today will repeat, he was not just vital, but instrumental to a lot of what made – and makes – Google, well, Google. He's one of just two Google "Senior Fellows" – the company's highest technical honor.1
Of course, the other Google Senior Fellow, Sanjay Ghemawat, is also leaving. Less well-known externally, but just as critical to the history of the company (and industry) for his technical work alongside Dean over decades, he's now joining his compatriot in forming a new startup.
So yes, just news of his departure alone would also be massive. But together, as Steven Levy so colorfully puts it: "that’s like Mick Jagger and Keith Richards ditching the Rolling Stones to start a new band."
But those two are also joined by two other seemingly critical people to Google's AI efforts in Oriol Vinyals, the VP of research at DeepMind (and a technical lead for Gemini), and Quoc Le, a cofounder of Google Brain. Both were famously co-authors of a 2014 paper about scaling data for pre-training AI models – one of the key papers for the foundation of LLMs. The other co-author? One Ilya Sutskever.
[As an aside, sort of harsh that Sundar Pichai's note calls out Dean and Ghemawat but not Vinyals and Le? Then again, a lot of senior people have been leaving Google recently... Which leads me to...]
These departures were announced the same day that it was announced that Demis Hassabis would be stepping back – sorry "stepping up" or "aside"? – from his day-to-day work at DeepMind and into a more nebulous and higher minded role as Alphabet's Chief Scientist (though also still leading day-to-day at Isomorphic Labs, his other Alphabet project where he's CEO). I mean, what is going on at Google?!
The answer, at least in what you read right now, is perhaps less dire than it may seem on the surface. Dean and team, after collective decades on the inside, seemingly wanted to try their hands at a startup in the hottest space during the hottest time to have a startup: the AI Boom. And they seemingly picked the hottest current sector to start with: recursive self-improvement.
Yes, RSI is the new AGI, at least until RSI leads to AGI. Maybe. One day.
Anyway, on the DeepMind side, everything you read suggests that Hassabis had already effectively handed off day-to-day operations to DeepMind SVP Koray Kavukcuoglu. The fact that Kavukcuoglu was (and will remain) Google's 'Chief AI Architect' – a title which he may or may not have received when a certain Mark Zuckerberg came calling with a "Godfather" offer – lessens any blow to actual operations even further. He already reported directly to Pichai. He will continue to report to Pichai.
At the same time, he will not be the new CEO of DeepMind – because no one is getting that title. Because DeepMind is now clearly going to be run less autonomously and more as a part of Google's overall AI efforts.2
And that's probably a good thing, to be honest. Because after already having to course correct when they fell behind in AI once, well, it sure seemed to be happening again. The reporting there this time points less to timidity and more the usual corporate in-fighting and bureaucracy. Google always has too many cooks in the kitchen, but the actual chef can't seem to keep them corralled and on the mission to make the meal. Because there were also too many chefs! That's why we used to get a dozen chat apps, many of which are half-baked. And now a dozen AI coding projects. Hopefully Kavukcuoglu has the mandate to unite the cooks. And to get the meal prepped and plated.
We'll see. But what remains wild to me is that all of these departures were announced at once. Couldn't Google convince either Hassabis or Dean's team to wait a week or two to blunt the impact – certainly on the market, but also internally? Maybe something was leaking. Or maybe they decided that packaging them together would actually help soften the blow versus the alternative.
Because the alternative not only includes all of the above people, but also includes Noam Shazeer, unexpectedly jumping ship a couple months ago – to join OpenAI, no less. And John Jumper – another Nobel Prize winner at Google alongside Hassabis3 – john-jumping ship, to join Anthropic, no less. And Peter Norvig – Google’s head of research for the past 25 years – heading to Recursive Superintelligence, working on, you-know-what.
You add all this up and it doesn't look just like a trend, it looks like an exodus. A brain bleed the likes we perhaps have never seen in such a short amount of time.
So yeah, had we gotten hit with Dean, Ghemawat, Vinyals, and Le – followed by Hassabis (even yes, with him technically staying, but in a new, less day-to-day role), it would have actually seemed even more catastrophic. Like Google couldn't stop the bleeding. Better to rip the band-aid off, perhaps?
Regardless, it's still an optics nightmare. I mean, these people were just on stage together a couple months ago talking about the future of AI at Google as three of the four co-leads of Gemini. Now they're all gone – except for Kavukcuoglu.
Again, I think there's a way to read this that is not actually catastrophic and could actually end up as a good thing for the company and the AI efforts overall. Well, at least in so far as you think the need to productize and commercialize AI is vital to Google right now, with research and a push beyond LLMs (areas which have always clearly interested Hassabis far more) taking a bit of a backseat.
It's going to be hard for people to see this right now. I mean, just look at today's headlines matched with the other headlines about the departures and the fact that Google has clearly flubbed the Gemini 3.5 Pro model and... yikes.
But the flip side is that things seemingly weren't going great with the status quo so... a shakeup can be a good thing sometimes. You hate to lose talent, let alone historic talent, but the way forward is also likely to be paved with new technology and new talent. Now Google needs to rally the troops to ensure they don't lose even more. And that's probably what today's announcements were about. Take the lumps quickly, make it all seem quite orderly, and get everyone back to the kitchen. It's time to cook.
One more thing: Where is Google co-founder Sergey Brin in all of this? For all the reports about him being back and fully re-engaged around AI, interesting that he's not mentioned in any of these reports...
1 Cue the "these go to 11" jokes as that's the level inside Google at which a "Senior Fellow" sits. Again, there were two of them yesterday. Now there are zero. ↩
2 Will this lead to more retention issues? We'll see... ↩
3 Fun fact: there were actually three before Jumper left, with Michel Devoret, the 2025 Nobel Prize winner in Physics, leading up Google's quantum computing efforts. ↩
2026-08-05 22:33:18
Satya Nadella was going to make Google dance.
It was February of 2023 and Microsoft's CEO was feeling confident that the overhaul of Bing, now powered by OpenAI's models, was going to change the search game. To be fair, it sort of did. But only in so far as it was one of a number of things that perhaps kicked Google into gear to eventually disrupt themselves with their own overhaul of Search, powered by their Gemini models.
Of course, the far larger kick in the ass came from ChatGPT itself and not Bing's variant. Microsoft's version became more famous for suggesting that reporters leave their spouses and getting quickly reined in. Soon afterwards, ChatGPT would go on to reign supreme in AI and Microsoft was back to the drawing board...