Apple turns 50 years old tomorrow. I’ve been using its products for 48 of those years.

48 years. Over those five decades, my relationship with Apple has shifted as dramatically as its market cap. And not, I am afraid, in a good way.

I first used an Apple product in 1978. I was in middle school when my parents brought home a new Apple II. My mother, a teacher, took advantage of Apple’s focus on the education market. I remember writing papers using huge floppy disks and digging the dot-matrix printer, but that was about it. In the late 70s, Apple seemed like a cool new company at the forefront of a cool new industry. But what did I know, I was a kid.

By the time I left for college, my father had an IBM PC, which I rarely used, and my mother had upgraded to an Apple IIc, which came out at roughly the same time as the Macintosh. I’d taken a few programming courses at my high school – I could write a tic-tac-toe game in BASIC, run DOS programs from the C: prompt, and futz around with some Pascal. But I was a writer at heart, not a coder.

In college I knew enough about computers to cobble together a cloned IBM 286 machine, which I used to earn extra money scripting databases for a small software developer. I couldn’t afford the Mac – it was priced at around $2,500 in 1984 – roughly $8,000 in today’s dollars. But I had some wealthy friends, and when my boss asked if I had access to a Mac to beta test a new app he was building, I borrowed a friend’s machine and fired it up for the first time.

As I wrote in The Search, that moment changed my life. I instantly knew that this machine was the most important artifact ever created by humankind. I wanted to tell the story of its impact on the world. My first job out of college, at a startup magazine called MacWeek, was reporting on Apple and the ebullient industry surrounding it.

Apple in its early years was a pirate company, positioned as a counterweight to the hegemony of much larger companies that dominated the nascent computer industry. Microsoft and Intel were not just behemoths, they were evil, mendacious, and utterly corporate. Apple, on the other hand, represented creativity, human spirit, and independence. Those of us in the “Macintosh community” cast ourselves as morally superior underdogs – the heroes of an unimaginably exciting new story.

That’s not to say we didn’t view Apple as an adversary. My early career was driven by a reporter’s zeal to overcome the company’s famous secrecy. We reveled in scoops about the company’s new hardware, software, and business strategy. We scrutinized its every move and executive utterance. Our motivation was more more than reportorial glory – we felt it was our job to keep Apple on track, to ensure it would win in a world dominated by ugly companies with unsavory values. If, as we believed, all of society would someday be driven by this emerging digital industry, we wanted the good guys to win.

I map the rise of digital technology over the past half century in nine overlapping eras. Apple features prominently in most of them:

1970s–1984  — Early Personal Computing
1984–1990    — The Macintosh Era
1985–1993     — The First Online Services
1993–2001     — Early Web
1994–2002    — The Dot-Com Boom and Bust
2003–2012     — Search, Social and Web 2.0
2012–2018.    — Rise of the Oligarchy
2018–2022    — Consolidation and Political Power
2022–pres.    — The (Early) AI Era

I used Apple products in each of those eras, and I’m still using a Mac today. But I’ve avoided all of Apple’s products after it entered the smartphone market. I don’t use the iPhone, I’ve never relied on iCloud, and Apple’s app store remains a foreign destination. I switched to Google’s Pixel in 2012, and I’ve never looked back. The reason? I felt that Apple had taken its business strategy of vertical integration too far. With the iPhone, Apple began to act like all those companies it once railed against: A massive juggernaut bent on locking its customers into beautifully designed walled gardens.

The worst offender? The Apple App Store, where Apple dictated what software its customers could and could not use. Steve Jobs famously called mobile carriers “orifices” that locked their customers into paternalistic and deeply  misaligned relationships. With the App Store, Apple built the biggest orifice of them all.

Which leads me to why I decided to write about Apple today. As I laid out above, we’re now in the AI era of computing. Apple hasn’t exactly been a leading player in AI – but it’s certainly poised to be. The company wasn’t a player in search or social either, but thanks to its near death grip on distribution, it managed to profit handsomely from both those developments. The same strategy is playing out in AI. Those 1.6 billion active iPhones will all be running AI, AI that can only be accessed through Apple’s orifices. And Apple will happily make hundreds of billions in AI profit along the way.

Unless, of course, the AI ecosystem treats the app store as damage, and begins to route around it. That seems to be the case with AI coding apps, which allow end users to build, well, whatever the hell they want to build. That reads as dangerous to the Apple’s corporate interests, and yesterday, the company did exactly what one might expect a dinosaur to do when faced with mammals scurrying around its feet. It stomped. (It already stomps all over Mac-based AI coding, for more on that, see my last post).

50 years into the Apple revolution, the rebel has become the tyrant. There’s much, much more to say about how Apple is architecting control over the coming AI wave. But for today, on the cusp of the company’s 50th birthday, I’ll leave it at this: If Apple has its way, our industry, and our society, will be much the poorer for it.

—

You can follow whatever I’m doing next by signing up for my site newsletter here. Thanks for reading.

Last night I dreamt I was merging onto a rushing freeway. My on-ramp was far too short, a concrete embankment hemmed me in to the right. Faceless, speeding vehicles filled the lanes; integrating with them would require icy determination and perfectly executed timing. Missing the merge would bring certain death. The dream began after the point of no return – I was already accelerating into the flow, braking was not an option.

Do, or die.

My “writing brain” is often active during dreams, and as I sped toward that critical merge, a detached third-person narrator considered the meaning of my situation. This narrator simply knew that the speeding vehicles and the freeway itself represented the political economy of techno-capitalism – amoral, headlong, impersonal and ruthless. It also knew that my current reading of Sven Beckert’s Capitalism informed this perception – it’s a devastating history of the progress and the damage wrought by a revolution centuries in the making.

But what was I doing on this ramp, accelerating towards either certain death or exhilarating integration? My narrator had no theory on that question. I was simply acting. I looked over my left shoulder – an 18-wheeler barreled toward my path, I’d have to punch it and swerve in, hoping the truck would relent just enough to let me assimilate. I closed my eyes, floored it, and executed my play.

“We’ve seen this plot before, but not executing at this speed.“

That comment, left on my last post by a Professor of Entrepreneurship at Michigan, neatly sums up how I feel about the moment we’re in. He was responding to my observation that we’ve seen periods of extreme openness and experimentation in tech, but they’re always followed by consolidation and lock-in by corporations that leaves the ecosystem feeling poorer and less innovative. Today’s moment just feels much faster, and far more consequential.

Moving this fast is both exciting and troubling. It’s impossible to not break things at this speed. On the one hand, playing with AI feels exactly like the days of the early web – everybody’s tinkering, experimenting with new ideas and imagining new possibilities. But the pace is exhausting, as is the code-switching required to work with a strange new form of intelligence. We’re not taking the time to consider externalities or unintended consequences, and it often feels that we’re hurtling forward, slightly out of control, hoping it’ll all work out.

That dream is starting to make sense.

I started tinkering with Claude two years ago, but got serious about using it for projects just last year. One of my first ideas was to build a full-text database of everything I’ve ever written on this site. I’d then wrap that corpus with Claude’s chatbot interface. My goal was to use Claude to identify themes, arguments, and inconsistencies across the more than 1.5 million words and 5,800 posts I’ve written in the nearly 25 years since starting this site.

My initial attempts at building the “Searchblog Query Engine” ended in frustration and failure. Claude kept telling me it was possible, but I found it difficult to follow the steps it laid out – the technical chops required were beyond my skillset (and patience). I tried again this past January, after the release of Opus 4.5, and got much closer. At that point my issue was complexity – Claude wanted to do something called “vector embeddings” using OpenAI’s API. That made me nervous – I don’t like the idea of becoming dependent on anything from a company I don’t trust. After a few hours of noodling, I once again abandoned the project.

Earlier this week, I set out to try again. Instead of picking up from three months ago, I started fresh. This time Claude took a much more streamlined approach, walking me through the technical bits with patience and clear instructions. OpenAI’s API never came up  – I could use Claude’s instead. I knew just enough about API keys, the Mac’s Terminal application, and how scripts work to follow along.

45 minutes later, this was live:

I mean…holy shit. 

In less than an hour I built myself a tool I’d been dreaming about for years. What else might I build? What else might others build? Might this augur a world where tinkerers and dreamers once again lead us into an optimistic future, a future where platforms add more value than they extract? It’s hard to not hope for such an outcome, but harder still to reconcile such dreams with the present-day realities of platform policies, incentives, and power.

Toward the end of the Searchblog Query Engine project, Claude and I identified what felt like an interface bug. Each time I wanted to run the engine, I had to first open Terminal and execute a line of code. That felt non-intuitive, so I asked Claude for a workaround. “I can write you a small launcher,” Claude answered, “a double-clickable icon on your Mac desktop that starts the server automatically, so you don’t have to touch Terminal at all. Want that?”

“Hell yeah,” I responded. This would effectively turn my new creation into an app that lived on my computer. How cool is that?! Claude obliged, and a launcher called “Searchblog.command” just…appeared on my desktop. Huh, I noted, this is new. Claude now had root level access to my computer. That’s cool – I trust Claude and the work we’d done together.

But when I double-clicked on my new app, Apple begged to differ:

Apple gave me two choices: Move Claude’s offending piece of code to the trash, or hit “Done,” which ignored it altogether. Apple had taken control of my project, and left me unable to use it.

I have a fair amount of experience with how big tech platforms control access to their platforms, and while Apple’s warning was defensible, not offering me a way to bypass its prohibitions was inexcusable. Who’s in charge here – me, or Apple? I was on the final step of creating something I was genuinely excited about – a true bicycle for my mind, in fact – and Apple was treating me like an errant child.

I uploaded Apple’s admonition to Claude, which told me that Apple’s new MacOS, called Tahoe, “tightened Gatekeeper significantly.”

That got my attention. “What do you mean by “tightened Gatekeeper?” I asked.

“Gatekeeper is Apple’s system that checks apps and scripts before letting them run,” Claude responded. “It’s been around for years but each macOS version has made it stricter.”

Claude offered a workaround in Terminal, which I ran, and now the Launcher works flawlessly. But that experience left me a bit shaken. There’s an hidden world of code constraining what most of us can and cannot do with technology, and as we’ve seen again and again, those constraints almost always favor the business models of the tech platforms who enforce them.

I’ve researched and written about these constraints for years. But this one small example was the first time I’ve directly encountered their bite. As the world adapts and merges with the capabilities and complexities of AI, I’m certain it won’t be the last.

—

You can follow whatever I’m doing next by signing up for my site newsletter here. Thanks for reading.

It’s “phase two” of the AI boom, and the claws are out.

Back at the tail end of 2024, I wrote these words: “2025 will not be the year AI agents take off.  As the bloom came off the Generative AI rose in 2024, everyone started talking about AI agents as the Next Big Thing. Google, Apple, OpenAI, Microsoft, Meta, Amazon – all of them (and about a million startups) are trying to build user agents for both enterprise and consumer use cases. I’m a huge fan of the concept, but for now, it remains just that. Reasoning agents that book your travel, negotiate your insurance bills, or manage your calendar simply will not work if they are beholden to the same business models currently driving Big Tech.”

My prediction proved accurate – in 2025, anyway. But three months into 2026, it seems AI agents are not only everywhere, they’ve also got a mascot, and it’s a crustacean. Everyone in China, apparently, has gone all in on “raising lobsters” – using OpenClaw to automate nearly everything computer mediated. And as the Wall Street Journal reports this morning, lawyers, doctors, and entrepreneurs alike are racing to become power users of new agentic tools that lets them prompt their ideas into reality.

“Across the planet, everyone is tinkering,” notes Om Malik, usually one of tech’s most skeptical observers. Malik highlights what is perhaps the most important features of this year’s breakout trend: It’s not controlled by the tech oligarchy of Google, Meta, Apple, Microsoft, Amazon, and, more recently, OpenAI. OpenClaw, he writes, “represents a philosophy. The intelligence lives on your machine. You own it. You aim it. No subscription. No permission required.”

If this sounds familiar, you may have read my warning about the rise of generative AI three years ago: “We dream of genies, but who will they work for?” The piece lays out why I’m both excited and concerned about the potential of generative AI agents – they hold the promise of finally breaking us free of walled garden business models that have trapped all our data in profit-seeking amber. But if AI is driven by those same models, we may never have the chance to find out.

“Internet business models have been built to collect short term rent,” I wrote, then included a breakdown of OpenAI’s “terms of service” to prove my point. “Truly open systems rarely win over time,” I conclude, “regardless of whether the company uses the word “open” in its name.”

Add another one to the list: OpenClaw. Last month, aspiring tech oligarch OpenAI appeared to acquire OpenClaw — technically, its founder joined, but the narrative is clear. I understand why founder Peter Steinberger hitched his financial wagon to OpenAI’s rocket ship. He’s now a millionaire and his project will now have nearly unlimited resources. And OpenClaw has grown exponentially in the month since it became an “independent foundation” with OpenAI’s “support.” But let’s not forget — OpenAI itself was once an independent foundation “unconstrained by a need to generate financial return.”

That didn’t last.

For the moment, open, user-controlled systems like OpenClaw are dominating the tech conversation across society. It feels exactly like the early web – everybody tinkering, unconstrained by the dictates of corporations or governments. But we’ve seen this movie before, and it’s always ended the same way: Early enthusiasm for something new – be it home brew computers, web browsers, mobile phones, social networks, or app stores – always consolidates into the hands of ruthlessly capital-efficient corporations. It’s just never happened as quickly as it did with OpenClaw. I guess we have AI to thank for that.

—

You can follow whatever I’m doing next by signing up for my site newsletter here. Thanks for reading.

I was going to write a long piece on the implications of the ongoing cage match between Anthropic and the US government, but as I dug into the research, I realized that hot takes on subjects this complicated rarely add much value to the debate. I’m going to let things cool a bit and take another run at it down the road.

But something important kept tugging at me as I was reading up on what I believe is the most significant regulatory action ever taken in the tech industry (if you believe listing a major US company as a “supply chain risk” is NOT government regulation, you’re fooling yourself).

What kept coming up as I read all those hot takes was this: Anthropic finds itself at a unique and utterly novel moment in time, one that just might let it become a major consumer platform. To wit:

• Anthropic’s Claude is now the #1 downloaded consumer app for iOS, and recently broke into the top 5 on Google Play. Claude had languished below the top 100 at the start of 2025.
• Visits to the Claude website increased elevenfold in 2025 – from 16 million in January 2025 to 176.12 million in December 2025.
• While Claude’s increase in traffic in 2025 can largely be attributed to its original focus on B2B and enterprise usage, where it’s considered a leader, Anthropic’s management had the consumer in mind well before the current controversy. In January – before Anthropic’s cheeky Super Bowl ads, and well before the current Pentagon imbroglio, SimilarWeb estimated Claude.ai visits increased to 202.9 million, a 15 percent increase in one month. The February number, which will be published by mid-March, will likely show a much larger jump due both to the Super Bowl and the Pentagon news.
• The Super Bowl (on February 8) pushed Claude from #41 to #7 on the iOS App Store and drove a 32 percent US download spike and 15 percent global download spike. That momentum carried throughout February, with the app staying in the top 20 most of the month until the Pentagon conflict pushed it to #1 over the past few days.
• Yesterday, Anthropic announced two key upgrades to its Claude app, both of which are focused on the consumer: First, it added memory features to the free version of its app, matching OpenAI, and second, it added the ability to “absorb” the memory of competing apps, making switching from OpenAI or Gemini far less painful for consumers.

Anthropic has made its reputation – and its historic sprint to $14 billion in annualized revenue – on the back of a focused strategy that caters to enterprise clientele. It’s also built a brand based on being the more cautious and thoughtful of all the major model makers. It’s always had a good consumer app – I’ve been using it exclusively for more than a year – but until recently, the consumer market felt like an afterthought. In late 2025, OpenAI claimed 800+ million users, and Google’s Gemini had grown to 750 million. Claude’s users for the same period? A paltry 30 million.

But while Claude is tiny by comparison, it’s become a champion at converting free users to paid subscribers. Yes, most of those paid subscriptions were for business and enterprise use cases, but Anthropic is at a key inflection point: It’s got the world’s attention, it’s got a strong consumer value proposition – “we’re the good guys in tech, if you use us, your data won’t be used by the government” – and it’s already plowed the road to becoming a consumer brand with its Super Bowl ads and recently introduced competitive product features.  Kind of reminds me of another company in the early days of tech, one with tiny marketshare and a unique take on the world. (Yes, I mean Apple back in the 1980s).

Most observers of the AI industry estimate that Anthropic earns just 15 percent of its revenue from direct consumer subscribers. Given the past week’s news, I expect that number to change dramatically – if the Pentagon’s “supply chain risk” threat holds, enterprise revenue will slow dramatically, just as consumer revenue will inflect upwards.

What might it mean for Anthropic to become a consumer company at scale? For one thing, the company might have to reconsider its now-famous aversion to advertising. Time – and usage data – will tell. If Anthropic manages to retain the flood of new users checking out Claude, this fight with the US Government might prove to be the fulcrum to a major pivot in the company’s long term strategy.

You can follow whatever I’m doing next by signing up for my site newsletter here. Thanks for reading.

Earlier this week, as a major storm took aim at the little island where I live, I saw a story in which Sam Altman, CEO of OpenAI, defended his company’s energy use by comparing it to how much energy humans use to do similar tasks.

“…it also takes a lot of energy to train a human,” he argued, when asked about AI’s insatiable – and destructive – appetite for energy. “It takes like 20 years of life and all of the food you eat during that time before you get smart. And not only that, it took the very widespread evolution of the 100 billion people that have ever lived and learned not to get eaten by predators and learned how to figure out science and whatever, to produce you.”

It was late in the evening, and I was about to lose power, but the insanity of Altman’s comparison struck a nerve, and I wanted to call it out. I’ve mostly refrained from my old habits of dunking on idiotic shit through social media – back when I was on Twitter, I’d regularly engage in the practice. But I left Twitter when Altman’s fellow oligarch Elon Musk purchased (and ruined) the place, and in the past few years, I’ve started using LinkedIn as a home for various outbursts, most of them tame in comparison.

But Altman’s ridiculous statement got under my skin, and I reverted to my old Twitter ways. “What a total asshole,” I posted, along with a link to the TechCrunch piece.

The post began to gather steam, logging 15 comments and nearly 60 likes in its first half hour.  Most folks agreed with my sentiment, but a few pointed out that my comment was not entirely in character. “Not on Sunday please ” pled one commentator. Another responded, quite reasonably, with this: “Curious, given your background in writing and thinking through complex problems, there are some clear logical fallacies in Altman’s argument worth dissecting. What drew you to name-calling over that analysis?”

He was right. I didn’t have the time to write a proper post about the topic, as the storm had already taken down several trees nearby and we were busy laying in firewood for what turned out to be a four-day power outage. So I dashed off an apology of sorts: “You make a completely fair point. I’ll try to do better.”

I left it at that, the power went out, and for the next few days I forgot about the incident.

But checking my mail yesterday, I got my second-ever takedown notice from a social media site (we’ll get into the first in a minute). “Your post doesn’t comply with our Professional Community Policies on bullying and harassment. It’s been removed from LinkedIn and only you can access it.”

A social media site that removes posts for name calling?! I suppose it does make sense. LinkedIn is a professional network, and calling someone an asshole is certainly not professional. It’s not easy to impose “community standards” on a platform of 1.2 billion people, and I’ve got no issues with this particular slap on my wrist.

As for Twitter (nee X), well, that’s another story. My first ever violation of a social media site’s community standards came in late 2022, as I was both leaving Twitter and setting up an account at Mastodon, an open source, federated version of Twitter. Here’s the offending post, which was cross posted to Twitter from my Mastodon handle:

For those of you with a sense of history, I was poking a bit of fun at Elon with that post – the very first post on Twitter was from co-founder Jack Dorsey:

Well, Elon’s new team did not like my sense of humor, apparently, nor did they appreciate my linking to a direct competitor. My post was flagged for violating community standards, and degraded in search and X’s algorithmic feed.

What a bunch of assholes!

You can follow whatever I’m doing next by signing up for my site newsletter here. Thanks for reading.

Well, they’re here. Just a quick note for now (lots more to say later, but a board meeting in SF means that’ll be later) – OpenAI is rolling out ads to its free and “Go” paid tier. The ads look…harmless enough, just a sponsored link unit with small graphics at the bottom of the chat. Pretty much the exact launch playbook we saw from Google 25 years ago, and Facebook in 2012. A rudimentary prototype of what will become, over the next few years, an increasingly sophisticated monetization platform that, let’s face it, will probably make Instagram look tame.

OpenAI also rolled out some pledges: “We decide which ad to show by matching ads submitted by advertisers with the topic of your conversation, your past chats, and past interactions with ads. For example, if you’re researching recipes, you may see ads for meal kits or grocery delivery. If there are multiple advertisers, we’ll select the one that is most relevant to your chat to show you first….Advertisers do not have access to your chats, chat history, memories, or personal details. Advertisers only receive aggregate information about how their ads perform such as number of views or clicks.”

If you line up OpenAI’s pledges, it sounds awfully familiar: We won’t sell your data … but we will lease it in aggregate and target you personally. Ads won’t effect chat results … but we reserve the right to “evolve our advertising program to support additional formats, objectives and buying models and build new ways for businesses to interact with consumers in ChatGPT.” Truck, meet wide open hole.

You can follow whatever I’m doing next by signing up for my site newsletter here. Thanks for reading.

“Claude is a space to think,” the company announced in a blog post that promised to never let advertising creep into its core consumer product. “The history of ad-supported products suggests that advertising incentives, once introduced, tend to expand over time as they become integrated into revenue targets and product development, blurring boundaries that were once more clear-cut. We’ve chosen not to introduce these dynamics into Claude.”

This is exactly the point I was making in yesterday’s post – “Advertising Built Generative AI. Now Comes the Remodel.” And while Anthropic’s written post is both thoughtful and measured, the company also launched a four-pack of ads illustrating its point – ads that they will be running during the SuperBowl this weekend. Yep, the SuperBowl.

In the videos, Anthropic’s messaging is anything but subtle. Here’s one of them, “Betrayal,” where a pitch-perfect, dead-eyed AI therapist pivots from a question about a patient’s mother to a hard sell for a MILF dating site:

And here’s “Violation,” in which an eerily ripped AI assistant tries to sell shoe lifts to a young man looking to build muscle:

“Deception” plays on the same theme – a gratuitous AI chatbot tries to sell an entrepreneur on a payday loan scheme:

And in “Treachery” an AI professor counsels a student to celebrate turning in a good essay by treating herself to jewelry.

Anthropic knows exactly what it’s doing by tacking into the AI ads debate, and I can only imagine the fits these ads are giving its main competitor OpenAI. Actually, thanks to social media, we don’t have to wonder – OpenAI CEO Sam Altman couldn’t help but respond, and clearly, a nerve has been struck.

Well played, Anthropic. Now let’s see if that SuperBowl spend delivers a positive ROAS (that’s Return on Ad Spend, for those of you taking notes…).

—

You can follow whatever I’m doing next by signing up for my site newsletter here. Thanks for reading.

Last night my wife looked up from her phone, disgusted. “All I’m getting is Jeffrey Epstein and Peter Attia!” she said. “Why do they think I’m interested in this?!”

As the family’s resident interpreter of digital entrails, I felt responsible to hazard an answer, but given the prurient nature of the Epstein story, I sensed my thoughts might not be well received. So I backed into it a bit: “Have you clicked on any Epstein-related links recently?” I asked. She had, she rejoined, wary of the implicit judgement hovering over my question. “But that doesn’t mean I want my entire feed to be about it!”

For whatever reason – and there are many, many possible reasons – the algorithms responsible for producing my wife’s feed had determined that the most likely content to perform *at that moment* were posts about Jeffrey Epstein and the longevity influencer Peter Attia. Did that please her? No. But was it explainable? I think so, and the conversation that ensued helped sharpen a hypothesis I’ve been considering for weeks: We’ve been living with at-scale versions of “generative AI” for a lot longer than we thought – and if we want to understand how generative AI might shape us going forward, it would pay to study the impacts its early forms have already had on our world.

You might wonder what I’m on about – and given I’m thinking out loud, it might help if we define a few terms. I asked Gemini for a short explanation of “generative AI.” Here’s what came back: “a type of artificial intelligence that creates new, original content—including text, images, code, music, and videos—by learning patterns from massive datasets.” Sounds about right.

I then prompted Gemini with this question: “how do feeds work on Instagram and TikTok – what drive the decisions the algorithms make?” Now, I’ve studied the answer to this question pretty closely over the past decade or so, and Gemini’s answer rang true to me: “Instagram and TikTok feeds use sophisticated machine learning algorithms to curate personalized content, aiming to maximize user engagement by analyzing thousands of signals, including watch time, likes, shares, and comments.”

If today’s generative AI delivers content by “learning patterns from massive datasets” and today’s social media feeds use AI to deliver content by “analyzing thousand of signals” to “curate personalized content,” well, it strikes me that social media feeds constitute something quite similar to generative AI, just delivered in a different product envelope. Instead of direct prompts, platforms like Insta, YouTube, and TikTok use our actions, our personal data, and thousands of other inputs to determine what we might see next on our feeds. In essence, the AI behind social media are generating our feeds on the fly, billions upon billions of times a day. It’s an insanely complicated (and rather out of control) process. And it’s no wonder that the companies behind those platforms – Meta, Google, ByteDance, et al – have come to dominate generative AI. It’s also no surprise that the newest entrant in the race – OpenAI – is trying to push its way into feed-driven social media. GenAI and social media are two sides of a very expensive coin, minted in a forge fueled by compute, cash, and at-scale data.

In short, social media – scaled, AI-driven content engines – catalyzed the revolution we now call generative AI. And what drives social media?

Advertising.

It’s no secret that advertising is coming to generative AI. OpenAI has already announced its plans, and Google quietly incorporated advertising in its “AI Overviews” over a year ago. Without the advertising industry’s massive revenue, AI providers will never be able to justify the hundreds of billions of dollars in investments they’ve already made in consumer-facing AI applications.

But a commitment to advertising comes a commitment to advertising’s imperatives – and we’ve seen exactly how those imperatives have played out through social media in the past decade or so. Will advertising impact future versions of generative AI in a similar fashion? That’s a question we should all be asking ourselves.  We may not like the answer, but there’s still time to imagine new models for how we engage with this new technology – and to demand more from the companies who provide it to us.

—

You can follow whatever I’m doing next by signing up for my site newsletter here. Thanks for reading.

It’s been fun to go back to Berkeley, where I first taught Journalism more than 20 years ago. I’m leading a seminar on how technology impacts journalism, with a particular focus on AI. The class asks students to read a bit of history – it’s hard to understand where we are if we don’t know how we got here. Search is a big part of that history, so I included a chapter of my first book – The Search – as a reading assignment.

As I prepared for class last week, I dug through my archives and unearthed The Search’s original manuscript. In the first chapter, “The Database of Intentions,” I opine on how search might lead to the development of AI that passes the Turing Test. Written 22 years ago, the passage anticipates the rise of generative AI. I start by drawing a distinction between data that is on our personal machines and data held in the cloud by large technology companies like Google. Then I think out loud a bit about where that all data might take us. Even though the writing is two decades old, it prompts some interesting questions about the moment in which we find ourselves.

—

“It is somewhat awkward to say this as the CEO of an AI company, but I think the next tier of risk is actually AI companies themselves…the governance of AI companies deserves a lot of scrutiny.”

So much to say about this. But read the essay. Then we’ll talk.