🤓 The Atlantic’s “Deal with the Devil” Carries a Warning for Every Leader
PLUS: Hollywood’s AI problem starts in the corner office; Tubi brings streaming recommendations into ChatGPT; The real reasons OpenAI bought a media company—and more
What do you do when the company that could help gut your business wants to buy access to the thing that makes it valuable?
That’s the choice that’s been sitting in front of newsrooms and publishers for the last two years.
The Atlantic. The New York Times. Condé Nast. The Wall Street Journal. The Financial Times.
Companies built on decades of original reporting, investigative journalism, and storytelling—work their audiences trusted enough to pay for—were asked to make licensing deals with AI companies.
To train models on their reporting.
To surface their work in AI search products.
To use their work as ingredients that generate the answers people get from chatbots instead of their websites.
And get paid for it.
On paper, it looked like a business decision.
In practice, it felt existential.
For a lot of people inside these organizations, saying yes felt like feeding the thing that was coming to eat them.
It felt like surrender.
Refusing felt more principled.
But it also meant no money, no seat at the table, and no influence over how your content showed up in AI products that were being built with or without you.
Suing felt like the clearest stand to take.
But lawsuits take years, and the technology doesn’t wait for the courts.
Some signed.
The Atlantic did. So did the AP, Vox, the Financial Times, and dozens more.
Some blocked AI companies from crawling their sites entirely.
The New York Times and Ziff Davis—the company behind CNET, Mashable, and IGN—sued.
And now the CEO of The Atlantic is looking back and saying he didn’t move fast enough.
A Deal with the Devil
Nicholas Thompson turned The Atlantic from a publication losing over $20 million a year into a profitable one with the highest subscriber count in its history.
He signed the OpenAI deal despite fierce internal resistance.
His editorial team saw it as “a deal with the devil.”
He’d do it again. And he’d go further.
More deals. Earlier. With every AI company.
“One of the most controversial things I did was the deal with OpenAI.
The editorial side saw it as a “deal with the devil.”
I wish I had introduced it differently so the staff didn’t reject it so hard.
But in retrospect, I would have done deals with every AI company.
What’s happened to the price of training data? It’s gone way down.
Thank God we did that deal.
I feel like we missed out on opportunities in that period because of all the anger over AI.
I wish I had said, ‘I see where this is going, and I’m going to act on it.’”
Read that again.
Because of all the anger.
And the anger, however justified, shaped the strategy.
It created delay. And that delay had a cost nobody was tracking.
While some publishers were still working through the internal politics, the ground was moving underneath them.
Every new deal gave the labs more leverage and left the companies still waiting in a weaker position.
The value of what publishers were selling—training data, content access—was declining as labs found alternatives like synthetic data.
The publishers who were waiting for the courts or for the dust to settle were actually watching their negotiating power erode.
If you’ve been reading this newsletter for the past few years, this argument will sound familiar: for most publishers, taking a smaller check than they thought their work was worth wasn’t giving up—it was the most clear-eyed option in a window that was closing fast.
What Thompson is describing now is the cost of that hesitation from the other side of it.
Someone who moved, and still didn’t move far enough.
So, what does this have to do with you?
The first round of the licensing fight has already played out clearly enough to teach us something.
But the decision-making problem is just getting started.
The same tension is already showing up in decisions about enterprise access, internal builds, vendor partnerships, proprietary data, and how much to commit before the technology shifts again.
With AI, timing becomes a much bigger part of the strategy.
The technology is advancing fast enough that the thing you’re deciding about—its value, its cost, its competitive dynamics—can shift meaningfully between the time you start deliberating and the time you act.
Some get clearer if you wait.
If you’re building an expensive internal system on top of technology that is still changing every few months, patience can save you from building the wrong thing and rebuilding it again six months later.
In those cases, a more flexible short-term plan can be smarter than charging ahead just to feel like you’re doing something.
But in other situations, your position weakens every month you wait.
Leverage shifts. Terms get locked in by competitors who move. Access gets defined without you. Opportunities narrow.
In those cases, waiting doesn’t buy clarity. It increases your risks and costs you ground you may not get back.
It’s the partnership terms that were available last quarter but aren’t now.
The workforce your competitors upskilled while you were still figuring out rollout.
The role-specific workflows they’ve already built, iterated on, and scaled while you were still stuck in pilot mode.
The strategic hires and scarce advisors who got scooped and are now harder and more expensive to land.
The whole game right now is knowing which kind of decision you’re facing.
Who Actually Knows What They’re Doing?
Most leaders do not have the time or bandwidth to track model releases, research progress, lab roadmaps, and where the major players are actually heading over the next three to six months.
Fair enough. They are running businesses.
The bigger problem is that many of the people being paid to advise them don’t have that depth either.
A lot of high-priced consulting firms rushed into AI, staffed up fast, and sold expensive transformation work to companies that knew they needed a strategy but couldn’t clearly define what that strategy should be.
The Wall Street Journal also reported that many clients came away feeling consultants lacked meaningful AI expertise and were learning at the client’s expense. Even Deloitte had to partially refund the Australian government after an AI-assisted report included fabricated citations and other errors.
In too many cases, one side collects millions while the other is left with failed pilots, wasted months, and projects that are obsolete before they launch.
That doesn’t mean those firms have no role to play. OpenAI is now partnering with McKinsey, BCG, Accenture, and Capgemini because this work does need strategy, integration, workflow redesign, and change management, not just good models.
But it does mean the expertise that matters most right now often doesn’t come from the places companies instinctively turn first.
The people most equipped to help with these calls are often under the radar.
They don’t have the big firm names.
But they’re the ones close enough to the technology to ask better questions, spot what’s shifting, and tell the difference between a decision that rewards patience and one that punishes it every month you wait.
So how do you find them?
Look for four things.
1️⃣ Proximity. How have they actually spent the last three years getting close to the technology? What are they doing week to week to stay current on the research, the models, capabilities, and what the labs are actually building toward? How are they using the tools themselves? (This last one is the most important one.)
2️⃣ Translation. Can they connect what’s changing in the models, the research, and the product roadmaps to actual business decisions?
3️⃣ How they think through uncertainty. Are they honest about what they still don’t know? What are they watching closely, what would change their view, and what do they believe is still foundational enough to focus on while the rest is moving?
4️⃣ What they have to show for those years of immersion. Writing. Experiments. Product work. Sharp analysis. A body of thinking that clearly comes from someone who has actually been at it.
A lot of people can talk around AI for hours.
The difference shows up when you push past the language and ask what they know, how they know it, and how it should change the decision in front of you.
Even then, you still can’t outsource all of your judgment here.
You also need to get closer to the technology yourself.
Not from presentations or strategy reviews.
From using it. A lot.
There is no shortcut to building better judgment about what these systems can actually do and how they’ll transform your business and role.
None of this is easy.
These are some of the hardest strategic calls leaders have had to make in a long time. Maybe ever.
The ground is shifting, the rules are still being written, and even very seasoned decision-makers are being asked to make consequential calls without the kind of clarity they’re used to.
That’s exactly why a deeper read on what’s changing—and better guidance around these decisions—matters so much.
What Thompson is describing is not a one-off.
It’s an early warning of a pattern that’s going to repeat.
In more industries. On bigger decisions. With higher stakes.
The real work now is preparing for the harder calls before they arrive.
📈 Work with Me
AI Advising and Consulting: Strategic guidance to build sustainable AI strategies and adapt to shifting audience behavior
AI Training: Practical training that helps teams use AI more strategically, effectively, and safely—so it becomes a trusted thought partner, not just a task assistant
What You Need to Know About AI This Week ⚡
Clickable links appear underlined in emails and in orange in the Substack app.
🎬 Hollywood’s AI problem starts in the corner office.
A new Hollywood Reporter piece found assistants across studios, agencies, and networks already using AI for everything from scheduling to script coverage just to keep up with heavier workloads and smaller teams.
A lot of that use is happening through personal accounts, with very little training or oversight, and with real anxiety about what it means for their jobs.
The bigger issue is that they’re the ones being left to figure it out first.
If the people at the top still don’t understand what these systems can do for higher-value work like research, analysis, strategy, decision-making, and high-stakes communication, they’re not in a strong position to make smart calls about how this technology is changing the core of the business.
That makes it much harder to make informed decisions about AI budgets, hiring, training, tools, or policy.
And without proper training, clear policies, and the right systems in place, you get exactly what this article describes.
People put sensitive or confidential information into tools no one approved, and they produce more mediocre work at greater speed and volume—work that might look like productivity but actually creates more problems than it solves.
To be clear, some companies are already moving.
Over the past several months, I’ve been meeting with leaders across the industry, and a few are genuinely much further along than others.
They have cross-functional task forces.
They have well-resourced ops teams working with departments to build role-specific, repeatable workflows around the kinds of administrative and tactical work AI handles well.
This is a smart place to start.
But it’s only a start.
Do you think the Pentagon and Big Pharma are using AI mainly to speed up routine and administrative workflow tasks usually performed by their support staff?
Of course that’s part of it.
But the bigger value sits higher up: better research, better analysis, stronger strategy, and smarter decisions.
That’s one reason a large share of my clients over the last two years have been senior executives.
They could see needs emerging faster than the organization could respond, and they wanted a way to help their teams move forward while sharpening their own judgment and AI skills, especially as those skills become more important to staying competitive.
A lot of them were also trying to get ahead of a more personal question: if AI is starting to reshape their function, their role, and the value they bring, how do they build enough understanding to figure out their next move?
That’s also why I always start AI workshops with the leadership team.
Once leaders understand these higher-value capabilities firsthand, the conversation changes.
They stop thinking only about staff training and tool adoption.
They start asking harder questions:
🔷 How does this reshape our core business and how we work?
🔷 How should the company itself be structured for what is coming next?
🔷 As AI gets better at parts of my job, where do my expertise and experience matter most?
Related Post:
📺 Tubi is bringing streaming recommendations into ChatGPT.
It’s the first streamer to launch a native app there, letting users ask for something to watch in plain English and get recommendations inside the chat.
Tubi already tried this inside its own app with Rabbit AI, then discontinued it a year later.
Moving that experience into ChatGPT instead says a lot about where discovery is headed.
As I’ve been writing, AI is increasingly becoming the place where people discover, research, compare, and decide what deserves their time and money.
These tools pull together reviews, fan discussions, and the wider web, then turn all of that into personalized answers built around the exact question being asked.
And this changes the power dynamics: The companies that own the path to the decision start owning more of the relationship with the audience.
Related Posts:
📈 Why did OpenAI buy tech’s hottest news show?
Fresh off the biggest funding round in Silicon Valley history—$122 billion at an $852 billion valuation—OpenAI acquired TBPN for an undisclosed amount.
This marks OpenAI’s first major media acquisition.
If you’re not familiar, TBPN is a daily live tech-and-business show often referred to as SportsCenter for business.
The show averages about 70,000 viewers per episode, which is still small by mass-media standards, but much more influential than its size would suggest.
It has real pull inside Silicon Valley.
Its guests have included Sam Altman, Mark Zuckerberg, and Microsoft CEO Satya Nadella, and other high-profile executives, some of whom rarely speak to legacy media outlets.
So, why buy it?
1️⃣ First: distribution.
TBPN reaches exactly the kinds of people OpenAI cares about most: founders, engineers, operators, investors, and tech executives.
The show helps shape what this crowd pays attention to, who they take seriously, and how they talk about the companies driving AI forward.
And because the show lives natively on X, its clips travel fast and far beyond its live audience.
2️⃣ Second: the IPO.
That distribution matters even more this year ahead of an expected IPO, and TBPN already sits unusually close to the exact audience OpenAI wants paying attention: tech finance people, ambitious builders, decision-makers, and future shareholders.
Its exclusive NYSE partnership only strengthens that case.
3️⃣ Third: TBPN knows how to package a company for this audience without sounding like corporate comms.
The hosts are AI optimists who are fluent in startup and tech culture, and good at turning company messaging into content their loyal audience actually watches and shares.
OpenAI’s own announcement makes clear that this was part of the appeal: beyond keeping the show running, it specifically called out TBPN’s “comms and marketing instincts” and said the team will also contribute inside OpenAI.
Chris Lehane, OpenAI’s Chief Global Affairs Officer, framed the deal less like a media acquisition and more like bringing a top digital-first marketing and communications shop in-house.
OpenAI says TBPN will remain editorially independent.
But a show largely covering AI while being owned by the most powerful AI company, during its IPO year, raises obvious trust questions.
One more notable detail: TBPN did about $5 million in ad revenue in 2025 and was on pace for more than $30 million this year, yet OpenAI is winding that business down. That alone is telling.
But with AI labs especially, I tend to assume we’re usually seeing only a small part of the strategy.
The technology is moving fast enough to reshape the terms, incentives, and power dynamics in ways that aren’t fully visible in real time.
That’s one reason I rarely trust overly certain takes on deals like this.
🚨 Anthropic says its newest model is too dangerous to release.
Anthropic announced Claude Mythos Preview, which it says is its most powerful model and represents a “step change” in AI performance.
But the company also says the model poses serious new cybersecurity risks, largely because of its advanced reasoning capabilities.
So instead of releasing it widely, the company is sharing early access with more than 40 companies including Apple, Google, Microsoft to help find and patch vulnerabilities in critical systems.
This marks a striking and unsettling moment in AI development: the same capabilities every frontier lab is chasing are now raising risks serious enough to hold a model back from public release.
💡 Why Isn’t AI as creative as humans yet?
Jack Clark—Anthropic co-founder, former researcher, and now the company’s lead on policy and public benefit—shared one of the sharpest explanations I’ve heard for why AI still hasn’t come up with truly original, intuitive ideas.
His point: today’s models can help scientists make advances in math, biology, and physics.
But they still haven’t had a CRISPR or theory-of-relativity moment of their own.
Clark’s theory is that some part of human creativity may come from “idling”: working hard on a problem, stepping away, going for a walk, swimming, living your life, and then having the idea show up somewhere far from your desk.
He also suggests there may be an improvisational element here that these systems still lack—the kind of off-script, intuitive leap that doesn’t come from pushing straight through the work.
AI can keep working.
It can even grind 24/7.
What it can’t do, at least not yet, is step away and let a problem simmer.
It has no presence, no way of interacting with the physical world, or moving through the world while a thought keeps forming in the background.
I’ll let him explain it in his own words👇.
If understanding that gap gets us closer to understanding what creativity actually is and how it happens, that may turn out to be one of the most interesting things AI helps us uncover.
📰 Reporters are drawing a line at AI-generated rewrites.
A new newsroom AI tool can turn reporters’ previous work into fresh stories with new headlines and angles for different audiences. But Sacramento Bee journalists are refusing to let those pieces run under their bylines — the line that tells readers who wrote the story—saying it puts credibility, source trust, and the public’s faith in local news at risk
In case you missed the last edition, you can find it 👇:
That's all for this week. See you in 2 weeks.
Thoughts, feedback and questions are always welcome and much appreciated. Shoot me a note at avi@joinsavvyavi.com.
Stay curious,
Avi
💙💙💙 P.S. A huge thank you to my paid subscribers and those of you who share this newsletter with curious friends and coworkers. It takes me about 20+ hours to research, curate, simplify the complex, and write this newsletter. So, your support means the world to me, as it helps me make this process sustainable (almost 😄).