After four years covering AI as a daily beat — first at VentureBeat starting in April 2022, then at Fortune starting in March 2024 — I’ve decided to do the thing many journalists whisper about but few do: go independent.

What an incredible privilege it has been covering the society-shifting story of AI at this moment for Fortune, such an iconic brand. It has been an extraordinary platform to report on AI from the ground up — from C-suites to data centers to Capitol Hill — and along the way I even landed my first Fortune cover story! I'm deeply grateful to my editors and colleagues for the opportunity, the support, and the friendship. And while this chapter is ending, I hope to continue contributing to Fortune in the future.

But I've always been drawn to building things, starting with the family newspaper I launched at age 11 called The Gardenia Gazette (I grew up on Gardenia Lane), for which I was an editor, reporter, food critic and comic writer — and I got to boss around my little sister who served as an (unpaid) assistant editor.

Thanks to you for reading AI Side Notes

AI Side Notes has long fed that builder side of me, and I’m incredibly grateful to everyone who has subscribed and read along.

This little Substack experiment started out as Sharon’s AI Extra, and some of you have been here since those earliest posts in the fall of 2023. Along the way, I’ve shared reporting tidbits, analysis, observations from the AI beat, and occasionally some unfiltered thoughts about where all of this is headed.

Whether you’ve been here from the beginning or joined more recently, thank you!! The support and encouragement have meant more than you know—and helped give me the confidence to take this next step.

Doubling down on the AI beat and Substack

I’m still going to be an AI reporter, but I’m going to be doubling down on Substack and launching something new — right here.

On June 16 — five days from now — AI Side Notes will become something bigger, with a full rebrand, a new name, a sharper mission, and a podcast to go with it.

I won’t reveal much more than that until next week, but I can tell you this: I’ll be focused on original reporting about an AI era that is reshaping business, technology, policy, and society in real time.

As a thank-you for being early supporters, current AI Side Notes subscribers will receive a special launch offer in a separate email next week before the new publication officially launches. The publication will continue to offer a free subscription option, alongside additional benefits for paid members.

The door opens June 16. I hope you'll walk through it with me!

❤️
Sharon

This is my last post before I head out on vacation tomorrow (🥐🇫🇷🥖).

I had never heard of the Explorers Club. Turns out it is an international organization dedicated to the advancement of field exploration and scientific inquiry, which is also a fabulous private club on the Upper East Side of Manhattan located inside the landmark Lowell Thomas Building. Founded in 1904, it has counted the world’s most notable explorers as is members, including the first to the North Pole, first to the South Pole, first to the summit of Mount Everest, first to the deepest point in the ocean, and first to the surface of the moon.

This morning, journalists and podcast hosts Jennifer Strong and Jeff Wilson led an “AI and leadership salon” called The Drawing Room at the Explorers Club, and I was happy to be asked to join a panel of reporters to talk about all things AI, of course.

To be honest, I was feeling quite cranky and curmudgeonly about all things AI, even after two cups of coffee. It’s only Wednesday, and it’s already been a week chock-full of blowback, pushback and general gloom and doom about anything related to artificial intelligence.

“The American Rebellion Against AI is Gaining Steam,” a Wall Street Journal headline screamed yesterday. In leaked audio, Meta CEO Mark Zuckerberg was tone-deaf and heartless in leaked audio from an all-hands meeting in which he told employees that he was training AI on them ahead of mass layoffs (which happened today). Oh, and college grads are booing their commencement speakers, who don’t seem to get that only 18 percent of Gen Z-ers feel hopeful about AI, and almost half say the risks outweigh the benefits.

So today, when I looked at this poetic tribute to the Explorers Club written in 1921, hung in a room that included Teddy Roosevelt’s desk and two African elephant tusks), I felt a bit sad. Who would not admire explorers, who “proceed to seek the stars, through vast expanse, and revel in the cosmic dance?” I, for one, admire those that “scale the peaks where eagles soar, and sail the seas where waves all roar, confronting challenge, without fright, illuminating dark with light.”

I would like to think of the AI industry as these kinds of explorers — a “fellowship of seekers, who are bound by fervent flame.” I certainly know many AI researchers and engineers who qualify as explorers, venturing into systems that behave unpredictably, mapping capabilities no one fully understands, and building tools that may reshape science, medicine, language, robotics, and human knowledge.

But as OpenAI and Anthropic prepare to go public, it’s becoming harder to hold onto that romantic image of exploration. Big Tech and AI are increasingly entangled with Wall Street expectations, geopolitics, labor fears, and battles over massive infrastructure projects in small towns across the U.S.

My hope is that the kind of curiosity and humility that once defined generations of explorers can still survive amid all the money, power, and hype.

On that note, au revoir! The AI news will continue without me while I head off on a small expedition of my own across the pond. I’ll be back in a couple of weeks.

When I started covering AI full-time in 2022 for VentureBeat, one of the first things that struck me was the popularity of stories about a company called Nvidia. At the time, I didn’t even know how to pronounce the company’s name (I said NI-vidia instead of EN-vidia), much less understand why it seemed to sit at the center of so many highly technical conversations about AI.

Today, of course, Nvidia is the world’s most valuable public company and has become a proxy for the entire AI boom. As usual, its earnings announcement this Wednesday will be something of a cultural event, with all eyes on the Santa Clara-based behemoth as investors look for signs that the extraordinary AI spending boom can continue.

But in January 2023, I mostly wanted to answer a basic question: What exactly did Nvidia do? So I reached out to the company and said I wanted to write a deep-dive feature on how it came to dominate AI — and how it planned to capitalize on the generative AI boom that had erupted after OpenAI’s ChatGPT stunned the world in late 2022.

The core Nvidia story hasn’t changed

Looking back at that story now — published in February 2023, just three months before Nvidia’s stock began its history-making ascent — I’m struck by how much the underlying storyline remains the same.

The history, of course, has not changed. Nvidia’s dominance in AI emerged partly by accident. The company’s famed GPUs, or graphics processing units, were originally created in 1999 to power ultrafast 3D graphics in PC video games. But as the chips became optimized for more general computing tasks, they turned out to be remarkably well-suited for running deep learning algorithms.

The generative AI boom sparked by large language models and ChatGPT would likely not have happened at this scale without Nvidia. Today’s massive AI models require hundreds of thousands of GPUs to train and run — and Nvidia still controls roughly 90% of the discrete GPU market, even slightly more than it did in 2023.

Back then, people were already asking how long Nvidia could sustain its AI dominance. Would anyone eventually topple it from its perch? Not anytime soon, experts told me. They agreed Nvidia would not have free rein forever — especially with competitors like Advanced Micro Devices and Google developing rival chips, alongside startups like Cerebras Systems and SambaNova Systems.

CUDA is Nvidia’s defining moat

However, Nvidia’s platform strategy and software-first approach also remain remarkably difficult to dislodge. As CEO Jensen Huang likes to emphasize, Nvidia is not simply a chip company. “We stopped thinking of ourselves as a chip company long ago,” Huang said during Nvidia’s 2025 annual shareholder meeting.

Nvidia’s ecosystem extends far beyond chips themselves, encompassing hardware, software, networking, and development tools all optimized to work together. The company’s CUDA compute platform, introduced in 2007, is the software and middleware layer that allows researchers to tap into the immense parallel processing power GPUs can provide. Experts both inside and outside Nvidia emphasized to me that the deep learning revolution likely would not have unfolded the same way without CUDA.

Today, CUDA remains Nvidia’s defining moat. Just last week, WIRED published a story about the platform Huang has described as Nvidia’s greatest “treasure.”

Geopolitical tensions loom larger

And just like today, geopolitical tensions already loomed over the company. Under the Biden administration, the U.S. had imposed export controls preventing state-of-the-art Nvidia chips like the A100 and H100 from being sold to China.

But one thing that also hasn’t changed is Jensen Huang’s determination not to give up on the Chinese market (not to mention his love of leather jackets and Asian street food). Last week, Huang joined the delegation of American business leaders traveling with President Donald Trump to Beijing, a sign of how aggressively Nvidia continues to pursue business in the country despite years of export controls and political uncertainty. Even after the Trump administration approved Nvidia’s H200 AI chip for sale to China late last year, Beijing has yet to approve purchases.

My VentureBeat story from February 2023 ended by saying that “Nvidia is still in an AI sweet spot.” “Just as the company’s GPU was at the center of powering the deep learning revolution of a decade ago,” I wrote, “Nvidia’s hardware and software are running behind the scenes of today’s GPU-hungry, hyped-up generative AI technology.”

Three years later, that remains true. In many ways, Nvidia’s position looks even stronger. But every move the company makes now also carries enormous geopolitical weight. Today, AI is infrastructure, economics, and geopolitics all intertwined — and Nvidia sits at the center of all three.

Last week, Cerebras — once one of several ambitious AI chip startups trying to challenge Nvidia — pulled off one of the most closely watched IPOs in the AI market. But even now, Nvidia’s moat looks stubbornly difficult to crack.

Nvidia has remained remarkably consistent

So much has changed since I started covering AI in 2022. OpenAI is nearly unrecognizable from the research lab I first encountered when the DALL-E 2 image model was released during my first week on the beat. Anthropic was still a scrappy startup fueled in part by money from the now-disgraced Sam Bankman-Fried. Google had not yet fully realized it was falling behind in the AI race.

Nvidia, however, has remained remarkably consistent in its focus on the infrastructure layer fueling the AI boom — much like Jensen Huang’s ever-present leather jacket.

The above photo from my trip to Saline Township, Michigan, in March tells you a lot about this agricultural community 30 minutes south of Ann Arbor: classic red barns, dirt roads, and a farming tradition dating back generations.

When I returned from the trip, I teased the story of how a massive AI data center project landed in this rural corner of Michigan. But it took far more reporting — and many more interviews — to fully understand the complicated legal and zoning dynamics that allowed the project (now to be a $16 billion campus for OpenAI and Oracle,) to move forward, even though the vast majority of residents opposed it and local officials voted against it.

Yesterday, I finally published the story in Fortune (gift link), and I hope you’ll take some time this weekend to give it a read. I admit it’s a long one, but I also really think it’s an engaging and thought-provoking saga of what happens when the AI boom collides with rural politics, legal pressure, and fears that local communities are losing control over what gets built in their backyard.

That said, I didn’t think a deep-dive story into what is essentially a zoning fight in a tiny Michigan town would take off online. But to my surprise, when it was published yesterday it immediately hit Google Search and Google Discover like lightning, quickly racking up nearly 300,000 views at last count.

If you recall, I had already published two previous stories on AI data centers and rural communities — one on the chaos at Meta’s mega-site in northeast Louisiana, and one on a proposed project deep in the desert near Phoenix, Arizona. Neither of them got that kind of massive online traffic out of the gate, or at all.

It makes me wonder if something has changed in the last few months since I visited Arizona, or even in the past three months since I was in Louisiana. I'm becoming this topic has entered a peak moment, as backlash to AI data centers, and, frankly, AI in general, continues to heat up. In addition, there was something about this particular story (Midwest, farm, legal issues, power struggle) that struck a chord. Certainly, this trend is part of the public consciousness now.

I do sometimes get the sense that people reading these stories assume I’m against AI data centers. I’m not. Data centers have powered our online lives for decades, and AI infrastructure is poised to expand that dramatically in the years ahead. Reporting these stories, I’ve met hard-working and thoughtful people at development firms, construction companies, utilities, and Big Tech companies who genuinely believe they are building something important. And on a sheer engineering level, these facilities can be astonishing to see up close.

But while the strategy in Saline Township may have worked — at least for now — I also came away believing that pushing through projects over overwhelming local opposition is probably not a sustainable answer to the computing demands of the AI industry. And visiting these communities in person made clear just how complicated, emotional, and consequential these fights are becoming across the country.

As Kathryn Haushalter, one of the residents opposed to the Saline Township project under construction, told me, “It’s a very complicated situation and a lot to take in,” she said. “I’ve had to learn more about ordinances and state law and zoning than I ever thought I would want to. But now I realize how important these nitty-gritty, seemingly boring things really are. They can upend your whole community.”

When I showed up at my freshman dorm at George Washington University, I had something almost no one else did: a personal computer and printer (I’ll let you do the math to figure out how long ago it was).

Luckily, my parents had pushed me into typing class as a kid, where I’d built up to a speedy 80–85 words per minute. So while everyone else was hand-writing papers or hunting and pecking on electric typewriters, I saw an opening to earn some cold hard cash.

I started typing papers for other students, charging per page. It was my first job in college — and, for a while, a very good one.

Over the next year or two, however, nearly every student acquired their own desktop. My little business went bust.

That wouldn’t be the last time — my personal boom-bust job cycle was just getting started. For decades, new technologies kept showing up, upending whatever I was doing — both helping and hurting — and always forcing me to figure out what came next.

There were my college-era receptionist jobs, where I sat at a desk answering phones for executives. Those gigs disappeared pretty quickly once voicemail and phone trees came into vogue. My post-graduation, entry-level jobs in PR and advertising required IRL things like photocopying, sending faxes, and shepherding file folders from office to office. None of those job “skills” lasted once email came along.

My typing skills hung on into the mid-90s: I once had a weird part-time gig as a transcriptionist, where I sat in a small conference room with headphones, a tape deck, and a foot pedal to control playback, typing interviews from television news shows like 60 Minutes and Dateline as fast as I could. I eventually got carpal tunnel syndrome and had to give it up — but it was just as well. Human transcribers hung around for years after that, but today I use Otter AI for my own interview transcriptions.

When I got into journalism, desktop publishing was all the rage — I’d just missed the era of cutting and pasting galleys onto layout boards, and had taken a class in Quark XPress, in which I spent much time making silly newsletters for my family. Print magazines and newspapers were still king, so we printed out layouts, spent hours copy editing and fact-checking, held headline-writing sessions for magazine covers, and stayed up late — once a week for a weekly, once a month for a monthly — when the issue finally closed and went to the printer.

As someone who had always dreamed of being published in The New Yorker (still dreaming), I couldn’t imagine any of that changing. But, of course, it did. The internet came along and began pulling everything — readers, ads, attention — online. Yes, I had a Pets.com puppet. Yes, I ordered a single pint of ice cream from Kozmo.com. Oh yes, HTML tags became a way of life.

At that point, print magazines were still part of the publishing mix, but the internet was where the money was. And the work itself was different — faster, constant, driven by clicks and search traffic. Stories were meant for page views, headlines were optimized for Google, going “viral” became a thing. Many people I knew in the business hated it. I found it kind of thrilling.

When I eventually took another staff job, the weekly print magazine I joined had already shrunk to bimonthly and focused mostly on its online business. I could see where things were headed, and I headed for the door once more.

I got deeply into things that had never existed before, like content marketing and blogging and listicles, riding the wave of digital and social, then eventually circled back to journalism.

I had been writing about technology for years — it was one of the best-paying subject areas out there, and if the most secure jobs were in technology, particularly after the pandemic, surely covering the tech beat was a safe place to be.

So when I started covering AI in 2022, I felt like I had won the jackpot. What could possibly go wrong? And when ChatGPT came out that November, and I was covering the biggest story on the planet, I felt unstoppable. This, I was sure, was where I was meant to be.

Yet by mid-2023, doubts were creeping in. The very beat I was covering every day was the one that could send me back to job-hunting. AI was going to write, edit, and publish articles, forcing me into my always-on-standby backup plan — being a dog walker. (I have never actually been a dog walker, but I love dogs and I love walking, so I’ve always considered it a well-laid contingency plan.)

The same beat was supposedly killing media in general, too. Those print magazines that had valiantly pivoted online, figured out SEO, cracked social media — their business models were suddenly upended all over again, thanks to AI Overviews and chatbots.

These days, the first question most people ask me about AI is: “Will it take my job?” Then they often wink at me and, with a nudge, suggest that my industry in particular is ripe for the tanking.

There are headlines about AI causing a “permanent underclass.” I report regularly on economic studies and entry-level job fears, and I often test myself on other non-dog-walking jobs I could do if I were laid off due to AI.

But I’ll be honest: I actually don’t buy it. The latest tech has made many parts of my job obsolete for decades, and I’m sure it’s going to do that for millions of people many industries. But I’ve also been using the latest tech all along to get to the next thing. And in my experience, there is always a next thing.

These days, I use AI for a boatload of tasks in my daily work. All the major tools — like ChatGPT, Claude, and Gemini — help me with everything from editing and research to images and video. But none of it does my work for me. In fact, I have more work to do than ever.

I’m not naive about it, though. I do believe that new types of jobs will develop that we can’t even imagine now, but I also know that I’m not predicting the future. After all my shifts and pivots, I know that no job — no matter how well-paying, how seemingly necessary, how high-status or fundamental — is guaranteed. And I know that not everyone is as lucky as I am. I’m several decades into my career. I’m on my husband’s health insurance. I don’t have kids depending on me. I’ve had time to build a network that took years to cultivate.

But every wave of technology that was supposed to end me just pushed me somewhere more interesting. I have no reason to think this one will be any different. Tech has been killing my jobs for decades. I’m still here.

I started covering AI the same week that OpenAI’s DALL-E 2 image model was released on April 6, 2022.

It was considered so groundbreaking, such a massive step up from the original DALL-E released a year earlier, that Cade Metz wrote about it in the New York Times, calling it the “the AI That Draws Anything at Your Command,” with an image of the avocado teapots that DALL-E 2 could output that were all the rage.

But it didn’t take long to realize that there were massive limitations to AI image models. Hands had too many fingers, text was garbled, and charts were impossible.

Subscribe now

Another leap forward

Over the past four years, that’s changed dramatically. OpenAI’s new Images-2 model, released last week, feels like another leap forward.

The day it came out last week, I was flying back from Washington DC. With a (fabulous) Starlink connection, I could play around with it on my short flight. The images it produced were striking: polished, detailed, with text that was suddenly surprisingly coherent. Like many people trying it out for the first time, I came away impressed.

Sam Altman and me:

Me in a Peanuts-style comic talking about AI:

But after sharing my images with my sister, who is an anatomy professor at a medical school, she suggested a tougher test.

We had a funny exchange that I shared on X:

The late-night exchange, and the image (which you can view at the beginning of this post) went viral, drawing more than 180,000 views. It seemed to hit a nerve, with many people reacting to how something that looked so polished could have so many errors — that weren’t, to be clear, immediately obvious to a layperson like me giving it a quick glance.

A benchmark of AI image accuracy

I asked my sister why she chose the thorax image as a kind of benchmark of AI image accuracy. She said that a couple of years ago, at meetings for the American Association for Anatomy, one of the speakers was talking about the potential for AI in education, and used an AI-generated image of the thorax as an example of the random placement of structures that would happen (in that case, she recalled, the image had two aortas and other blatantly made-up structures). Now, she said, she likes to try it to see how things are coming along.

And things are, clearly, coming along. In a blog post, OpenAI said that the new model offers “more precise image generation for complex creative tasks like small text, UI elements, diagrams, and dense layouts.” But it’s still not enough to satisfy anatomy experts. The question is, why is that?

The issue isn’t detail—it’s structure. Image models can generate something that looks anatomically correct, but they don’t reliably capture how parts relate to each other in space. My sister pointed out that while the latest thorax image is better than it used to be—there aren’t two aortas, after all—it’s still surprising how far off it is given how many accurate examples exist.

The problem is that image models learn statistical patterns from images rather than an explicit understanding of underlying structure. They’ve seen thousands of anatomy diagrams, but those vary in accuracy, and the model doesn’t have a reliable internal model of how structures relate in 3D. In effect, it’s drawing on patterns it has seen before rather than reasoning from a consistent anatomical framework.

Labeled diagrams are especially difficult. Text, arrows, and image content all have to align correctly—a different challenge than generating a plausible-looking image. Without reliably capturing those relationships, the result is errors like misplaced vessels, incorrect labeling, and anatomically incorrect structures.

Almost right, but not quite

That circles right back to my sister, whose immediate ‘Oy’ wasn’t just about what was wrong with the image, but how the image could be used.

“What concerns me most is as it gets closer to correct, it will become more likely to be used,” she said. “Then it will be harder to know what to trust since things will be ‘almost right’ but not quite.”

When it comes to anatomy, ‘almost right’ will never be right enough. Even if it is still cute, tho.

Since the attempted firebombing on OpenAI CEO Sam Altman’s San Francisco home on April 10, AI backlash has been impossible to miss. You know it’s reached a moment when people are even mad at the hard-to-be-mad-at Reese Witherspoon for supporting some use of AI.

But as any terminally online AI industry watcher knows (myself included), anti-AI groups have been part of the conversation for years. I’ve covered the rising use of the word “doomer” since mid-2023, when I wrote this story for VentureBeat about AI experts challenging the ‘doomer’ narrative, including claims of a risk of human extinction from advanced AI if its development is not properly managed.

In that piece, many told me this 'doomsday' take, with its focus on existential risk from AI, or x-risk, is happening to the detriment of a necessary focus on current, measurable AI risks — including bias, misinformation, high-risk applications and cybersecurity. The truth is, most AI researchers are not focused on or highly-concerned about x-risk, they emphasized.

In a separate Q&A for VentureBeat in June 2023, top AI researcher Kyunghun Cho put it more bluntly, saying “I’m disappointed by a lot of this discussion about existential risk; now they even call it literal “extinction.” It’s sucking the air out of the room.”

Of course, these days, whether they are “doomers” or not, people are concerned about every kind of AI risk — from advanced cyber attack capabilities and deepfakes to predictions of massive job loss and the effect of mega AI data center construction on local communities. And while social media was filled with concern for Altman’s safety after the attempted firebombing allegedly carried out by 20-year-old Daniel Moreno-Gama, the action has drawn attention to two anti-AI groups with similar names: Pause AI and Stop AI. I had heard of both, but I dug into their histories and tactics for a Fortune story this week.

Both organizations have condemned the violence and said the suspect is not and never was a member of their organizations. Still, the incident, in which Moreno-Gama also went to OpenAI’s headquarters and tried to shatter the building’s glass doors with a chair and threatened to burn the facility, surfaced his activity on Pause AI’s Discord server and renewed scrutiny of Stop AI’s direct actions targeting OpenAI last year.

I hope you’ll consider reading the whole thing: This is a gift link. But here are a few details that stood out:

• The alleged attacker had posted 34 messages in Pause AI’s public Discord—including one saying, “we are close to midnight, it’s time to actually act”—but the group says none explicitly called for violence.

• Pause AI says it actively vets volunteers and tightly controls messaging to avoid being associated with extreme views.

• The group’s name traces back to the Future of Life Institute’s 2023 open letter calling for a pause on advanced AI—and that organization is now its largest funder.

• A key internal fracture: Stop AI was formed by people pushed out of Pause AI over disagreements about tactics—especially around civil disobedience and breaking the law.

• Stop AI’s activism has included serving a subpoena to Sam Altman mid-event and staging protests directly targeting AI labs.

• One Stop AI cofounder allegedly threatened violence, disappeared after an internal conflict, and is still missing.

• Another cofounder, Guido Reichstadter, has a history of dramatic protest tactics—including a hunger strike outside Anthropic and chaining himself to infrastructure.

• Stop AI says the suspect once joined its forum and explicitly asked if advocating violence would get him banned—and stopped engaging after being told yes.

For months, I’ve wanted — no, needed — to try Anthropic’s Claude Code. How does it look if Fortune’s AI reporter couldn’t wrap her mind around the tools everyone is raving about?

“Everyone should be using Claude Code more,” wrote Lenny Rachitsky in a post featuring “50 ways non-technical people are using Claude Code in their work and life.” And hundreds of influencers across social media raved about the power of “vibe coding,” in which a beginner like me could clearly create apps, games and tools by just chatting with Claude Code, Codex, Cursor or Google AI Studio.

Even my Substack journalism peers, like Big Technology’s , made me feel FOMO, posting that he was “building some Big Technology internal tools with Claude Code.”

Developers I asked were encouraging, even reassuring. It’ll take just a few minutes to get started, they told me. One pulled up their terminal and showed me how to download Claude Code. Another pointed out that the new desktop app made getting started more seamless than ever.

That’s sweet. But the truth is that Claude Code is nowhere near ready for normies like me — desktop app or not. The developers (or even the power-user hobbyists) hyping these tools aren’t wrong, exactly. Claude Code and its ilk are genuinely changing their lives. But they’re also the ones driving the narrative, and their experience is not everyone’s experience. What transforms a developer’s workflow in minutes can leave the rest of us completely lost. Hype, unfortunately, does not mean “accessible.”

I did not get farther than downloading Claude Code on my own. Shortly after that, I was stuck. My software developer husband had to explain everything: how to install Node.js, the runtime Claude Code requires; what a directory is; how to set up a project folder; how to read error messages; and why I couldn’t get my simple app to work — a reminder app to nudge me to move every 30 minutes, with encouragement from a dancing teddy bear. Even after thousands of tokens and many minutes of ruminating, Claude Code seemed stuck. And even my developer husband was temporarily stumped.

Turns out to get my app going I just needed to copy the preview URL into a browser. Claude Code hadn’t said a word about that — or much else that was helpful along the way. And once my teddy bear started dancing, with only one attached arm and the other dangling, it was clear that my little app was not ready for prime time.

My husband grumbled about it for the rest of the evening, insisting that the Claude Code desktop experience was clearly marketed for non-developers, but that it fell well short of that promise. To really target the average consumer, a setup flow could not assume terminal familiarity. It would need plain-language explanations of every dependency. It would require proactive prompts that tell you what to do next. And, of course, clear explanations about why an app isn’t working and what to do about it.

Sure, I could take a coding class. But isn’t that the opposite of what everyone in AI is promising — that coding will soon be unnecessary, that software engineers are headed for extinction? And isn't democratizing AI supposed to be the whole point?

We’ll need developers for a long time to come — no doubt. The bar isn’t a working teddy bear whose arm isn’t dangling by an imaginary thread. It’s being able to get started without a developer sitting next to you. When that day comes, I'll be first in line, teddy bear and all. But for now, I’ll stick to my trusty chatbots.

Hello from the Grand Hyatt SFO, where I am staying overnight after flying in from Newark in preparation for AI conference overload — in this case, a massive Enterprise AI extravaganza called HumanX.

I’ve got a fantastic runway view from my room, which would please my Dad, who is an avid airline memorabilia collector. Personally, I’m mostly interested in the Fiji Airways plane parked outside my room. I’m tempted to book a flight to the South Pacific rather than spend the next few days at the Moscone Center, but duty calls.

I’ll be moderating a few panels, tackling a bunch of meetings, hitting a boatload of happy hours and dinners, and maybe, just maybe, having a moment to attend a session or chat with folks totally randomly? We shall see — especially since the news cycle never ceases for a Fortune AI reporter!

My flight was a typical coast-to-coast jump. Naturally, my row-mate is an AI founder building in stealth. Yet, as usual the Starlink wifi barely worked. Of course I am almost ready to hit the hay, and it’s only 7 pm PT. Yawn.

Tomorrow, I’ll prepare for three fireside chats on the main stage. I’m excited to speak with Andrew Ng and Greg Hart (DeepLearning.AI/Coursera), Michael Truell (Cursor) and Francis deSouza (Google Cloud).

Other than that, I assume I’ll be dodging the hundreds of PR folks I did not respond to in advance of HumanX. I’m not sure I have ever been bombarded with reach-outs like I have here. It was total overkill and mostly felt like AI-written spam. I had to put up a post on LinkedIn begging for it to stop.

That said, I am looking forward to the part that actually matters — real, in-person conversations with people I choose to talk to. It’s been a while since I’ve had that kind of density of IRL reporting, and I’m hoping to come away with some real signal (and not just noise) to share later this week.

How AI is affecting journalism was big news this week. There was the Wired story about how tech reporters on Substack are using AI to help write and edit their stories. There was the Wall Street Journal story about how one Fortune staffer is churning out a half-dozen stories a day prompted by AI (as you know, I work for Fortune).

Meanwhile, my 4000-word tome on how Meta’s $27 billion new AI data center in northeast Louisiana is reshaping the region published this week. Bad timing for a deeply-researched, on-the-ground reported feature?

I say not.

The discourse, not surprisingly, has tilted towards a familiar argument: that AI is hollowing out journalism, that attention spans are collapsing, that the era of long-form, on-the-ground reporting is sputtering to an end, especially as tech reporters are struggling to keep up with the sheer pace of news.

Let’s be clear: I use AI in my work. All day, every day. AI transcribes my interviews, helps write my headlines, tightens my sentences, digs into research, translates technical jargon, offers feedback on hot takes, provides structural notes, and helps break up run-on sentences like this one, that threaten to race off into nonsensical oblivion and risk turning off even the most dedicated readers (maybe I should have had ChatGPT fix this example, but then it wouldn’t be an example, right?)

But I write my own stories, and some of them are damn long. There are those who say GenZ doesn’t read, that they get their news from short-form video, that if you’re struggling to pump-out long-form, original reporting, you’re not innovating. In the age of AI, even if isn’t slop, it needs to be short, sweet, and easy to digest.

I say that’s hogwash. I say that as respectfully as one can say “hogwash.” I say that as someone who has run the gamut of journalism/content over two decades. From deep dives and analysis to quick news hits, aggregation, humor pieces, trade mag write-ups, consumer mag features, blogs, white papers, sponsored brand content and oh, that stint writing tabloid celebrity write-arounds for InTouch Weekly where I sat on a train to Philadelphia with my cell phone tucked into the crook of my neck, desperately taking notes from a plastic surgeon about what he thought about Pamela Anderson’s side-boob.

There will “always” be room for long-form, original reporting, that journalists struggle to write. Reports of its death are greatly exaggerated. In fact, it might even be the future. Sure, it won’t be in print newspapers or magazines. Funding sources might differ, its focus might be more niche, the number of authors may shrink.

But just like Timothee Chalamet’s comment doesn’t mean ballet or opera are going anywhere, there will always be an audience for writing that emerges from original reporting on the ground. The value of being in-person is only increasing. And short-form video is not enough to get the full story.

This brings me back to Richland Parish, Louisiana, where the story of Meta’s mega AI data center is complex and messy and, in many ways, can only be experienced on the ground.

There, I could witness the hundreds of trucks hurtling in and out of the 4 million square foot Meta site, kicking up rocks that damage residents’ windshields. I could meet the local businesses that are thriving as a result of the project, as well as the ones that are struggling to get in on the action. I could meet the residents that are thrilled to see their relatively poor parish get Big Tech money; as well as those who are being priced out and even evicted to make room for thousands of out-of-state construction workers. I could drive by massive new “man camps,” RV parks meant to house those workers.

This isn’t the kind of story that emerges from a model output, even though AI helped me at many points along the journey to publication. No, it comes from being there and spending time and energy to report a complex, nuanced story. No churning out anything in this case.

I’m lucky that I had Fortune’s support to fly to Louisiana, rent a car and stay in a hotel. And, of course, I’m lucky that they considered the topic to be worthy of a deep dive.

But I also believe people will read this kind of story. No, not everyone, and not all the way through (though I highly recommend it!). But enough.

The real question is whether journalists will be allowed to keep doing original, in-depth reporting, whether it be deep dives like my Meta story or Big Tech scoops or long-form commentary and analysis. I say they should, or should do it independently — as so many here on Substack do. I am a paid subscriber of many, including , , , , and .

In my AI beat, I think it’s essential, because the more AI reshapes the physical world and the people and companies living in it, the more we need journalism that goes out, sees it up close, and writes about it.

People will read 4000 words — or at least try — when it’s the kind of reporting they can’t get anywhere else. Books didn’t disappear, and neither will deeply-reported stories that have some real non-AI effort attached to them.

At least, that’s the bet I’m making right now: Read the full Meta story here.

I’ve been covering AI as my daily beat for four years — first at VentureBeat, and now at Fortune. For most of us covering the space (shout-out to y’all), AI beat coverage has orbited around a small number of companies building frontier models, including OpenAI, Google, Anthropic and Meta.

The drama has been undeniable: The model launches! The product features! The benchmarks! The leadership soap opera! The talent wars!

Of course, I have always made it a point to publish about other issues, including AI policy and regulation, the chips powering the AI boom, and how top companies are adopting AI. Still, it always seems like the latest tit-for-tat between Sam Altman and Dario Amodei has sucked up the oxygen. Every investment round, every embargo, every hyped-up beef on X has been at the center of the 24/7 news cycle.

Part of the reason, I think, is because this style of reporting presents a simple narrative of believing, building and battling for attention and growth. But I think the complex messiness of AI is becoming harder to ignore as AI gets out into the wild, into the real world. The daily AI news cycle is drama for Silicon Valley. But the story has expanded far beyond the Bay Area—into the workplace, infrastructure, supply chains, communities, politics, security, and people’s daily lives.

None of this is to say that the frontier model companies don’t matter—they absolutely do. They’re still setting the pace of the technology and shaping what’s possible. But the story is getting far more complicated and consequential.

That’s where my reporter ears really perk up. Once AI is out in the wild, we’re not just talking about model capabilities, or product features, or even what a frontier AI billionaire thinks. We’re talking about control, risk, cost, consequences.

“Why'd ya have to go and make things so complicated?” sang Avril Lavigne back in the early-aughts. I say bring it on, because once you start looking at AI through that lens—control, risk, cost, consequences—you start to see the same story showing up everywhere.

In the AI data center boom and its impact on local communities. In the new challenges facing cybersecurity teams as AI-powered attacks ramp up. In how companies are navigating backlash, how elections could be shaped, and the explosion of products trying to operationalize AI across every industry.

That story is what happens as AI moves from contained systems—labs, models, companies—into complex, real-world systems, and all the consequences that follow. And none of it will play out exactly as even the best experts expect.

I sometimes think about this when I’m sitting forlornly on a New Jersey Transit train, trying to get from my central New Jersey home to Fortune’s offices in downtown Manhattan.

It’s often a total shit-show: decades-old trains, aging tracks and wiring—and rail tunnels under the Hudson River that are more than a century old, were seriously damaged during Hurricane Sandy, and still need major, long-term repairs.

Everyone knows they have to be fixed. But they can’t just be taken offline, so a new tunnel has to be built first—and that process is expensive, slow, and politically fraught, with years of funding fights and delays. And meanwhile, I sit with hundreds of fellow commuters, frustrated, staring at the marshes of the Meadowlands, wondering how my life turned into this when Japanese commuters have enjoyed bullet trains since 1964.

Yup. It’s never just about the technology.

Even with all the news dropping in AI this week, my Fortune editors were kind enough to let me go on an extended reporting trip for the series on the mega AI data center boom I’ve been working on (my article on Arizona ran in December and my feature about Meta’s Hyperion site in northeast Louisiana will publish in a couple of weeks). These next two stories should publish in the next month or two.

Subscribe now

Saline Township, Michigan: A $7B AI data center collides with rural farmland

First, I headed to Saline Township, Michigan, which is a centuries-old farming community outside of Ann Arbor. A massive $7 billion AI data center backed by developers tied to Related Companies founder Stephen Ross and partners including OpenAI and Oracle, was initially rejected by the township board after residents raised concerns about power demand, water use and the loss of agricultural land. But the developers and landowners sued, arguing the township’s zoning effectively blocked data centers altogether. Facing a costly legal battle, local officials quickly settled the lawsuit and agreed to allow the project to move forward under a consent agreement that includes farmland preservation funding and other concessions—an outcome that has left many residents feeling the town was forced to accept a hyperscale AI facility it didn’t want.

Now, there are 300 trucks hauling dirt to and from the over 500-acre site three times a day as construction heats up.

I visited one of the project’s most vocal critics, Kathryn Haushalter, a former US marine and mother of five, who lives on a 60-acre property across the road from the data center. Haushalter filed a motion to join the lawsuit tied to the data center project, but a judge ultimately denied her request to intervene in the case. She is particularly concerned with the environmental effects the data center could have on her well water, the farming soil and the health of her children.

I was amazed at the classic red barns all over Saline Township:

And you can already see the work happening on the site:

I also attended a Town Hall meeting where council members heard from a local consultant, Barry Lonik, who specializes in conservation easements, which are one of the main tools farmers can use to protect agricultural land as development pressure rises. It doesn’t help with the data center already underway, but could help the community prevent more projects coming to the area.

Ross, Texas: When a neighboring city controls the data center vote

Next, I flew to Austin, Texas and then drove north to visit a proposed data center site near Ross, Texas, outside Waco, which has become a local flashpoint because the project’s fate is largely being decided not by Ross itself but by the nearby city of Lacy Lakeview, whose extraterritorial jurisdiction (ETJ) covers the rural land where the facility would be built.

Developers are seeking annexation and zoning approvals from Lacy Lakeview for what residents say could become yet another massive AI data center campus on agricultural land, raising concerns among nearby farmers about noise, power infrastructure, water use and the long-term industrialization of the area.

The situation has created tension in McLennan County, with some Ross-area landowners arguing that a neighboring city effectively holds the power to approve a project that could reshape their rural community—even as officials and economic development advocates emphasize the potential for jobs and long-term investment tied to the AI infrastructure boom.

There are signs from the opposition all over Ross:

At the moment, the proposed site is beautiful grazing and farmland that has been sold to a developer:

I visited one of the members of the opposition community, who introduced me to her family’s beautiful cows. Her land is directly adjacent to the data center site.

I have more meetings today in Lacy Lakeview and Ross, and then tomorrow it’s back to Austin for something totally different: I’ll be moderating a fascinating panel at the SXSW conference on how AI is learning to code biology.

Meanwhile, I’m enjoying the fabulous early-spring weather here in Texas. Have a wonderful weekend!

This time I reached out to of , who I have followed for the past year as I have started reporting more on all things AI data center (and turns out to be good friends with my next-door neighbor! Small world!).

Rich is, yes, a “rich” well of information on this topic. He has over 25 years of experience reporting about data centers, so the recent craze for mega AI data centers is not his first rodeo — though he agreed that the scale of what’s going on right now is unprecedented. He has a great measured take on the space: He agreed with me that the data center and Big Tech companies have a lot of work to do in dealing with local communities with transparency, but he also emphasized that data centers have become essential infrastructure — and there is no getting around the fact that more are needed in more places.

Let me know what you think! And as always, I’m always open to ideas on who to invite to another AI Side Chat!

Last week, Anthropic gave me an exclusive opportunity to speak with Logan Graham, the leader of the company’s Frontier Red Team —a group of researchers tasked with stress-testing Anthropic’s most advanced AI systems and exploring how they could be misused—about a new product the company was about to launch based on the team’s findings, called Claude Code Security.

I’d met Graham and his colleagues months earlier at Black Hat and DEF CON—an annual one-two punch of conferences focused on cybersecurity and hacking—where Anthropic had been sharing security research on the capabilities of its AI model, Claude. By the time that work turned into a product, it felt like a natural progression, one I already understood. It made sense that I’d be the one to write about it.

Subscribe now

In my story, I focused on what Claude Code Security actually does: an AI system designed to hunt software bugs on its own, including the most dangerous ones humans often miss. What I didn’t anticipate was that the real headline would come after the announcement. Investors were spooked by what the tool seemed to imply—that traditional security software might be less necessary—and cybersecurity stocks sank, including incumbents like CrowdStrike and Zscaler.

I don’t think the investor panic was warranted. But the episode was another reminder that covering AI as a daily beat is strange. It’s a constant rush of product launches, hot takes, marathon podcasts, influencer threads, and deeply technical arguments about what matters and what doesn’t. And no matter how closely I follow the research—or how many people I talk to—the list of things I don’t see coming keeps getting longer.

I’ve written about this feeling before: the day ChatGPT landed in my inbox and felt, at first, unremarkable; the moment Matt Shumer reached out ahead of publishing his viral essay, “Something Big Is Happening”; the shock of reading that Sam Altman had been fired from OpenAI just after I’d written about the company’s nonprofit board. Each time, I thought I understood the story—until it suddenly became something else.

I know I’m not alone in this—if you’re constantly reading about AI or working in the industry, I’m sure it happens to you, too. And I’m not claiming to be any kind of fortune teller when it comes to AI (even though I work at Fortune, ha). But as an AI reporter constantly trying to break news, analyze trends, and dig deep, it’s disconcerting to realize how often the story you think you’re covering isn’t the one that ends up taking off. 🤷‍♀️

Can you really run your AI agents while you get your beauty sleep? Or is it more like a Mom trying to keep her toddler from running amok?

Finding that out became my goal over the past week, after my Fortune editors asked what actually happens when AI agent tools like the viral OpenClaw—or whatever OpenAI builds now that it has hired OpenClaw’s creator, Peter Steinberger—are allowed to run 24/7.

In my new Fortune story published today (here is a gift link!) I have the answer, at least for the moment: Toddler. Or, best-case scenario, junior intern.

The problem is while tools like OpenClaw and Claude Code make it technically possible for agents to run for long periods, in practice these tools are typically fragile, unpredictable, and labor-intensive to manage. Rather than replacing human work, today’s agents often require constant monitoring, guardrails, and intervention, especially when the stakes rise beyond low-risk experiments.

That’s why if you’re on X a lot, you might have come across a cautionary tale from Summer Yue, who works on safety and alignment on Meta’s superintelligence team.

In a post on X Monday, Yue described how her OpenClaw autonomous AI agents—built to run locally on a Mac mini computer—deleted her entire inbox, ignoring instructions to pause and ask for confirmation first.

“I had to RUN to my Mac Mini like I was defusing a bomb,” she said. It was, she added, a “rookie mistake.” The workflow had been working in a test inbox she used to safely trial the agent for weeks, she explained, but in the real inbox the agent lost her original instruction.

I spoke to a fantastic array of experts for my Fortune piece, including Shyamal Anadkat, who previously worked as an applied AI engineer at OpenAI; Yoav Shoham, a former principal scientist at Google, a professor emeritus at Stanford and co-founder of AI21 Labs; Bret Greenstein, chief AI officer at consulting firm West Monroe; and Aaron Levie, CEO of cloud-based content management and collaboration company Box.

The bottom line: For now, working with AI agents may have less to do with sleeping while they work than with staying half-awake while they do. Tools like OpenClaw can run for hours at a time, but for many early users, that autonomy comes with a new kind of vigilance—checking logs, reviewing outputs, and stepping in before things go wrong.

That dynamic was captured in a recent viral post titled Token Anxiety, in which investor Nikunj Kothari described a friend leaving a party early—not because he was tired, but because he wanted to get back to his agents. “Nobody questions it anymore,” Kothari wrote. “Half the room is thinking the same thing. The other half are probably checking the progress of their agents. At a party.”

Do you have any stories of AI agents gone rogue? Or how OpenClaw is your latest obsession? Are there use cases you love so far?

On Monday, AI influencer and CEO Matt Shumer reached out to say he was writing an article to explain to the average person “what’s going on in AI and how it’s going to affect them.” He asked whether Fortune runs opinion pieces, and I passed him along to the appropriate editors.

He also asked if I wanted to read a draft. I declined—and I’m glad I waited to read it until “Something Big Is Happening” had already racked up nearly a million views on X. (Full disclosure: Fortune did end up running a shorter version of the piece.)

In hindsight, I’m glad I didn’t read it before publication, because I probably would have pushed back on the breathless tone (“bigger than Covid”), the hyped-up framing (“The people I care about deserve to hear what is coming, even if it sounds crazy”), and the sweeping claims built on shaky assumptions (“We’re telling you what already occurred in our own jobs, and warning you that you’re next”). As a longtime editor, I also might have suggested some tightening. The piece runs more than 5,000 words.

Of course, then it might not have gotten even close to those 75 million views.

As it happens, I’m glad my colleague Jeremy Kahn took the time to weigh in with a critique of the essay today, focusing on Shumer’s assertion that what’s happened in coding is a prequel for what will happen in other fields—and that this shift will unfold within just a few years. Kahn sees that timeline as wrong, even though he himself wrote a book—Mastering AI: A Survival Guide to Our Superpowered Future—arguing that AI would massively transform knowledge work by 2029, something he still believes.

“I just don’t think the full automation of processes that we are starting to see with coding is coming to other fields as quickly as Shumer contends,” Kahn wrote. “He may be directionally right, but the dire tone of his missive strikes me as fear-mongering, and based largely on faulty assumptions.”

What’s been most frustrating to me is how difficult it’s become to separate real industry signals from overwrought hype. Every hot take explodes into a narrative cascade—mega-threads, podcasts, YouTube takedowns, extravagant quote-tweets—until it’s hard to remember what the original claim even was.

Shumer says he wrote the post as a warning to friends and family about how their jobs are about to be radically upended. In a follow-up post, he clarified: “To be very, very clear: I didn’t write this to scare people. I just think it’s our duty—those of us in the AI space—to share what we’re seeing so people at least have a sense of what may be happening.”

But the reality is that we don’t yet know how this story will play out—and I worry that framing uncertainty as inevitability risks leaving people not prepared, but disoriented and confused. The hardest thing to do may be to sit with and work with uncertainty—and resist turning it into prophecy.

I returned last night from a reporting trip to Richland Parish, a rural county in northeast Louisiana, where cotton and soybean fields have been replaced by a mammoth-sized Meta AI data center called Hyperion, that is so large that it even has its own street cutting through the site called Far Far Away Lane, as you can see from the above photo.

But this was no galaxy far far away, but right here in the US, so I was dying to see it for myself — and I am also working on the second story in a series of articles about the AI data center boom and how it is impacting communities around the US (the first, a story about a project in Arizona, ran in December — here’s a gift link). The new piece will publish on Fortune online and in the next print issue. Stay tuned!

But meanwhile, I thought I’d share some nuggets about my trip to this incredibly welcoming and tight-knit community which, by the way, suffered intense damage from a heavy ice storm last week and is still recovering. There were downed trees and power lines everywhere you looked and there were many people in the area that still didn’t have power or were still in clean-up/recovery mode. I was amazed how they all pulled together to help each other through the crisis.

As for the Meta site itself, I’ve been scouring the thesaurus, trying to land on the right word to describe just how large, loud, and chaotic this 2250-acre construction site is. Colossal? Mammoth? Sprawling? Let’s put it this way: it takes a while just to drive the length of the site—it stretches roughly five miles from top to bottom, and is another mile or so wide.

I get the feeling some parts will be quite tall, too:

Oh, and there’s an elementary school ball-throwing distance from the construction site:

Meta’s Hyperion is quietly getting even bigger

During that drive, I discovered that Hyperion is getting even bigger. While the expansion had long been suspected—and was something of an open secret among some locals—I confirmed for a Fortune scoop that Meta has quietly purchased roughly 1,400 additional acres, an area nearly twice the size of Manhattan’s Central Park, adjacent to its already-mega 2,250-acre campus. I also observed active construction underway on the newly acquired land—when I wasn’t worrying about getting mowed down by the endless parade of 18-wheelers hauling materials in and around the site.

I had help on my trip — thank goodness, because it was much harder driving around the site than I thought. My husband served as chauffeur, while I took videos of (part) of the perimeter like this one:

I also had amazing help from a local community journalist and activist named Amber Perez, and her friend Erika — the two met doing community work to get help to Richland Parish residents after the ice storm. We had a blast touring the site and eating red beans and rice and chicken at Opal’s Orange Food Truck that serves some of the thousands of workers on the construction site:

Overall, the Richland Parish community was incredibly welcoming and I wish I could have stayed longer to eat more amazing food, have more amazing chats and see more of this beautiful area.

I’ll have much more to share about this part of the AI data center boom in my upcoming feature, hopefully with amazing photos and video to accompany it.

Meanwhile, back to the rest of the AI news cycle — have a great weekend.

Happy Friday!

It’s only the end of January and the joint is already jumping when it comes to AI mega-deals, mega-rounds, mega-earnings announcements and much more, well, MEGA $$$$$.

Psst: Did you hear that OpenAI is in talks to raise as much as $100 Billion in fresh mega-bucks? According to the New York Times, the company is in talks with tech giants and Middle Eastern sovereign wealth funds for a funding round that could value the company at $750 billion or more (!!!).

And two days ago, Anthropic closed its latest funding round at over $10 billion which, somehow, could go even higher.

Oh, and just yesterday, Elon Musk’s Tesla announced an agreement to invest $2 billion in — yup — Elon Musk’s xAI, which competes with OpenAI. The money-go-round won’t end there: SpaceX is reportedly in merger talks with xAI ahead of a potential IPO this year. Some are even salivating about a potential $2 trillion IPO!

I know these are a lot of exclamation points, but how else can I communicate the sheer exclamation-pointy-ness of these deals?

Spend on AI like it’s 2026

And that’s just the fundraising. How about the spend? I reported this week that on Meta’s 4th quarter earnings call, CEO Mark Zuckerberg predicted his company would experience a “major AI acceleration” in 2026 as it races to catch up after falling behind Google, OpenAI and Anthropic in leading AI models in 2025.

Now, Meta forecast its capital expenditures could rise to as much as $135 billion this year, nearly double the $72 billion it reported in 2025, as it unveiled its most radical AI spending plans to date, largely driven by increased investment in AI infrastructure costs and talent. (Speaking of talent, I’m still waiting to be able to actually speak to Meta’s highest-profile talent, former Scale CEO Alexandr Wang, now head of Meta Superintelligence Labs. I’m told he won’t speak until Meta comes out with its new models. I’ll wait.)

Much of that spend will be on Meta’s massive AI data centers and related infrastructure, like the $6 billion deal they just cut with Corning for the 175-year-old company’s fiber optic cables — I wrote about that this week, too.

That kind of data center spend can worry investors, though. Microsoft’s stock shares dove yesterday as the company’s data center spending overshadowed their massive earnings surge.

Raising billions to spend billions

But never mind — apparently there is still room in 2026 for AI startups that raise billions even though they have no products and no revenue, just some eager investors.

This Wall Street Journal piece digs into Flapping Airplanes (a reference to the biological cues future AI should take from nature), which it says is “part of a new wave of startups some have dubbed ‘neolabs,’ which give priority to long-term research and developing new AI models over immediate profits.” Profits, shmofits, amirite?

The bottom line: across much of the AI industry, from tech giants to well-funded startups, the cycle is clear—raise or earn billions, spend even more, rinse and repeat. For now, anyway.

Dog walker.

My go-to idea for the past two decades. I lived in New York City for 20 years and nothing seemed to offer better job security. Can also scale # of dogs based on the size of one’s hand and grip strength. No matter how robotics improves, there is still no way an Upper West Side dog owner is going to let a robot walk Susie the Shiba Inu.

Street musician.

I did this a few times back in the day: I brought my guitar to a New York City subway platform (nowhere near my apartment), opened my case, and had at it with acoustic covers of Oasis’ “Wonderwall,” Patsy Cline’s “Crazy,” and Madonna’s “Borderline.” I don’t think AI will ever master the subtle art of knowing exactly when to stop singing as the train screeches in—and when to repeat the chorus as commuters stream out to the booming refrain of “stand clear of the closing doors.”

Hair stylist.

I’m not saying I’d be good at it. But every time I get my hair cut, I want to tell my stylist she’s got it made. There will be no AI Edward Scissorhands in her future. Even if Claude someday becomes conscious, she’ll still be there for his (her? their?) artificial locks. Also, in 10th grade, I took a personality test that told me “movie director” and “hair stylist” were both excellent career options — so I feel like my life would come full circle.

Buddhist monk.

Even in the age of AI, you still have to do your own meditating. Um…right?

Writer.

Surprise! Seriously, I think I can pull it off even once AI — slop or not — takes over all media. The key is to keep zigging and zagging away from whatever writing style AI is hyping at the moment. M-dashes? I’ll go all semi-colon. X, then Y? I’ll counter with with nonsensical organization that will make your head spin. And no paragraph breaks ever.

Founder.

Ideas I’m considering:
Perplexity, but for birds.
Cursor, but for preschoolers.
Runway, but for community theater.
Nvidia, but for Lego.

AI therapist.

Anthropic’s AI constitution, mostly written by its in-house philosopher Amanda Askell, sets the stage for an AI that is growing up and just wants to be loved. If humans now have AI therapists, maybe AIs will eventually need human ones. I’m an HSP (Highly Sensitive Person) and an INFJ—overflowing with empathy—so wouldn’t I be perfect to help AIs navigate their adolescence?
Hmm, should this go in the “founder” category?
a16z, call me.

Circus performer.

Do I have any talent in this realm? No. But when I visited the marvelous Venardos Circus last month, I thought deeply about why I think circus performers will be protected from AI. Neither a flawless robot nor a clumsy one would make a good act. The thrill of the circus comes from watching humans push the limits of what a body can do—and from the very real possibility of failure. If a robot juggles knives perfectly, yawn. Of course it can. If it drops them, yawn again. Just another robot needing a reboot. The circus only works because someone — a human — might miss.

Pilates teacher.

In 2010, I trained to become a Pilates teacher. I spent two years doing dozens of courses and over 500 hours of student teaching on all the crazy Pilates machines, like the rack-looking Reformer, the Cadillac (which looks like a medical table), the Chair and the Tower. I only taught Pilates for a year — turns out I didn’t want to wake up early to train demanding people (usually women) at 7 or 8 am and give half my earnings to the Pilates studio I rented space from. But hey, if I get laid off due to AI, you’ll find me telling people to scoop those abs in no time!

Sharon Bot.

No, not on OnlyFans! But if I got laid off, couldn’t I use my charming and adorable AI avatar to help me power-charge my influencer campaign? I’m thinking live bot hot takes across dozens of platforms; moderating bot-heavy panels; and sending my Sharon Bot to cover media briefings. OR — just spit balling here — Sharon Bot as a stand-in for other people’s Bots? When they’re tired of their own voice and likeness? I’m seeing $$

Just to be clear, I’m hoping to keep my job at Fortune. 😀

I’ve been covering AI as a daily beat since April 2022.

From the start, the AI beat felt more fast-paced than anything I had covered in more than two decades as a journalist, writer, and editor. Then, when OpenAI released ChatGPT that November, everything accelerated. And since then, the sheer velocity of new products, infrastructure, research, companies, investments, societal shifts, geopolitics, and policy battles has only kept climbing—week after week, month after month, year after year.

Just this week, for example, the AI news cycle felt nearly impossible to keep up with. I began with a story on the AI chip startups sitting pretty after Nvidia’s Groq deal—work I’d done before New Year’s—then rushed to cover Nvidia CEO Jensen Huang’s CES keynote and OpenAI’s launch of ChatGPT Health. Meanwhile, my inbox kept exploding with pitches from top companies and investors. The flood of smart, provocative posts on social media—each pointing to a story I might want to chase—felt almost crushing.

By Friday, I felt a kind of burnout no one should experience just one week into a new year, when you’re supposed to still be coasting on holiday energy.

I know I’m not alone in this. I didn’t speak directly with any of my AI reporting peers this week. But as I watched (and envied) them share scoop after scoop on companies like OpenAI, xAI, Anthropic, and Google; report from the frenzy of the robotics and semiconductor scene at CES; try to make sense of the seemingly endless stream of announcements from Big Tech’s marketing machines; and parse the cryptic posts and reposts of figures like Elon Musk, I began to wonder: How sustainable is all of this for anyone trying to do real journalism?

Last April, a fellow AI reporter invited me to join a group chat called the “AI journalism support group,” with about a dozen members. It became a great place to let off steam, trade notes, and commiserate about the overwhelming nature of the beat.

By December, the leader of that group chat had quit their job to “take some time off from the grind,” telling everyone to “keep fighting the good fight.” We wished her well — and totally understood.

Other AI journalists I know have struggled with burnout and mental health issues, left for what are presumably less stressful PR jobs, or — particularly those hired for their ability to “get the scoops”— have simply questioned whether they can withstand the pressure.

For me, it’s been more a matter of worrying whether I can keep up with, let alone fully understand, so many new developments. While I was reporting deeply on the world of AI data centers, I missed how software developers were rocketing into the stratosphere with Claude Code. As soon as I make headway on the latest in AI inference, a wave of new consumer products hits the market. And that leaves precious little time to grapple with things like xAI’s “undressing” image debacle or the geopolitical AI race between the U.S. and China.

Some might say this is just a legacy media journalist complaining about competition—or that if I can’t stand the heat, I should get out of the kitchen. But these days, no matter how many AI-focused reporters there are on the beat, it’s becoming hard for any outlet to report and investigate deeply enough to truly uncover the contours of what may be the biggest tech story of our time.

And it’s not about legacy versus new media. I’m confident independent journalists on Substack feel this impossible pace too. Anyone trying to do serious, source-driven, investigative work on AI—regardless of platform—is struggling to keep up.

And believe me, I—and my peers—do use AI to help us cover AI. Not to write our stories, but to research, transcribe, bounce off ideas, analyze data, and more. We are working faster than ever. But a trillion-dollar industry like AI is working ten times faster.

I don’t want to get out of the kitchen. I love working the AI beat, and I’m excited for what’s to come in 2026 and beyond. I believe in the value of covering AI with care, context, curiosity, and skepticism. But I also believe that if the pace keeps accelerating unchecked, we risk not just burnout, but something far more consequential: our ability to truly focus on—and understand—the technology and the companies reshaping the world.