2026-09-02 03:58:52
Sometimes I feel like I'm The Architect in The Matrix. You know, the guy in the room full of monitors in Reloaded who seems exasperated to have to explain to Neo that this is not the first Matrix nor is he the first version of 'The One'...
As the hype builds around AI "personal assistants", there's a real sense of deja vu. Basically since the dawn of technology, this has been viewed as the holy grail, certainly for consumer-facing services. From Apple's 'Knowledge Navigator' concept video to Microsoft 'Bob' to Clippy to Cortana to Copilot and everything in between, one day, such technology is just going to do all the tedious tasks for us.
Sadly, that day never comes. Often for a whole host of reasons, but ultimately, the technology simply isn't good enough. While it might be able to execute a demo or two, it quickly breaks at basically every other seam. It doesn't even seem to really matter what the underlying technology is, if it works at all, nothing can scale. Enter AI, the latest iteration of 'The One'.
Once ChatGPT cracked the code on chat-based AI, the talk quickly started shifting to "agents". That is, the prospect that the AI would be able to autonomously do things for you in the very near future. But on the way to this – yet again – inevitability, a more enterprising use case got in the way: coding. While OpenAI was focused on various consumer use cases, Anthropic found product/market fit with Claude Code (and Cursor, to some degree, largely powered by Anthropic's models).
OpenAI didn't seem to realize what a big deal this would be at first – to their credit, seemingly neither did Google! – but if Anthropic's surging business didn't quickly prove it, zooming past OpenAI in valuation certainly did. And that's because coding is both a big market, but also because it seems like Claude Code was the entry point to broader agentic uses. Thus, 'Cowork' was born.
OpenAI quickly reoriented with Codex, while SpaceX bought Cursor (after merging with xAI, of course). Google and Meta are also rushing towards this opportunity. But it seems like it's going to be hard to dislodge those big three in coding specifically. So the talk is now once again shifting back to the original and forever goal: personal assistants. That is, consumer-oriented agents.
This too required a wake up call, in the form of a lobster. When OpenClaw exploded out of the cage – perhaps to the point where Apple was selling out of Mac minis! – it was clear there was something to the notion of personal agents. Of course, that was always clear – again, see the past 50 years or so of computing – but the difference is that the technology may be right this time. May.
While OpenClaw was cool and fun, I quickly pointed out the challenges in getting mass consumer adoption for such projects. Allow myself to quote... myself (because it's a good line, if I do say so myself):
"People tend to fall in love with great movements but end up marrying great products."
OpenClaw was a movement, we still needed that great product.
All the powers that be were off to the races again. Meta (smartly) tried to jump ahead by buying Manus. But China had other plans. NVIDIA tried to layer on top of OpenClaw, recognizing the obvious security issues, with 'NemoClaw'. But despite the Jensen Hype Machine out in full force, we were still left searching. Amazon, Apple, Google, and Microsoft all had their talking points ready, if not exactly their products. As did Anthropic and OpenAI.
To their credit, SpaceXAI may have actually been the first of the big players to move here. Their Grok Bot is getting a lot of buzz in beta testing. Still mainly around work-oriented tasks, but it looks and feels awfully consumer-y. And it comes just as a handful of startups, notably Instinct and Town, have entered the chat. It took approximately 13 seconds for them both to achieve unicorn status. Why? Personal assistant AI, baby!
Of the two, Instinct seems to be getting the most love right now, as it's more consumer-focused and thanks to its unique onboarding flow and super proactive nature. That, in turn, has led to some inevitable backlash around security and trust. Regardless, we seem to be in a full-on race once again towards the mythical personal assistant.
Any day now, we should see 'Project Hatch', hatch. That is, Meta's 'OpenClaw for Normies'. Amazon is clearly cooking up something similar, as is Microsoft as they work towards their inevitable "Super App" – seemingly a key ingredient in this race. Apple is on the verge of launching Siri AI – for real this time – with many of these capabilities. And Google has 'Gemini Spark' – because everyone must have some sort of 'Spark' AI product – rolling out in phases. Meanwhile, OpenClaw is back with version 2.0 of their offering, which they promise has learned from the lessons of complicated onboarding and the like.
So the stage is set. But will it actually work this time? Are we really going to get viable personal assistant technology? Products that are actually useful? For certain tasks, I think so. But for many things, it's still hard to see it.
Obviously a lot of it will depend on the underlying technology. The models. While Anthropic and OpenAI maintain their leads in terms of agentic flows, SpaceXAI and Meta are clearly gunning for them with new releases due soon. But the question could ultimately come down to how much it matters to control the whole stack, as it were – the model and the harness – versus having the best harness which can leverage the best models, no matter who they're made by.
This has become more interesting in recent weeks with the surge of Chinese "open" models. While they may not be quite at the frontier, they're close. And they're currently offering such closeness at fractions of the American prices. (Or, of course, given the open weights, you can download the models yourself, if you have machines powerful enough to run them locally.) As things stand right now, there's certainly a case to be made that being model agnostic may be advantageous. And that would point to the startups here standing a chance.
While OpenClaw's founder was already "hackquired", the open source project lives on, though some seem skeptical of how at-arms-length it can actually be from OpenAI given that relationship. Or NVIDIA, given that relationship. But OpenRouter has certainly showcased how you can drive immense value in such an approach. And while they too were acquired, the fact that it's Stripe, a company not directly in the AI model space, seems like a good situation in terms of maintaining a "Switzerland"-like status.
On the flip side, OpenAI just terminated their contract with Cursor after the SpaceX deal, showcasing the risk of playing up this agnostic nature. And while Anthropic was quick to note their continued support of Cursor, you may recall when they helped torpedo OpenAI's Windsurf deal by pulling their models...
I point that out because it seems quite likely that Instinct and Town will have offers to sell soon, if they don't already. And depending on the acquirer, it could help or hurt or both (Cursor) their chances of truly winning the space.
There's also the question of if owning even more of the stack will matter. By that I mean, the actual devices on which these services are used. Beyond being able to tailor the models and software to run more efficiently, there's the privacy angle here as well. Obviously, Apple, Google, Samsung, and to some degree Microsoft are best positioned in that world. Though Meta and OpenAI would argue that new devices are needed for these new use cases.
Is voice the unlock to all of this?
It is interesting that seemingly no one has the absolute combination of model (that they fully control), harness, and hardware... At least not yet!
Maybe that doesn't matter. There's certainly a case to be made that all that will actually matter in the end is a killer product. Again, a few of the current offerings have interesting elements, but no one has nailed the true "personal assistant" experience. And while you can finally at least squint to see elements of it, it's still hard to see how anyone will in full.
Even if we get comfortable having these agents handle our email, will we ever fully offload say, travel? From research to booking. Certainly the former, but the latter still feels like a minefield of options and edge cases. Also, that's the example that is always cited forever and ever around such technology. But how much does that actually matter to the "average" consumer anyway? How many trips are they really booking in a year? Regardless, that use case does not pass the "toothbrush test".
I keep coming back to the idea that the first killer use cases may be more "fun" in nature. Letting agents do (lightweight) tasks for you that result in absolute delight. Opening up your to-do list and getting more comfortable over time with the technology taking over – again, assuming it can. It will be baby steps, not one big handover.
Is a startup well positioned without the baggage of other products weighing them down? The models protected OpenAI and Anthropic, what's the moat here? Is being 'Switzerland' enough? OpenAI has a strong product delight history. Meta always talks a big game and billions of users – not to mention WhatsApp – but some major trust issues, and they keep missing. Google has all the data – and Gmail! – but also some trust-in-themselves issues. Can Apple actually do it this time? No but seriously.
There's a long way between even just the technology and being a fully trusted and useful personal assistant. And we've been promised such things many times, for decades. There needs to be a killer product and a killer use case. Ideally many, but let's start with one?
To quote The Architect, "Hope, it is the quintessential human delusion, simultaneously the source of your greatest strength, and your greatest weakness."
2026-09-01 00:19:16
An OpenAI cover story gets thrown for a loop when their models hack a partner (one which NVIDIA may or may not be buying). And throwing some ice cold water on the blockbuster movie year for Hollywood. Read on for thoughts on...
2026-08-29 03:54:00
Tim Cook's watch over Apple officially ends this coming Tuesday. And while I'm no OneRepublic, I will just commend and congratulate him on one hell of a run. He had perhaps the most impossible task imaginable in succeeding Steve Jobs and he executed what will undoubtedly be a new textbook example/case study of how to follow up a beloved founder/literal savior of a company. And now his watch has ended. ✌️🍎
Inklings is a newsletter featuring links and commentary from M.G. Siegler on timely topics found around the web.
💰 NVIDIA's Regains the Profit Crown – While some headlines focused on the company closing in on $100B in revenue a quarter, plenty of companies do that – Walmart is nearly double that number, Amazon actually is. A far bigger deal is the $60B in profit NVIDIA just posted for the quarter. It has been bonkers for a while now, but it keeps getting more bonkers. And while Samsung briefly surged ahead last quarter due to the memory chip boom, these results puts NVIDIA back on top. And it's almost entirely actual profit, money NVIDIA gets to keep, not the silly (GAAP) net income figure which includes moment-in-time paper gains on investments – accounting which pushed Google's "profit" to a comically stated $112.2B last quarter. Their actual profit? Again, the money they keep? Closer to $30B. Roughly half of what NVIDIA just took home. [NYT]
🤗 NVIDIA's $12.9B Hugging Face Deal – Poolside. Perplexity. And now this. If you wanted a hint of how much profit NVIDIA was about to post, these deals – even if just rumored – were a good sign. I joked that NVIDIA clearly can't spend their cash quickly enough, but also, they clearly can't. Between bankrolling and backstopping data centers and investing in seemingly every company even tangentially related to AI at the moment, NVIDIA is essentially a bank. Yes, so was Cisco back in the day, but I digress... This particular deal actually makes quite a bit of strategic sense. I'm not sure the price does but again, does that really matter when you're making $60B in profit a quarter? Not if there was competition for the deal and you could keep it out of the hands of others to further your "open" AI narrative – not to be confused with OpenAI, which is also tied closely to NVIDIA, but is not open, and seems hell-bent on competing with NVIDIA in chips with the apparently quite spicy 'Jalapeno'. Or maybe you better know them as the company which just inadvertently hacked... Hugging Face. Wild times. [Information 🔒]
🦆 Hugging Face's $400 Duck Robot – Speaking of... while I don't think the 'Microduck' is contributing much to the $13B price tag here, it's also pretty fun and clever! Given the focus on robotics as a next logical extension of AI, I'm surprised nearly everyone seems focused on either the humanoid or warehouse variety. Why not just have a little fun while igniting some imaginations? Clearly they are over in Asia, and Amazon tried this to some extent – but as always, they were too early – but I bet there's a growing market in the US now too for people who want a Star Wars droid-like device, not an Ultron or Terminator-like robot – at least not yet. Especially as voice AI keeps leaping forward. Start small and fun. What's the old saying about big things that start out looking like toys? What if they actually start out looking like toys? Apple should be all over this. If for no other reason than it now seems like their longtime nemesis NVIDIA will be! [Bloomberg 🔒]
🐣 Meta's 'OpenClaw for Normies' Nears – I mentioned 'Project Hatch' earlier today as Meta's big bet on productizing AI for consumers. AI "Personal Assistants" are so hot right now – at least in the eyes of VCs, which, speaking from personal experience, at best doesn't mean much and at worst is negative signal. Everyone obviously wants agentic AI to work in consumer, but we need actual killer use cases beyond sorting your desktop icons or drafting email responses. Again, I would focus more on fun. What can these things do that is actually a delightful experience? Sure, easing the nightmare of booking travel sounds great, but, at least in my experience, that's also incredibly bespoke given various family requirements. Which is to say, I'm honestly not sure I'd ever trust AI to do that, even if it could. I realize that's a dangerous thing to say, but really I just want some obvious use cases for regular people and I have yet to hear any. Meta can shove this in front of 3 billion people, but without those use cases, it won't matter. Certainly if Meta is really going to charge money for this. Though I will say, being able to leverage WhatsApp as the potential UI here is interesting... Perhaps that's the real play for 'XChat' – Xitter's superflous chat app. Maybe it can be the Grok Bot interface? They just need to gain about 3 billion users first. [BI]
"I’m very pro-A.I. in music creation. I don’t see the downside at all. There will be some crappy music. There’s crappy music now."
– Jimmy Iovine, with his assessment of the AI backlash. At which point in the interview, Dr. Dre enters the room to drop: "I don’t see it as a threat. I think the only people that see it as a threat are the people who have trouble creating."
Ah yes, billboards that look like disclaimers are really going to resonate with people. Especially the intended audience. Nothing to see here, why do you ask? At least they double as an eye chart, I guess? This is not trending well for Meta...
🎶 Listening to "Release" by Pearl Jam (happy 35th to Ten)
🍺 Enjoying a Meantime Pizza Pale
🇬🇧 Sent from London, England
2026-08-28 19:38:06
There's a lot of positive buzz around Meta's AI efforts at the moment – if nothing else, it's impressive the speed at which they've zoomed back at least near the frontier. Can the forthcoming 'Watermelon' model fully get them there? Or set some kind of new mark? We should see soon enough, but more important for Meta will be creating an actual product people want to use AI for. Will 'Hatch' be the consumer harness that matters? The "OpenClaw for Normies" that I've been waiting for?
I remain skeptical simply because of Meta's recent history. When they can't buy, they've had trouble building products that actually resonate of late. It seemed like they found their buy-in here with Manus, but, well, China had other ideas. So the question in my mind is if they controlled the company long enough to effectively clone the IP before they had to give it back?
Anyway, this dive into Meta's history with AI by Harry McCracken gives some important context for the go-forward story. (And I swear I'm not just saying that because I'm quoted twice in the piece.) Mark Zuckerberg certainly gets credit for some early insight into AI being important back in 2013 when he hired Yann LeCun – but this was also because he had just lost the bidding to buy DeepMind. (And in pretty embarrassing fashion, according to subsequent reporting.) This directly led to the creation of PyTorch, the open source deep learning library which Meta absolutely does deserve credit for and propelled the entire field forward.
But as for the "open" strategy with their actual early Llama models, people forget, but that was decidedly more complicated:
While Zuckerberg was selling the world on headsets and smart glasses, a twist of fate helped propel Meta near the forefront of AI labs—at least for a time. In February 2023, Meta announced a new LLM called Llama (for Large Language Meta AI). At first, it planned to share its creation with academic researchers on a case-by-case basis. But a week later, as the world was still wrapping its mind around the three-month-old ChatGPT, a leaked version showed up on the dark web site 4chan. Suddenly, Llama was available to anyone who could figure out how to download and install it. And anyone who did could modify the code to suit their own purposes. Llama’s unintentional release led to fears of it being misused by bad actors. Some of that played out, including the creation of a Llama-powered Discord bot that spewed hate speech. But there was also excitement about the freely available model’s potential to democratize the spread of useful forms of AI.
Meta had been distributing the Llama weights to outside researchers, which is undoubtedly how they leaked, but there was no indication that this was the actual strategy for the general public. In fact, when it happened, everyone was scared shitless that Meta has just recklessly endangered the world! That was overblown, of course, and so it (retroactively – though undoubtedly guided by Meta's previous success with opening up some of their core technologies, including the aforementioned PyTorch) became the strategy:
OpenAI, Anthropic, and Google tightly control their primary models. Zuckerberg realized he could drive improvements more quickly by sharing Meta’s work. The prospect rattled other companies. “Paradoxically, the one clear winner in all of this is Meta,” wrote Google engineer Luke Sernau in an internal memo published by SemiAnalysis. “Because the leaked model was theirs, they have effectively garnered an entire planet’s worth of free labor.” Less than five months later, Meta leaned into that advantage. Rather than vetting users, the company simply released Llama 2 to the public as an open-weight model. (AI purists use this term for models such as Llama, whose training data remains proprietary, making them less than fully open source.)
Then came Llama 3, which was an even bigger success. Meta seemed to be off to the races with Zuckerberg talking up "open" models non-stop on every stage (and every podcast) that he could. Then the record scratch...
Instead of cementing Meta (and "open" models) as the leader in AI, Llama 4 was an unmitigated disaster. The gaming of leaderboards was a symptom, not the cause. While it was easy to blame "open", the reality was that Meta got outflanked by other models using mixture-of-experts (MoE) techniques to train. That was perhaps the biggest revelation of the "DeepSeek Moment", for Meta at least. Well that, and the notion that distillation of models can work quite well – perhaps especially if your model happens to have open weights.1
And so "open" was closed. And Llama was killed. Scale was "hackquired". And billions were burned (on both talent and compute). Again, the initial results look pretty good. But then came the "Kimi K3 Moment" mixed with the "AI Overkill" movement (and the Anthropic heel turn, and ongoing OpenAI backlash), and suddenly "open" was cool again! As such, Meta quickly pivoted back – though not fully this time – just in a way to try to take advantage of the marketing and framing as they try to re-establish themselves externally – and re-align themselves internally.
Back in 2015, Zuckerberg had proudly told me that his success as a CEO had come from the effort he put into “building a culture where people think about the mission in the same way that I do.” Now that synchronicity seems to be gone, at least for the moment. According to a current MSL staffer, employees in the lab regard Zuckerberg’s “personal superintelligence” talk as sloganeering, not a road map for building useful products. One Silicon Valley insider says the richly compensated AI recruits he knows at the company are “mercenaries,” there “to ride this out for as long as Zuck’s interests hold and take as much money off the table as possible.” Morale among longer-serving staffers, who have seen friends and mentors depart by edict or choice, has cratered, says the MSL employee.
"Sloganeering" is a great term. I still think if they nail 'Watermelon' (and can they come up with a better name than 'Muse Spark' for it?) and certainly 'Hatch' these problems will fade away. Success has a funny way of changing perception, fast and clearing up any internal morale issues. We'll see, it's nearly the end of this particular race...
1 Models which many other Big Tech companies were also happy to use, but not pay for, which was obviously an issue for Meta as well given the costs. It's almost like they needed a cloud... ↩
2026-08-27 21:36:13
🔗 Meta to Pay Up to $17.1 Billion in Landmark Settlement Over Social Media Addiction Claims • NYT
On one level, it's just math. $17.1 billion is less than $200 billion (the amount the states were seeking in damages). And certainly less than $1.4 trillion (the amount Meta claimed they could be on the hook for if they lost such a case – which happens to also be roughly their current market cap). Meta generates $17.1 billion in revenue every 26 days or so. Yes, Meta makes more money in a month than these settlement damages.
Also, those damages will be paid out over a decade. Also, also, they may not even be $17.1 billion, but instead more like $12 billion unless TikTok and YouTube also agree to similar settlements including damages and product changes! Oh and need I mention that Meta's market cap jumped about $50 billion on the news of the settlement?1
So yeah, this is one of those losses that seems like a win.
Yes, yes, technically it wasn't a loss at all, just a settlement without admitting any wrongdoing. But come on. Unlike the (pointless) antitrust case against Meta, which they rode through to the (easy) victory in the end, the company clearly read the writing on the Facebook Wall here. Money aside, optically, this was doing yet more damage to Meta's brand and reputation. Every story every day was equating their products to the new tobacco. And Mark Zuckerberg was about to have to take the stand. The media frenzy would have turned into Shark Week, with Zuck as the chum.
Oh would you look at that, a new trailer for The Social Reckoning just dropped. What an inconvenient coincidence!
So yeah, the settlement was a no-brainer. To the point where I'm honestly shocked the states agreed to it. Yes, $17 billion seems like a massive headline amount of money, and the payouts to each state ranging from hundreds of millions to billions (depending on population) will be meaningful to their budgets. But it's simply not a meaningful amount of money to Meta.
Far more interesting are the product changes that Meta is promising alongside the settlement. "Night Mode" and "School Mode" implemented at a service-level sound fine on paper, but in practice, are they actually going to work? A big part of the abbreviated trial was all about how such tools don't really work. Yes, the default positions will help, but how hard will it be for little Timmy to ask a parent to click this thing for them while they're busy on a million other things?2
Even more interesting is Meta's jiu-jitsu maneuver to turn this whole thing against their main rivals. The states will all now obviously be heavily incentivized to go after each of them to maximize their payouts. Meanwhile, Meta gets to frame themselves as the leader in teen safety here. As Casey Newton writes in Platformer:
Whether American teenagers should be harassed after 10 p.m. by engagement-baiting push notifications ought to be a matter for Congress, not the social media cartel. But with characteristic gall, Meta is seeking to position its game of prisoner’s dilemma with YouTube and TikTok as moral leadership.
The wolves are leading the call to stop other animals from attacking the henhouse! With an "open letter" running in national newspapers! This is some real Don Draper-level shit. Anthropic, eat your heart out.3
Naturally, such moral leadership only applies to the US, it seems.
So Meta can get back to work with the money printer largely intact. And this matters as that money needs to flow directly into their AI buildout. And potential damages aside, had Meta lost this case, it would have undoubtedly further spooked investors away from helping Meta to fund that buildout.
Yet risks remain, as Jyoti Mann and Aaron Holmes of The Information point out:
But unlike the big tobacco settlement, which effectively shielded firms from most future lawsuits by states and individuals, Wednesday’s agreement doesn’t resolve the thousands of open lawsuits brought against Meta by consumers, school districts and other parties. Also unlike individual lawsuits against tobacco firms, which didn’t ultimately succeed, individual lawsuits against Meta have already resulted in verdicts that found Meta liable.
At the same time, it also doesn't set any sort of precedent for other cases. That's certainly another reason to settle (if you think there's even a chance you're going to lose). As is this from the same piece:
However, the agreement includes a provision that it shouldn’t be cited against the company in other cases, noted Eric Goldman, associate dean for research and professor of law at Santa Clara University School of Law. That avoids giving plaintiffs in Meta’s other lawsuits a straightforward argument that the company has already acknowledged the allegations against it.
And so Meta walks away with what looks like a poke on the wrist. For now...
Not exactly the same, but some parallels here pic.twitter.com/5YGWHexCuw
— Arpit Gupta (@arpitrage) August 26, 2026
1 Though yes, it came back down after a down day for the market – but still ended the day up $15B. Pretty close to the settlement amount! ↩
2 To the US-only point, how much do you want to bet that there will be some VPN workaround? Too complicted for kids? Come on. What about changing the time on the device? You laugh now... ↩
3 But really, how long until the AI companies use a similar playbook? ↩
2026-08-26 22:57:12
Uh, Bill Gates sounds a lot like... Dario Amodei in writing up his thoughts on where he thinks the world is headed with regard to AI. It's a good essay, worth the read, but it's certainly not going to help the AI perception problems! (And it's also worth wondering why he is weighing in right now...)
Also, the Mac mini lives! But not the version you remember. It has been reborn as Apple's AI PC. Read on for thoughts on...
