Hi, I’m Jamaal. I’m a CFO, investor, and advisor working across technology, labor markets, and finance. I write about the future of work, money, and AI — and how those systems shape who gets opportunity.

If that’s relevant to you, subscribe below.

Subscribe now

Read time: 4 minutes

TL;DR

• Jane Street generated nearly $40 billion in revenue last year with 3,500 employees, roughly $9 million in profit per person.

• The firm just committed $7 billion to AI cloud infrastructure and is still hiring.

• If AI were actually replacing jobs, Jane Street would be among the first firms to cut headcount. It’s doing the opposite.

• The companies doing mass layoffs in 2026 — Meta, Amazon, Microsoft, Oracle, Block, Snap — are public companies with bloated headcounts, languishing stock prices, and investor pressure

• AI is being used as a disguise for corporate retrenchment, a way to free up capital for infrastructure spending, and a signal to Wall Street

I started on Wall Street at J.P. Morgan in 2005. Back then, the most ambitious paths for a new finance grad were a couple of years in investment banking before moving to private equity, or a stint in sales and trading before jumping to a hedge fund. During my time, I’d never heard of Jane Street, the secretive quantitative trading firm, which was then only a few years old. I knew a few people at “prop trading” shops, but those firms didn’t appear to attract the same caliber of talent as the banks. From my vantage point, they always seemed a little sketchy. I wouldn’t hear about Jane Street until years later. It really entered public consciousness with the rise of Sam Bankman-Fried, the now-imprisoned founder of crypto exchange FTX, who’d started his career at Jane Street.

Last year, Jane Street generated nearly $40 billion in revenue, roughly $31 billion in EBITDA, and about $13 billion in profit. It did this with just 3,500 employees, roughly $11 million in revenue, and $9 million in profit per employee.

Jane Street has become known for attracting deeply technical, highly credentialed quantitative talent from the best schools and best employers. It runs tens of thousands of high-end GPUs and builds the deep learning models that directly power its trading strategies. Earlier this month, Jane Street committed $7 billion to CoreWeave, the AI cloud provider, $6 billion to computing infrastructure, including Nvidia's next-generation chips, and a $1 billion equity stake, making it one of CoreWeave's largest shareholders. This is a firm that is betting heavily on AI.

So if AI is replacing jobs, how come Jane Street isn’t laying anyone off?

If any firm had the talent, the infrastructure, and the financial incentive to replace humans with AI at scale, it would be Jane Street. Instead, the firm has grown headcount roughly 15% annually through 2024, accelerating to over 20% in 2025.

The Companies That Are Cutting Aren’t Cutting Because of AI

In 2026, some of the most high-profile tech layoffs have been explicitly tied to AI and automation, including major cuts at Meta, Amazon, Microsoft, Oracle, Salesforce, Block, and Snap. In many cases, executives framed these reductions as efficiency moves.

I’ve been writing about this for over a year. Much of what gets labeled an “AI layoff” is a financially motivated signal to investors, a distress signal in disguise, or a necessary reprieve for unsustainable AI capital expenditure. These layoffs are happening at public companies with large headcounts, high labor costs, languishing stock prices, and explicit or latent pressure from investors. They are not happening at firms like Jane Street, where AI is actually central to the business.

Jane Street is one firm, and one firm is an anecdote. But it’s an illustrative one. When a company that lives at the intersection of AI and profit is still hiring, and the companies claiming AI forced their hand are the ones with bloated headcounts and impatient shareholders, the pattern starts to look less like technological disruption and more like financial convenience.

Here’s what’s actually happening.

AI is a disguise for corporate retrenchment. Economic growth is tepid. Geopolitical instability, volatile markets, and shifting trade policy have made long-term planning harder than at any point in recent memory. And the tech sector is still unwinding the massive overhiring of the COVID era. Meta tripled its headcount during the pandemic. Amazon added roughly 800,000 employees between 2019 and 2021. Jack Dorsey admitted he overhired at Block because he built two separate company structures instead of one. In each case, the cuts map more cleanly to the scale of the overhiring than to anything AI is doing. You could remove “in the age of AI” from any of these announcements and the logic would hold up fine.

Companies are freeing up capital for AI infrastructure spending. Amazon, Google, Meta, and Microsoft alone are expected to spend roughly $650 billion on capital expenditures in 2026. That money has to come from somewhere. Meta’s internal memo said the layoffs were necessary “to allow us to offset the other investments we’re making.” TD Cowen estimated that Oracle’s layoffs alone could generate $8 billion to $10 billion in incremental free cash flow. As I wrote last year, AI isn’t directly taking jobs, but AI investment may be cannibalizing hiring. Companies aren’t firing people to replace them with AI. They’re spending so much building AI systems that there’s less money left for salaries.

These layoffs are a signal to Wall Street. The stock market has been consistently rewarding headcount reductions since at least 2022. Block’s stock jumped more than 20% on the day of its layoff announcement. Snap’s cuts came weeks after activist investor Irenic Capital demanded changes. Profitable companies now routinely do layoffs. Half of the most profitable companies on the planet did mass layoffs last year. Investors see headcount reductions as a sign of capital discipline, and CEOs respond accordingly.

What Jane Street Actually Shows You

Jane Street doesn’t answer to public market investors. It doesn’t have a languishing stock price. It isn’t unwinding a pandemic hiring binge. And it isn’t cutting people because AI can do their jobs.

The current conversation around AI layoffs is almost entirely about efficiency — fewer workers, lower costs, wider margins. You don’t hear anyone saying AI will double revenue or dramatically expand capacity. That asymmetry is a tell. It tells you that AI “efficiency” is a smokescreen for corporate retrenchment in a market defined by uncertainty, massive infrastructure spending, and relentless investor pressure.

The next time a CEO tells you AI forced a round of layoffs, ask yourself why many of the firms most equipped to replace humans with AI are the ones that keep hiring them.

Hi, I’m Jamaal. I’m a CFO, investor, and advisor working across technology, labor markets, and finance. I write about the future of work, money, and AI — and how those systems shape who gets opportunity.

If that’s relevant to you, subscribe below.

Subscribe now

Read time: 4 minutes

Last week, I spoke with Maxime Recoquillé, a tech and digital economy reporter at L’Express — one of France’s oldest and most widely read weekly news magazines — about the cost of AI inference and agents and what it means for organizations. The piece was published over the weekend. I was quoted three times.

The piece explores whether the explosion of agentic AI is consuming tokens at a rate the labs can sustain — and what that means for the flat-rate subscription model most users rely on. The article covers a lot of ground in a short space, and naturally, only a portion of what I shared made it in. I’ve shared an expanded version of my answers and my thinking on the subject below.

How long can the real cost of AI stay hidden inside flat-rate subscriptions?

The subsidies can’t last much longer. These companies’ capital needs will require IPOs soon, and preparing for those IPOs will reveal the real economics. Users are getting a deal right now that doesn’t reflect the actual cost of inference. When that ends — and it’s going to be sooner rather than later — flat-rate subscriptions will either get more expensive, get throttled harder, or both. My guess is we see pricing tiers segment more aggressively before we see across-the-board hikes.

Enterprise users will see cost increases before consumers. Consumers are more price-sensitive, and the broader media narrative is tied to consumer prices, giving the labs an incentive to protect that tier for as long as possible. Enterprise can absorb more, complains less publicly, and has fewer substitutes.

Total cost to end users is going up.

What does your own token consumption look like? Do you use agents, and has the cost surprised you?

Personally, I rely heavily on the fixed-rate, subsidized plans from Claude and ChatGPT, and I wait out the rate limit caps when I hit them. I’ve been on ChatGPT Plus since it launched, and last year I upgraded from Claude’s entry paid tier to the mid-tier plan after hitting the limits too often. When I’ve been impatient, I’ve paid for more inference or bumped up further. If I need something now, I pay. If I can wait an hour, I wait.

I’ll call something an agent if it makes me sound smart to the audience I’m talking to, even if it’s only modestly autonomous. The term has gotten elastic.

Professionally, the budgets I oversee are fixed month to month with a very high bar for increasing them.

Is tokenmaxxing real? Does consumption go hand in hand with productivity?

I think agents are both overhyped and genuinely promising. The conversation is confused because there’s no objective definition of an agent. Anything that doesn’t require continuous prompting is getting labeled as one. There’s cachet in deploying agents and talking about it publicly. You paint yourself as further ahead than everyone else, and widespread AI anxiety fuels demand for that signal.

Where agents are real and productive is in engineering. Software teams are seeing actual gains. That’s created a bifurcation of worlds. Engineers feel like they’re living in the future and can see how an agent might take their job. Everyone else, doing work that’s less rules-based, less structured, less deterministic, is much further from disruption than the headlines suggest.

Consumption does not go hand in hand with productivity. There’s a lot of conflation between AI activity and AI productivity. People are doing lots of stuff with AI. The activity is real. How much of that activity is actually productive remains an open question.

I don’t think tokenmaxxing is real. It’s shown up in a few media articles, and leaderboards at places like Meta read more as management theater than anything else. No serious organization is doing this. Inside companies flush with cash and trying to push employees to experiment, tokenmaxxing might work as a short-term proxy for helping people become comfortable with the tools. This isn’t sustainable, serious, or rational as a long-term practice.

When token spend approaches or exceeds an employee’s salary, how should organizations think about budgeting for it?

The typical capital allocation frameworks still apply. These are expensive, highly variable cost software programs. Treat them that way.

Step one is understanding how all of this actually works. The difference between token usage embedded in fixed-price subscriptions, which are heavily subsidized, and token usage via the API. The difference between AI usage for products, where the cost sits inside the cost of sales and ROI is measurable, and AI usage for workflows, where the line from inference spend to revenue is indirect. Most people don’t have a clear picture of any of this. There’s no shame in that. Your job is to figure it out.

Step two is visibility. I know of several software and data science teams that spent five and even six-figure sums on inference in very short periods without meaning to. None of it was sustainable, and all got dramatically scaled back once finance saw the numbers. You can’t budget for what you can’t see.

Step three is ownership, and this is where most companies are flying blind. The organizations figuring this out have an AI lead who is AI-native, or close to it, overseeing usage across the company. When AI lives inside the legacy IT department, you get process and compliance. Experimentation and cost discipline come from giving it its own home.

The labs are pricing below cost to capture market share. How should enterprises prepare for when that ends?

That moment is coming sooner rather than later. Plan for it now. The most valuable expertise to build in-house is inference efficiency. How do you get the same output at a lower cost? I expect this to become a well-paid specialization in the next few months.

Practically, that means two things. Build visibility into your inference spend now. Stress-test the ROI of your workflows at 2x and 3x current prices and see which ones still pencil out.

Hi, I’m Jamaal. I’m a CFO, investor, and advisor working across technology, labor markets, and finance. I write about the future of work, money, and AI — and how those systems shape who gets opportunity.

If that’s relevant to you, subscribe below.

Subscribe now

Read time: 6 minutes

I’ve got three scam recruiting emails in my inbox right now. Three, in one week.

I must have ended up on a list somewhere. Typically, I get the occasional LinkedIn DM about an executive or board role that looks credible at first glance but falls apart after a single read. Often, the scammer had hijacked a real person’s LinkedIn profile. Other times, the profile itself is obviously fake. But these three emails were different. They landed directly in my email, unprompted, within days of each other. Something changed.

They’re all from “recruiters” pitching me for board or executive roles that I’d come to realize almost certainly don’t exist, each designed to get me engaged enough to eventually hand over some money or some information. But they’re not all the same quality. The gap between the worst and the best tells you something about where fraud is headed.

The Obvious One

The first is sloppy. Someone claiming to be a recruiter from Korn Ferry pitches me three different roles at UAE-based organizations. The language reads like a LinkedIn cold DM from someone who spent thirty seconds on my profile. And the dead giveaway: he’s emailing from a Gmail address while linking to Korn Ferry’s leadership page as if that makes him credible. In fact, the scammer was pretending to be a real Korn Ferry employee.

Better, but still obvious

The second is better. The sender claims to be with The Addison Group, a firm I’d never heard of but one that — after a quick Google search — appears to be real. His affiliation with the firm is awkwardly stated, but his explanation of what he does is tighter: “I specialize in direct placement recruitment for top-tier organizations.” While this sounds like a real recruiter, the outreach was generic. Nothing specific was mentioned about my background, just “I was impressed by the level of operational precision and executive support you’ve demonstrated throughout your experience.” You could say that to anyone.

The Gmail address was the biggest red flag. I knew then it was likely fake, but I played along. When I pushed back and asked for a working LinkedIn link and firm verification, he pivoted to “I came across your profile while reviewing a number of professionals with strong leadership and operational experience.” Lastly, the LinkedIn link didn’t even work.

The One That Almost Got Me

The third one — which was actually the first to arrive — is the one that matters.

The sender emailed me regarding a confidential mandate for an audit committee chair on behalf of a PE-backed digital financial services platform. This is a role for which I’m well-positioned. It hit several keywords in my background: fintech, capital markets experience, board-level audit leadership. The compensation was broken out across board retainer, committee chair premium, and equity in a way that tracked with how real board packages are structured. The job description ran several pages and read like a legitimate board spec.

The positioning was smart. He called himself a “Career Coach / Employer Relationship Specialist” and linked to the LinkedIn profile of a real person with genuine recruiting experience. He didn’t oversell himself like the fake Korn Ferry guy. Instead, he framed himself as an independent practitioner who had “transitioned to a more focused, relationship-driven model” after leaving a broader search platform. I’ve met people who actually work this way.

The drip campaign was textbook. First email: here’s the opportunity, let me know if you’re interested. I replied. Second email: a detailed brief. Third email: send me your resume and board bio. Each step got me more invested. Each step felt like a real search process.

When I said I’d need a couple of days to update my materials, he moved fast: “Rather than spending time refining the board bio or resume upfront, feel free to send over whatever version you have on hand. I can review it in the context of what the Board is specifically prioritizing.” And then: “If helpful, I can also point you toward a couple of specialists who work specifically with senior executives and board candidates, in case you decide you want a more formal refresh after we align on positioning.”

That’s where it clicked. The eagerness to see my resume early, the quick pivot to coaching, the referral to resume-writing services. A quick Google search revealed this is a known scam pattern. The “opportunity” exists to funnel you toward paid services that have nothing to do with an actual board seat.

I’ve long prided myself on being good at spotting fake stuff on the internet, although I’ve recently conceded that we’ve entered a new era where real and fake have too often become indistinguishable. In this instance, it took me a while to get there. The quality of the writing, the specificity of the role, and the calibration to my exact background. This wasn’t someone with a phishing template. This was someone, or something, that had studied how I show up online and crafted an approach to match.

The Tells

I still caught them. Here are a few red flags that gave it away:

• Gmail addresses. The single biggest giveaway. I’ve worked with plenty of independent recruiters and one-person shops. Every one has their own domain. These scammers are using AI to write polished, industry-specific outreach, but can’t be bothered to register a domain name. I expect that gap to close soon. It doesn’t take much more AI to build a credible website with generated headshots and bios. When that happens, these scams get much harder to spot.

• Confidentiality that never goes away. A PE-backed board search starts confidential for good reason. You don’t want to signal governance issues or tip off the market during a capital raise. But relatively quickly, a legitimate recruiter names the PE firm. That’s how they build credibility. This one never did. When I asked directly and offered to sign an NDA, he deflected with another paragraph about his “relationship-driven model.”

• Coaching that was too eager and too early. Real recruiters coach candidates. But it comes after vetting, after the recruiter has decided to present you to the client. This guy jumped straight to offering to review my resume and connect me with “specialists” before we’d ever spoken live. The combination of eagerness and referral to paid resume services is the fingerprint of this scam.

The Bigger Problem

My inbox is one data point. The broader numbers are worse.

The FTC reported $12.5 billion in total fraud losses in 2024, up 25% from the prior year. Job and employment scam losses alone tripled from $90 million in 2020 to $501 million in 2024. Those are just the reported numbers; the FTC says the real figure is likely much larger.

The scariest recent case: Arup, the British engineering firm behind the Sydney Opera House. In early 2024, a finance worker in their Hong Kong office joined a video call with people who looked and sounded exactly like the company’s CFO and several colleagues. All deepfakes. The employee transferred $25.6 million across fifteen transactions before anyone caught it. Arup’s chief information officer called it “technology-enhanced social engineering.” No systems were hacked. No passwords stolen. The attack targeted the employee’s trust in what he could see and hear.

The scam emails in my inbox are at the low end of the spectrum: well-written but easy to spot. The Arup case is the high end: real-time deepfake video, impersonation of named executives, tens of millions moved in a single attack. The distance between the two is shrinking.

What Most of the AI Conversations Miss

So much of the popular discussion around AI is about whether it will take your job, write your novel, or pass the bar exam. Very little addresses the fact that the same tools are available to people who have been running scams for years and now have a massive upgrade. Anthropic just launched a limited release of its Claude Mythos model, which found thousands of previously undetected software vulnerabilities in every major operating system and web browser, some dating back decades. The company is restricting access specifically because the capability is dual-use: the same model that helps defenders find bugs helps attackers exploit them. That tradeoff applies to every AI capability, not just cybersecurity. The tools get better for everyone, including the people you don’t want to have them.

The bad guys have always been there. They just write better emails now. They produce more convincing job descriptions. They respond to pushback with polished paragraphs that sound like a real professional on the other end. They don’t need to hack your systems. They need to earn your trust long enough to get you to hand over your resume, your bank details, or your wire transfer authorization.

The scam no longer has to sound sloppy to be fake. In recruiting, in finance, in corporate communications, that’s the new problem: not just fraud, but fraud that sounds like business.

Hi, I’m Jamaal. I’m a CFO, investor, and advisor working across technology, labor markets, and finance. I write about the future of work, money, and AI — and how those systems shape who gets opportunity.

If that’s relevant to you, subscribe below.

Subscribe now

Read time: 6 minutes

The Takeaway:

• Anthropic’s viral AI jobs chart doesn’t say what everyone thinks it does. It measures whether AI can speed workers up, not whether it can replace them.

• The underlying research was written by three OpenAI employees. Anthropic, which profits from inflating AI’s impact, built its chart on top of it.

• No job category hits 100%. There are tasks in every occupation that AI can’t even speed up, let alone do.

• The gap between what AI could do and what it’s actually doing is a product roadmap, not a pink slip.

• AI might handle 97% of a knowledge worker’s job. The last 3% is the hard part — and it’s going to resist automation far longer than people expect.

A few years ago, I was working on a middle-market private credit deal. My client, an impact investment firm, was lending a few million dollars to an Atlanta-based small business. There’d been months of diligence and negotiations. ChatGPT had been launched about a year earlier. Somewhere near the closing, while responding to an email from an associate at the law firm we’d hired to finalize the paperwork, I remember thinking: no matter how good AI gets, it is going to be a long time before human beings who are investing millions of dollars will completely trust AI to do this kind of legal work.

I didn’t think this because AI couldn’t draft the documents. It definitely could. Or because it couldn’t then suggest necessary edits, or even add that unconventional but necessary clause to the final version. It could do that too. In fact, AI could have almost certainly done all of it. The constraint was how much humans would trust this process entirely to AI. I wondered how much we would trust AI to cover us in the edge cases. If the deal goes well, I thought, AI’s output will be more than adequate. But the entire reason you spend so much time and money on legal documents isn’t for when things go right, it’s for when things go wrong. It’s for when the company is struggling, can’t pay you back, or when one party believes another is in breach of contract. It’s for when things go sour. And until we trust AI enough to be your lawyer in those hopefully unlikely yet all too common scenarios, there will always be humans in the loop.

I thought about that deal when I saw the Anthropic chart that flooded my feeds a few weeks ago.

The Chart Everyone Shared, and Nobody Read

In early March, Anthropic published a research paper titled “Labor Market Impacts of AI: A New Measure and Early Evidence.” Buried in it was a radar chart mapping AI’s “theoretical coverage” across job categories. The chart spread everywhere. The paper didn’t. The general sentiment, at least anecdotally, was: “This is our progress report on which jobs AI can take.

Fortune’s headline: Anthropic mapped out which jobs AI could potentially replace, warning that a “Great Recession for white-collar workers” is possible. CBS framed it as Anthropic tracking which jobs are “most exposed to artificial intelligence.” TheStreet: “Anthropic finally reveals which jobs AI cannot replace...and those most at risk.” Social media distilled it further into listicles and reels about “which jobs AI is actually replacing right now.”

I was skeptical the moment I saw it.

First, the chart was created by Anthropic, a company with a significant financial stake in selling you the idea that AI’s economic impact will be enormous. Second, we are so saturated with AI hype and marketing that the default position should be that anything you hear about AI is driven by an ulterior motive beyond pure information. And third, the phrase “theoretical limits of AI” is so nebulous that it should immediately raise the question: how is anyone defining it?

So I spent a few minutes doing something that, based on the coverage, almost nobody else did. I read the chart, the footnotes, and the source paper.

The Chart Actually Measures the Limit of How Much AI Can Speed a Human Worker

The blue bars in the chart represent an exposure measure from a 2023 paper by three OpenAI researchers and a Wharton professor who studies the economic impact of digital technologies. The measure captures whether an LLM, or LLM-powered software, could, in principle, cut the human time needed to perform a task by at least 50 percent while maintaining similar quality. The researchers defined “exposure” at the task level, then aggregated those scores to the occupational level.

Anthropic used those occupation-level exposure scores as the “theoretical coverage” baseline in their chart. The blue bars represent the share of tasks in each occupation category that could technically be sped up by that much, regardless of whether any firm is actually using LLMs for those tasks today.

Here’s the critical distinction: this is a measure of whether a worker using an LLM could cut task time in half. Not whether an LLM could do the job on its own. The measure is one of LLMs aiding workers, not replacing them.

There’s another detail in the chart that most people missed. None of the blue bars extend to 100%. For every occupational category, the researchers concluded that LLMs were incapable of speeding up workers by 50% across all tasks. Even in the most optimistic reading of the paper, there are tasks in every job category where LLMs offer no meaningful time savings. That’s a built-in constraint on the idea that an LLM can fully take over a job. If it can’t speed up all the tasks, can it even do them? And if not, can it do the job as a whole?

Nowhere in the paper does it estimate which jobs are most likely to be taken.

You wouldn’t have known that from virtually any of the social media posts. Instead, the chart became ammunition for more AI fear-mongering, stripped of context and repackaged as a replacement risk scorecard.

It’s also worth flagging that three of the source paper’s authors were employed by OpenAI at the time of publication. There is ample reason to question the paper’s independence before you even get to Anthropic’s use of it.

What the Chart Could Actually Be Useful For

But assume you take the paper at face value. What’s the most productive reading?

I see this primarily as an opportunity gap map. The delta between the red bars (what AI is actually doing in practice) and the blue bars (what it could theoretically assist with) represents the space to build tools that fill the gap. That’s a product roadmap, not a pink slip.

Here’s Where I Might Be Wrong

The paper was published in 2024. In large language model development, two years is an eternity. The models available today are meaningfully more advanced than those available when the research was conducted. If anything, the findings likely underestimate the theoretical potential of AI assistance.

And while the ability for an LLM to reduce the time to complete a job’s tasks is not the same as doing the entire job, it’s potentially a proxy for it. If my job is comprised entirely of spreadsheet analysis, and every task can be done in half the time, you could make the case that it won’t be long before the LLM can do the job outright. If you truly believe LLMs will replace jobs wholesale, this chart is a rough proxy for which jobs get replaced sooner rather than later.

But one of the trickier issues in predicting how AI will affect the job market is that it’s hard to predict when LLMs will be able to do the hardest, final few percent of the work. AI might be able to do 97% of an accountant’s job relatively soon. But that last 3% might require significantly more time. It might be something beyond AI’s inherent capabilities. Or it might be something humans simply never allow AI to do.

That’s what I kept coming back to during my private credit deal. The lawyer’s job is all of that stuff: the routine drafting, the complex structuring, and the edge-case protection. AI might handle 97 or 98% of it. But making AI good enough to do that final 2%, the part that matters most when things go wrong, is going to be much harder than people think.

Hi, I’m Jamaal. Welcome to Jamaal’s Letter, where I share personal reflections, insights, predictions, and strong opinions on the future of money, work, education, policy, and more.

The people who read this newsletter are the people who run the world. If you aren’t yet a subscriber, consider joining them by subscribing below. Please share it with your friends and colleagues if you find it helpful:

Subscribe now

Read time: 4 minutes

TL;DR:

• OpenAI killed Sora yesterday, six months after launch, despite topping the App Store and landing a billion-dollar Disney deal

• The stated reason is compute costs ($15M/day in inference against $2.1M in lifetime revenue), and that’s partly true

• But companies fund products people love, Sora’s 45% download drop and collapsing retention suggest this is also a demand story

• People enjoy AI-generated content when they don’t know it’s AI (~22% of TikTok is substantially AI, and half of it isn’t labeled)

• When people know content is AI, the entertainment value drops — Sora was the first platform built entirely around knowingly AI-generated content, and people left

• Short-form video runs on authenticity (”this actually happened”); AI-generated fiction may work eventually, but not in that format yet

• Prediction: someone will build a social platform where AI content is prohibited, and it will do very well

I’m in a group chat with some members of my family that is almost exclusively funny videos, memes, and random internet links. I usually catch up on whatever’s piled up during the day after my kids go to sleep. But more and more, I’m the one killing the vibe, pointing out that a video someone sent isn’t real. It’s AI. Every time, the replies go quiet. Someone usually hits me with some version of “you’re no fun,” and the chat moves on.

The interesting thing is that knowing a video is AI doesn’t change the content. The cat opening a door on its own is still impressive. The toddler saying something weirdly adult is still hilarious. But knowing the video is AI almost always changes the entertainment value. The cat isn’t as impressive when you know it never actually did that. The toddler’s line hits different when you know no kid actually said it.

The content is the same. The experience of it isn’t.

I kept thinking about this when OpenAI killed Sora yesterday. The app launched in September, topped the App Store charts, landed a billion-dollar Disney deal, and six months later, it’s done. The Disney deal is off, too.

A lot of people are treating this as a sign of distress at OpenAI, or evidence of an impending AI bubble. For the record, I think we’re in some version of an overheated AI market. What it looks like when it cools remains to be seen. OpenAI will undoubtedly be a generational company, even if its current cash burn is probably unsustainable.

Sora’s economics were terrible, but that’s not the whole story

There’s also a straightforward business case for Sora’s death. By some estimates, OpenAI was spending $15 million a day on inference for Sora, against reported total lifetime in-app revenue of $2.1 million. The head of Sora himself publicly called the economics “completely unsustainable.” Its applications chief told staff they “cannot miss this moment because we are distracted by side quests.” Killing Sora was, in part, a compute cost decision and a pre-IPO cleanup. That’s fair.

But companies find a way to fund products that people love. If Sora’s retention had been strong — if users were coming back daily, creating and sharing and pulling their friends in — OpenAI would have found the compute. The fact that they didn’t fight harder to keep it alive tells you something about demand.

And that, I think, is the more interesting story: people don’t want to watch fake stuff when they know it’s fake.

Short-form video runs on authenticity, and AI can’t fake that

We love fiction. We watch comic book movies, animated films, and scripted TV. But that kind of content involves real craft, real investment, intentional storytelling. We sit down to watch it on purpose, and we accept its fakeness because the quality of the fiction earns our attention.

Short-form vertical video works differently. When I open an app to binge-watch videos on the couch, I’m not looking for a well-told story. I’m probably addicted, honestly, and half the time I’m doing it unintentionally. What I’m after is a short burst of entertainment, and that burst almost always depends on authenticity. The cat did something ridiculous. The kid said something unexpected. It’s compelling because it happened. Take away the “it happened” part, and you’re left with content that looks like the real thing but doesn’t feel like it. The cat is still funny, but less so. The kid’s line is still clever, but it hits different. People scroll past.

AI video may eventually get good enough to compete with intentional fiction. But Sora — the app — wasn’t trying to be a Marvel movie. It was trying to be TikTok, and in that space, authenticity is the product. I think AI-generated short-form content that everyone knows is fake has a serious demand problem, and Sora is the first major data point.

To be fair, there’s already a ton of AI-generated video on these platforms, and a lot of it is getting views. By some estimates, roughly 22% of TikTok content now has substantial AI-generated elements, and a quarter of the platform’s top search results contain synthetic imagery. But only about half of that content is labeled as AI. Most people watching it don’t know it’s AI. That’s what makes it work. Sora was the test case for what happens when you build an entire platform around content that is explicitly, knowingly AI-generated. And people left.

The opportunity is a platform where you know everything is real

It’s too early to draw definitive conclusions here, but my hypothesis is that human appetite for knowingly AI-generated entertainment, at least at current capability levels, is extremely limited. That’s what I think Sora’s shutdown actually tells you, not that OpenAI is in trouble, but that this particular bet on how people consume content was wrong.

And if that’s right, the more interesting opportunity is the inverse: a social media platform where AI content is prohibited, where you know that what you’re watching is real. I don’t know if anyone’s building this yet, but someone will, and I think they’ll do very well.

Hi, I’m Jamaal. Welcome to Jamaal’s Letter, where I share insights, predictions, and strong opinions on the future of markets, work, business, tech, and more.

The people who read this newsletter are the people who run the world. If you aren’t yet a subscriber, consider joining them by subscribing below. Please share it with your friends and colleagues if you find it helpful:

Subscribe now

Read time: 4 minutes

TL;DR:

• AI token spending is now a real line item in corporate budgets, with its own approval process and growing fast.

• A Business Insider article, prominent VCs, and OpenAI executives are framing AI token budgets as a “fourth component” of engineering compensation alongside salary, bonus, and equity.

• This framing is corporate propaganda. A tool your employer provides so you can do your job is not compensation. Your laptop isn’t. Your Slack seat isn’t. Your token budget isn’t either.

• The productivity argument — more tokens means more output means more money for you — only holds if you and your employer have truly aligned incentives, which most employees don’t have.

• Even when incentives are aligned, your employer is by definition capturing more value from that token spend than you are.

• Ask about token budgets in interviews the same way you’d ask about hardware or dev tools. Just don’t let anyone put it on your offer letter and call it pay.

I spend a lot of money on AI tokens. Both personally and professionally.

As a CFO, the bulk of that spend is token budgets I allocate to engineers who increasingly use AI in their daily work. Frontier models power our AI products. Agentic workflows handle everything from coding to fraud detection. And across the organization, employees at every level are using AI to make their own work faster and better.

So much so that the organization of AI tools and spending is becoming a meaningful piece of how I think about capital allocation and budgeting. AI token allocation now has its own line item, budgeting process, and approval workflow. It’s only getting more serious and more complex.

Silicon Valley Wants to Pay You in AI Tokens. Credible People Are Buying In.

There’s an emerging idea in Silicon Valley that engineers — or other employees — might soon be granted a token budget as part of their job offers. For those less familiar: tokens are the units AI models use to process language, where one token equals roughly three-quarters of a word. Every interaction with a model like ChatGPT or Claude consumes tokens on both sides — the input you send and the output the model generates — and companies pay for that usage. When engineers are running AI coding tools and automated workflows all day, those costs add up fast. Some people are starting to call this spending compensation.

A recent article in Business Insider reports that tech job candidates have begun asking during interviews what AI compute budget they’ll have access to if they join. Among other sources, the article cites a recent compensation submission on Levels.fyi from a software engineer that listed a Copilot subscription as part of their pay and benefits.

Tomasz Tunguz, a partner at Theory Ventures, has been the most vocal proponent of this framing. In a recent blog post, Tunguz argued that technology companies are adding a fourth component to engineering compensation: salary, bonus, options, and inference costs. He pointed out that with Levels.fyi pegging the 75th percentile software engineer salary at $375,000, adding $100,000 in inference brings the fully loaded cost to $475,000 — 21% in tokens. His punchline: “Will you be paid in tokens? In 2026, you likely will start to be.”

The piece also cites Peter Gostev, AI capability lead at startup Arena, who proposed that OpenAI and Anthropic should create recruitment sites where clients can advertise roles, listing the token budget alongside the salary range.

Your Employer’s Tools Are Not Your Compensation. They Never Were.

Many of my readers are highly compensated tech employees who collectively over-index on AI usage. I’d bet many of you are spending hundreds, maybe thousands, of dollars a month on AI tokens.

Listen closely. Do not be fooled into accepting AI tokens as compensation.

The idea that a tool used to make you more productive at work — work that ultimately benefits the organization — is part of your pay is corporate propaganda. Pure and simple.

I’ve written before about how corporate propaganda is so pervasive that we don’t even realize we’re immersed in it, and how its insidiousness has allowed it to be passed along as career advice. This framing of AI tokens as compensation is the latest version of that phenomenon.

The fact that this idea is being taken seriously and promoted by very credible people, with — as far as I’ve seen — no meaningful pushback, is absurd.

The Business Insider article discusses the ROI of token spend and how CFOs are starting to track it with the same rigor as headcount. Tunguz poses the question of productive work per dollar of inference. All fine. That’s a real budgeting conversation. But nowhere in any of this coverage does anyone stop to challenge the premise that this spending constitutes compensation. That absence is telling.

This is equivalent to telling you that the cost of the software and hardware you use is part of your pay. Your laptop is not compensation. Your IDE license is not compensation. Your Slack seat is not compensation. And your token budget is not compensation. These are costs of doing business. They are tools your employer provides so you can do the job they hired you for.

The Productivity Argument Only Works If You and Your Employer Are Truly Aligned

I know someone reading this is thinking: if I’m allocated more tokens, I could be more productive and ultimately get paid better. That logic only holds if you and your employer have truly aligned incentives. In most cases, that only happens in one of three scenarios: you own meaningful equity in the company, you get a precise and transparent share of profits, or you know the exact mathematical formula — and the objective inputs — behind how your performance bonus is calculated.

If one of those applies and you and your employer are genuinely aligned, token usage is, at best, a very indirect contributor to your compensation. And even then, by definition, your employer is making more money off of that token spend than you are. You’ll likely never have enough information about the organization’s ROI on token usage to calculate your personal ROI on more tokens. The way employer math works, the organization’s return, by definition, needs to be larger than yours. That’s how businesses work.

Ask About Your Token Budget in Interviews. Just Don’t Let Anyone Call It Pay

AI tokens might be part of your negotiation, the same way you negotiate other tools, hardware setups, or resources. Asking about your token budget in an interview is reasonable, as is asking about your laptop setup or whether the company pays for professional development. Those are legitimate questions about working conditions.

But do not confuse any of that with compensation. And do not let this idea spread without scrutiny and pushback.

Hi, I’m Jamaal. Welcome to Jamaal’s Letter, where I share insights, predictions, and strong opinions on the future of markets, work, business, tech, and more.

The people who read this newsletter are the people who run the world. If you aren’t yet a subscriber, consider joining them by subscribing below. Please share it with your friends and colleagues if you find it helpful:

Subscribe now

Read time: 7 minutes

TL;DR

• Prediction markets were originally designed to reward people with insider knowledge for trading on them. That was the entire value proposition

• Kalshi just disclosed its first public insider trading enforcement cases: a MrBeast editor and a California political candidate who bet on his own race

• Neither case would be illegal under traditional securities law, but Kalshi is treating them like securities violations anyway because trust is existential for the business

• Kalshi and Polymarket can’t afford for customers to believe the game is rigged

• The cultural definition of insider trading is shifting from earnings calls and mergers to reality TV outcomes and awards shows

• The contradiction at the center of all of this: the thing prediction markets were built to encourage is now the thing they’re spending millions to prevent

When I worked on Wall Street, I sat on the trading floor. My group was a weird hybrid of investment banking and trading. The kind of investment banking where people sit in cubicles building financial models and presentations, merged with the loud, raucous floor where people talked on phones and yelled at each other. I sat on the trading floor you’d see in movies and TV shows, but behind plexiglass walls. The plexiglass existed less as an actual security measure and more as a signal — a visible reminder of where the private side ended and the public side began. That’s what insider trading law is built on — the wall between what you know privately and what you’re allowed to act on publicly.

Often when I wasn’t actually working, I’d be trading on prediction markets. They were still in their infancy then, long before their current popularity, back when they were still illegal in the U.S., and you had to transfer money into an offshore middleman.

During this stretch — working behind the plexiglass by day and trading prediction markets by night — I read The Wisdom of Crowds, a New York Times bestseller that shaped how a generation of technologists, economists, and policymakers thought about collective intelligence. The book’s central argument was that large groups of people, under the right conditions — diversity of opinion, independence of judgment, and a mechanism for aggregation — are smarter than individual experts. He used prediction markets as a key example. The whole point was to give people with better information, whether because of expertise, local knowledge, or even early access to news, an incentive to reveal it through trading, so that the resulting price reflects the crowd’s best available judgment.

The entire value proposition of a prediction market was that people with better information — including insider information — would trade on it. On Wall Street, the plexiglass was the metaphor made physical — an attempt, however imperfect, to keep private information from leaking into public markets. In prediction markets, that information was supposed to flow freely. That was the whole design.

“Insider Trading” is Changing in Real Time

The idea of insider trading is changing as prediction markets eat up market share, investment and gambling dollars, and cultural mind share. The language of prediction markets is cannibalizing the language of financial markets.

Kalshi recently disclosed two enforcement cases that show how insider trading is already emerging on real-money prediction markets. In the first, an editor for YouTuber MrBeast used non-public information about upcoming videos to place roughly $4,000 of bets on Kalshi markets tied to MrBeast content. Kalshi’s surveillance systems flagged his near-perfect wins on low-probability outcomes. The platform froze his account, fined him about $20,000, ordered him to return roughly $5,400 in profits, and suspended him for two years. MrBeast’s company publicly emphasized a zero-tolerance stance and fired Kaptur. In the second case, a former California gubernatorial candidate wagered about $200 on a market tied to his own race and posted a screen recording of his Kalshi order on social media. Kalshi imposed a five-year ban, levied a fine equal to 10 times the original trade size, and referred the case to the CFTC.

We only know about these incidents because Kalshi proactively released information about them.

This is a reminder: most popular prediction markets do not fall in the same legal and regulatory area as financial markets. Insider information means something very different when you know the outcome of a MrBeast contest than when you know about an impending, undisclosed merger. Trading on your own gubernatorial race or on the secrets of a YouTube show isn’t illegal in the way that insider trading on securities is.

We are moving into a world where you’re more likely to hear about insiders betting on awards show outcomes, reality TV competitions, and their own political campaigns than about insiders betting on earnings calls. Lower stakes but higher volume. And in the process, the way we think about insider trading culturally is going to start to change.

Kalshi and Polymarket Are Spending Millions to Prove They’re Not Rigged

This tells me that inside these large prediction markets — Kalshi, valued at $11 billion and seeking a $20 billion valuation, and Polymarket, valued at $9 billion and seeking the same — the broader perception of insider trading is ringing a five-alarm fire. Proactively managing this perception is existential for these companies.

Here’s the tell: In a statement on LinkedIn, Kalshi CEO Tarek Mansour talks about working with law enforcement and having referred several cases to the authorities. He talks about building a proprietary surveillance system modeled on the NYSE and Nasdaq. He talks about hiring a former white-collar criminal attorney as head of enforcement and bringing on a former Under Secretary of the Treasury as an advisor.

But trading on your own gubernatorial race or on the secrets of a MrBeast show isn’t illegal. Not yet, anyway.

This is a signal to customers. They are treating this with the utmost urgency because they have to. The whole performance of enforcement — the fines, the suspensions, the referrals to the CFTC, the LinkedIn posts — is about one thing: trust. If customers believe insiders are cleaning them out, they stop trading.

No amount of internal policing, fines, or proactive investigations will change the underlying fact: making bets on these markets where insider trading is not illegal is a fool’s errand for the average participant. Not to say money can’t be made. But you will always be at a severe disadvantage compared to other market participants who know things you don’t.

Prediction Markets Were Designed to Reward Insider Knowledge

This focus on the unfair advantage of insider trading shows that prediction markets have evolved beyond their original purpose. The idea of a prediction market, as Surowiecki noted, was to elicit the best information, in part by tapping people with inside information. This evolution from an information-elicitation tool to a money-making business where mass participation is the incentive —not eliciting the best information — runs counter to the original intent.

This tension isn’t new. It’s baked into the history.

The idea of betting on events to infer probabilities dates back to 19th- and early 20th-century markets for elections and commodities, where prices implicitly encoded beliefs about future outcomes. Economists later formalized this, arguing that market prices could be read as collective probability estimates rather than just gambling odds. These early markets already relied on traders acting on whatever information they had — public or private — to move prices toward more accurate forecasts.

In the late 20th century, academics began building explicit prediction markets to test how well markets aggregate dispersed information. Corporate internal markets at companies like HP and later Google allowed employees to trade on outcomes such as sales volumes. The goal was not entertainment. It was to see whether allowing informed insiders to trade would yield better forecasts than official plans or expert committees. Over time, these internal markets often beat conventional forecasting processes.

Then came DARPA. In the early 2000s, the U.S. Defense Advanced Research Projects Agency launched the FutureMAP initiative, including the proposed Policy Analysis Market. The idea was to let traders buy and sell contracts on geopolitical events —coups, policy changes, regional instability — so that prices would reveal crowd-based probability estimates that intelligence analysts could use. The project was killed in 2003 after critics branded it a “terrorism futures” market. But the important intellectual legacy is that DARPA explicitly tried to harness potentially insider or expert information — held by analysts, regional specialists, or people close to the ground—as the feature, not the bug. The U.S. intelligence community understood that the value of prediction markets lay precisely in letting people with asymmetric information trade on them.

The Thing Prediction Markets Were Built to Do Is Now the Thing They’re Trying to Prevent

Today’s policy concern is that prediction markets let insiders with material, non-public information profit at the expense of less-informed participants, raising fairness questions that look a lot like insider trading debates in securities markets. Platforms and regulators are now experimenting with bans on certain “politically exposed” traders, surveillance of suspicious flows, and other measures to limit this advantage.

But go back to the intellectual roots — from early wagering markets, through corporate internal markets, to DARPA and Surowieck — and the raison d’être of prediction markets is precisely to encourage those with superior, even asymmetric, information to trade so the price reflects that information. The very mechanism that made prediction markets attractive as forecasting tools is what now gets redescribed, under today’s regulatory lens, as an “insider trading problem.”

In other words, prediction markets were designed to do the thing that prediction market companies are now spending millions of dollars to prevent.

The question is whether a prediction market that successfully stamps out insider trading is still a prediction market at all, or just a casino with better branding.

Hi, I’m Jamaal. Welcome to Jamaal’s Letter, where I share insights, predictions, and strong opinions on the future of markets, work, business, tech, and more.

The people who read this newsletter are the people who run the world. If you aren’t yet a subscriber, consider joining them by subscribing below. Please share it with your friends and colleagues if you find it helpful:

Subscribe now

Read time: 8 minutes

TLDR: I'm not writing about the war in Iran, I'm writing about how to read about it. Simplistic narratives are wrong, visual content can no longer be trusted, every source has an agenda, English-language coverage is filtered, and not all sources are equal. I rank them from worst to best: social media at the bottom, government sources above that (they're propaganda with polish), mainstream outlets in the middle (credible but milquetoast), and niche domain experts at the top.

I was a freshman at the University of Missouri when 9/11 happened. I heard the news while walking back from an 8:00 a.m. Spanish class. I was a sophomore when we invaded Iraq. I remember watching on the small television in my dorm room what the president dubbed “shock and awe.” I was in a red state, a right-of-center media environment — well before social media and podcasts — listening to local conservative talk radio in the background often. I was also studying at the world’s first and best school of journalism, learning about media independence and journalistic process from Pulitzer Prize winners and former war reporters.

That experience — consuming a war through talk radio while learning how journalism actually works — gave me an early sense of how wide the gap can be between what people are told and what’s actually happening. That gap has only gotten wider.

Right now, Israeli and American bombs are dropping on Tehran. Iranian bombs are dropping across the Middle East. This is one of several hot conflicts happening across the globe as I write this. I have many friends, acquaintances, and colleagues who are experts on war and geopolitics in the Middle East, and I’m part of half a dozen group chats with some of the brightest minds in the world on these subjects.

I built a digital media business reporting on the economic realities that overlay this type of conflict — work that taught me how information moves across borders and who controls the narrative when the stakes are high. In 2023, I traveled across Europe on the periphery of the then less-than-two-years-old Russian invasion of Ukraine, talking to activists addressing Europe’s misinformation problem and meeting with organizations on the literal borders of the war. That experience sharpened something I’d already believed: the information environment around a conflict is its own battlefield, and most people walk onto it unarmed.

This isn’t a piece about the war itself. It’s about something I’ve spent my career thinking about: how to consume information when the stakes are this high.

Beware the overly simplistic story

If someone is telling you a conflict is pure good versus evil, they’re telling you a story, not giving you information. This is not a Marvel movie. There is no supervillain and no hero who saves the day in the third act. Real geopolitical conflicts involve rational actors with competing interests, historical grievances, and internal politics. That doesn’t mean both sides are morally equivalent. It means if someone is telling you the other side is purely irrational — motivated only by destruction — that person is either propagandizing you or doesn’t understand the situation.

The “they just hate us” framing is almost never true. It’s a sign you need a better source.

Think back to the lead-up to the Iraq War. The narrative was streamlined into something digestible: weapons of mass destruction, a dangerous dictator, a clear threat. Nuance about sectarian complexity, intelligence credibility, or the consequences of regime change got bulldozed. The people asking harder questions turned out to be closer to right. The simple story was easier to sell, but wrong on the things that mattered most.

You can no longer trust what you see

Seeing is no longer believing. Recycled footage from old conflicts gets passed off as current: clips from Syria labeled as Libya, Ukrainian footage repurposed for a different war. In 2023, footage from a video game circulated widely as real combat footage from the Israel-Gaza conflict, convincing enough to fool millions scrolling on their phones.

Now add AI-generated imagery. Deepfakes, synthetic video, fabricated photos of atrocities that never happened. The technology deceives most people at the speed they actually consume content, fast, on a small screen.

If you see something jarring, emotional, or explosive on social media without a clear, verified source, assume it’s unverified until proven otherwise.

Every source has an agenda — figure out what it is

Every piece of information released during a conflict was released by someone for a reason. Governments have media strategies. Militaries run information operations. Insurgent groups, NGOs, and even humanitarian organizations have communication goals.

When you encounter a claim, ask two questions. Who released this, and what do they gain from me believing it? And how independent is this source, who owns this outlet, funds this think tank, employs this analyst? Is the source structurally connected to one of the parties in the conflict?

Dependence doesn’t automatically mean the information is wrong. A state-funded outlet can publish accurate reporting. An analyst with ties to one side can still have genuine expertise. But understanding independence gives you critical context for evaluating what you’re being told. You weigh information differently when the source has skin in the game.

Long-form debate is one of the best ways to learn during a conflict

There’s a growing media ecosystem built around debate — pro-Gaza versus pro-Israel, pro-Russia versus pro-Ukraine, anti-war versus hawk — and watching two credible, knowledgeable people argue is one of the best ways to sharpen your understanding of a conflict.

The caveat: the debate has to be long-form. An hour or more. The kind where claims get tested in real time, where someone can say “that’s not quite right, and here’s why.” Not the cable news version where two people get ninety seconds to shout past each other.

Organizations like Open to Debate are doing this well. They partnered with the Council on Foreign Relations for a debate on whether Congress should stop funding the war in Ukraine, featuring political scientist John Mearsheimer and retired Lt. Colonel Daniel Davis against former German Marshall Fund president Heather Conley and former Undersecretary of State Paula Dobriansky. That’s the kind of format where you actually learn something, because each side’s strongest arguments get stress-tested by the other.

Even in less polished settings — internet personalities debating on YouTube or podcasts, with varying levels of credibility and expertise — the format itself has value. The quality of the individual debaters matters less than the structure: when opposing ideas collide, weaknesses in arguments get exposed that you’d never see in a one-sided presentation. A brilliant argument can be completely convincing in isolation and full of holes when someone pushes back. The opposing side compensates for what any single source is missing. Long-form debate is one of the few formats that lets you see that happen.

If you’re only reading English, you’re getting the export version of reality

I’ve had a dozen experts on Middle East conflict tell me some version of the same thing: if you want to understand what’s really going on, pay attention to what leaders and organizations say in their own language to their own people, not what they say in English to an international audience.

That principle applies far beyond the Middle East. Governments and political leaders routinely say one thing to an international audience and something very different at home. If you’re consuming coverage exclusively in English about a conflict happening in a non-English-speaking region, you’re getting the export version of reality. Seek out translated sources, bilingual journalists, and analysts who work in the local language. The gap between what’s said in English and what’s said locally is often where the most revealing information lives.

Not all sources are equal. Here’s how to rank them

Not all information sources are equal during a conflict. Think of this as a ladder — each level has specific strengths and specific failure modes.

The worst: social media

The floor. Unsourced, unverified, algorithmically amplified content optimized for engagement, not accuracy. Recycled footage, AI-generated images, propaganda from state and non-state actors — it all lands here first. The incentive structure rewards speed and emotional intensity. Nobody gets rewarded for saying “I’m not sure yet.”

Treat everything here as unverified until confirmed by a higher-tier source.

Second worst: government and institutional sources

Pentagon briefings, IDF spokespeople, Iran’s IRNA and Press TV, Russia’s TASS, Chinese state media like Xinhua, State Department press conferences, UN reports. These carry an aura of authority that makes them more persuasive than media, and that’s exactly what makes them dangerous.

A government involved in a conflict is giving you its side. Full stop. These are sophisticated communications operations designed to shape how you understand what’s happening. The press conference looks neutral. The spokesperson sounds measured. The language is careful and official. This is propaganda from one of the parties involved. Whatever facts are present have been selected, framed, and sequenced to advance a position, not to inform you.

Better: major mainstream outlets

The New York Times, BBC, The Guardian, Washington Post, Reuters, AP, Der Spiegel, Le Monde, Al Jazeera English, Haaretz, The Economist. These organizations have real journalistic standards — fact-checking, editorial oversight, correspondents on the ground. They are meaningfully better than social media or government sources.

But they have structural limitations. They cast such a wide net that the reporter covering this conflict was probably on a different beat last month. They rely heavily on official sources — government spokespeople, military briefings — which makes them susceptible to whatever narrative the most media-savvy party is pushing. And the institutional process — layers of editors, legal review, organizational caution — tends to sand down the sharpest edges. The reporting will be credible and follow sound journalistic process, but it will rarely be especially provocative, novel, or predictive. The bureaucracy of a major newsroom whittles away sharp takes, bold predictions, and the kind of pointed analysis that might get ahead of the story. You get solid, reliable, somewhat milquetoast coverage.

Think of mainstream outlets as reliable for the what but often incomplete on the why.

The best: niche sources and domain experts

This is where real understanding lives. The historian who wrote the book a decade ago and predicted the dynamics now playing out. The journalist who spent their entire career covering this specific region. The former diplomat who served in that theater for fifteen years. The niche publication based in the country where the conflict is happening, staffed by people who grew up there.

These sources trade in context, not breaking news. They won’t tell you what happened in the last hour, but they’ll help you understand what it means and what comes next. Building a roster of these people is one of the most valuable things you can do for your information diet.

Hi, I’m Jamaal. Welcome to Jamaal’s Letter, where I share insights, predictions, and strong opinions on the future of markets, work, business, tech, and more.

The people who read this newsletter are the people who run the world. If you aren’t yet a subscriber, consider joining them by subscribing below. Please share it with your friends and colleagues if you find it helpful:

Subscribe now

Read time: 7 minutes

TL;DR:

• Margins, not machines: Block is cutting costs to get leaner, like every other tech company since 2022. The company tripled headcount during COVID and is now unwinding that. Dorsey admitted to overhiring.

• Profitability doesn’t protect you anymore: Half the most profitable companies on the planet did mass layoffs last year. The market rewards it. AI is just the most flattering excuse available.

• The stock move is arithmetic: Cut ~$880M in annual payroll, multiply by Block’s earnings multiple, and you get roughly a 20% bump. That’s what happened. No AI thesis required.

• The severance was decent, the structure was smart: Good financial terms and one big cut instead of death by a thousand rounds. Better for morale, retention, and execution.

Block, the fintech company behind Cash App and Square, laid off roughly 4,000 people this week, cutting its workforce from over 10,000 to just under 6,000. Within hours, a predictable crowd showed up to declare the beginning of the AI jobpocalypse. People are falling over themselves to say it. There’s no evidence of it. Many of the people saying it don’t believe it themselves. Many others are just naïve. Even if an AI-driven labor displacement is coming at some point, this particular event is not a sign of it.

Here’s why:

Block is laying people off to increase margins and increase efficiency, like every other big tech layoff of the last three years

There’s a meaningful difference between “AI is taking jobs” and “we want to be more lean and efficient…in the age of AI.” You could remove the “age of AI” part from Block’s announcement, and the logic would hold up just fine. The push for efficiency across tech is driven by COVID-era overhiring, increased competition, greater employer leverage over employees, higher interest rates, and pressure from activist investors. This dynamic has been playing out since 2022. I don’t think it has much to do with AI.

The overhiring numbers tell the story clearly. Block had about 3,900 employees at the end of 2019. By December 2022, that number had ballooned to roughly 12,500. The company more than tripled its headcount in three years, fueled by cheap capital and a pandemic-era boom in digital payments. Even after previous rounds of cuts in January 2024 (about 1,000 people) and March 2025 (931 people), Block still had over 10,000 employees heading into this week. The new target of just under 6,000 puts Block closer to where it probably should have been all along. Dorsey admitted as much. When challenged on X, he responded: “yes we over-hired during covid because i incorrectly built 2 separate company structures (square & cash app) rather than 1, which we corrected mid 2024.” He also acknowledged taking on complexity through lending, banking, and buy-now-pay-later services.

Wharton associate professor Ethan Mollick, one of the most credible voices in the AI research space, put it plainly: “given that effective AI tools are very new, and we have little sense of how to organize work around them, it is hard to imagine a firm-wide sudden 50%+ efficiency gain that justifies [massive] organizational cuts.” This is the right instinct and one I’ve previously written about:

The conversation around AI is almost entirely about efficiency — fewer workers, lower costs, wider margins. You don’t hear anyone saying AI will double revenue, triple earnings, or dramatically expand capacity. It’s all “headcount reductions” and “lower costs,” as opposed to “hiring sprees” and “revenue explosions.”

While leaders talk expansively about AI’s future potential, no one is rewriting forecasts today. No one is saying, “Throw out our old numbers, AI will blow them away.” Instead, they’re saying the opposite: AI will help us do more with less.

That blatant asymmetry of how these people talk is a tell. It’s a tell that AI “efficiency” is a smokescreen for corporate retreanchment in a market economy of unprecedented uncertainty.

I’ll concede one thing: Block’s cuts are larger as a percentage of workforce than what we’ve seen from most other tech companies in this cycle. Cutting 40% of your people in a single round is aggressive by any standard. Microsoft’s 15,000 layoffs last year represented about 3% of its workforce. Meta’s early-2025 cuts were under 5%. Block went much further. But the scale of the cut maps to the scale of the overhiring, not to the scale of AI’s impact.

Profitable companies now routinely do layoffs. Poor performance is no longer a prerequisite.

Many people bolstered their AI claims by mentioning that the company is profitable. That is naive. Gone is the era when something had to go wrong for layoffs to happen. This is a structural shift that has nothing to do with AI, and the people using Block’s cuts to tout the coming AI jobpocalypse are either missing this or ignoring it.

Block’s net income surged 150% year-over-year in Q4 2025, reaching around $280 million. Full-year net income topped $3 billion. This is not a company in distress. And that’s the point. In 2025, layoffs at profitable companies became standard operating procedure. Half of the eight most profitable companies on the planet engaged in mass layoffs last year:

• Microsoft laid off roughly 15,000 employees last year while posting record profits.

• Meta cut 3,600 people at the start of the year while revenue grew 27%.

• Amazon reported surging profits and still trimmed tens of thousands of roles, while Google followed the same playbook.

The market rewards these cuts. It has been rewarding them since at least 2022. Investors no longer see headcount reductions as signals of a crisis; they see them as a sign of capital discipline. That has created a feedback loop where CEOs are incentivized to run leaner organizations, whether or not there’s a crisis to justify it. Dorsey is operating within that loop, and AI is the most flattering narrative to explain what is, at its core, an efficiency-and-margin play.

The stock moved higher because that’s how math works, not because of AI

Block’s stock initially jumped more than 20% on the news of the layoff, adding roughly $6 billion to its market cap. Many people saw that move and interpreted it as the market endorsing AI-driven workforce transformation. It isn’t. The math is simpler than that.

Cutting about 4,000 roles likely removes somewhere in the range of $700 to $900 million in annual compensation and related costs, assuming average compensation of $200k-$250k:

• 4,000 laid-off employees x $220k average total pay per year = ~$880 million saved each year on payroll costs.

• Of that savings, about 75% flows directly to higher annual profit, adding ~$660 million to the bottom line.​

• Multiply those extra profits by Block’s current 23x earnings multiple, and you get $15.2 billion more company value.

• On Block’s pre-announcement value of around $75 billion, that jump equals a 20% rise in stock price, matching what investors did after the news

That kind of recurring operating expense reduction, capitalized at Block’s current earnings multiple, implies several billion dollars of incremental equity value. That’s exactly what you’d expect if investors believe most of the savings will stick and won’t be offset by revenue damage. The stock price rose because Block removed a large, recurring line item from its cost structure. That is arithmetic, not ideology.

I’ve heard people arguing that other companies will now make similar moves to push their own stock prices higher. That may well be true. But that same logic and math would apply to those companies, regardless of AI. Cutting costs raises margins. Raising margins raises valuations. This was true in 2015, and it’ll be true in 2035.

Zoomed out, the stock’s history makes the move look more like catch-up than a victory lap. Over the past year, heading into this announcement, Block had roughly matched or modestly outpaced the S&P 500. Over five years, Block remains deeply negative and far behind both the S&P 500 and the Nasdaq.

As far as layoffs go, these were relatively good ones

Block’s severance for affected U.S. employees includes 20 weeks of base pay plus one additional week per year of tenure, full vesting of equity through the end of May 2026, six months of employer-paid health insurance, retention of company-issued equipment like laptops, and a $5,000 cash payment for relocation or other transition expenses (with local adjustments for international employees).

Companies don’t offer generous severance because they’re kind. They do it so people won’t sue. That’s the reality. But measured by that standard, Block’s package is relatively good.

Another thing Block got right structurally: doing it all at once. Dorsey chose a single large round of cuts instead of the drip, drip, drip of smaller successive layoffs over months or years. Whether that’s the more “humane” approach is a debate I’m not going to have here. Moralizing layoffs is a conversation for another day. But from a pure organizational effectiveness standpoint, one clean cut is almost certainly better for the remaining employees, for continuity, and for morale. Repeated rounds of layoffs exacerbate anxiety, as nobody can focus because everyone is waiting to see if they’re next. Block had already done two previous rounds of cuts (January 2024 and March 2025), and Dorsey explicitly said he didn’t want to keep doing that. That instinct is correct. A company in perpetual layoff mode can’t retain its best people, can’t maintain institutional knowledge, and can’t execute on whatever the new strategy is supposed to be. Ripping the bandage off hurts more in the moment, but the organization recovers faster.

Hi, I’m Jamaal. Welcome to Jamaal’s Letter, where I share insights, predictions, and strong opinions on the future of markets, work, business, tech, and more.

The people who read this newsletter are the people who run the world. If you aren’t yet a subscriber, consider joining them by subscribing below. Please share it with your friends and colleagues if you find it helpful:

Subscribe now

Read time: 5 minutes

We are two and a half weeks removed from the Super Bowl, where the world’s biggest musical artist performed in Spanish, causing absurd and ideologically inconsistent backlash from a bunch of political conservatives.

I found the halftime show to be just OK. I’m not really a Bad Bunny fan. I do like “NuevaYol,” the video of which was shot partially on one of the campuses of the City University of New York, where I taught for a number of years. As someone who spent roughly 15 years in New York City, eight of which were in Spanish Harlem and the South Bronx, some of the halftime show visuals felt quintessentially NYC.

Among the nonsensical list of grievances I heard about the show was this notion that football is inherently American, and that the halftime show during football’s biggest game should follow suit.

I’m not surprised that the NFL, like everything else, has become a political Rorschach test. Everything is political now. What I find troubling is how effectively corporate propaganda has used patriotism to hide its motives.

Rather than seeing the NFL and its halftime show for what they are—the signature viewing event of a corporation looking to further expand its customer base outside of the United States—most of us have been tricked into thinking of football, and much of pro sports generally, as some sort of public good in which we have a civic stake. We don’t. The NFL is a private business. You are its customer, not its constituent.

The NFL has largely maxed out domestic growth. Traditional TV audiences have softened or plateaued in recent years, even as revenue keeps rising, so the league is leaning on more international games, commercial rights, and localized marketing in Europe, Latin America, and other markets as its next growth engine.

That international push is increasingly urgent because the sport’s violence and mounting CTE concerns threaten its long-term talent pipeline and social license at home, even as live games remain one of the few must-watch-now products in an on-demand world that is otherwise chipping away at the pricing power of U.S. TV sports rights.

A Spanish-language halftime show at the Super Bowl isn’t a cultural statement; it’s a customer acquisition strategy for Latin American and European Spanish-speaking markets. The fact that anyone connects football to American values or culture is a testament to how good the NFL’s marketing has been and how naive we are about corporate motives and influence.

Consider that no one is complaining that Netflix, an American company, now has non-English-language originals accounting for the majority of its new TV releases. Nobody is complaining that Disney, also American, relies heavily on locally produced titles for Disney+ in markets like India, Korea, and Latin America rather than defaulting to U.S.-made, English-language shows.

I’m not naive. I know why. The NFL, along with its Super Bowl and halftime show, occupies a different cultural space than Netflix and Disney. But economically, there’s little difference. All three are private American companies pursuing international revenue growth. The only reason we expect the NFL to perform Americanness while giving Netflix and Disney a pass is that the NFL has spent decades wrapping itself in the flag, and we bought it.

The Olympics Aren’t a Public Good. They’re a Television Production.

This confusion isn’t limited to football. The Winter Olympics just ended, and the two-week event featured several U.S. Winter Olympians pulled into patriotism controversies after using press conferences and social media to criticize ICE raids, attacks on immigrants, and setbacks for LGBTQ rights. Freestyle skiers Hunter Hess and Chris Lillis spoke about feeling “mixed emotions” and being “heartbroken” about representing the U.S. given current events. Figure skater Amber Glenn spoke about the struggles facing the LGBTQ community. President Trump called Hess “a real Loser” on Truth Social.

Unsurprisingly, a range of political controversies surrounded these statements, with the usual suspects charging the athletes with a lack of patriotism, ingratitude, and failure to appreciate their privilege. Politically, that argument is weak, but I’m more focused on its economic logic, or lack thereof.

If you didn’t know any better, you might think that the Olympics was some sort of inherently public good, owned by its member states like the United Nations or the World Bank, and that the apparatus that oversees selecting and sending athletes to the Games is federally funded. You’d be wrong. The Olympics is run by the International Olympic Committee, a private nonprofit based in Lausanne, Switzerland, that has only a loose affiliation with any government. The U.S. Olympic and Paralympic Committee is also a private entity.

The Olympic Games of today are not the ancient Greek religious festivals you might have learned about in school. The broad commercialization of the games exploded in the 1980s and has since generated billions from exclusive deals and broadcast rights.

That’s the relevant history: a steady arc from religious ritual to commercial spectacle. The Olympics is, today, largely a television show put on by a private organization. When people invoke patriotism to police what athletes can and cannot say, they’re conflating civic duty with entertainment consumption.

This is also a good time to remind you that the athletes who competed at the Olympics aren’t paid by the IOC. A study by Global Athlete, using IOC financial reports from 2013-2021, found that athletes receive only about 4.1% of the Olympic Movement’s roughly $3 billion in average annual revenues across all forms with just 0.5% of IOC funds going directly to athletes in any cash-like form before intermediaries take their cut.

So when a senator or a cable news host tells an Olympic athlete to shut up and be grateful, ask yourself: grateful for what, exactly? The privilege of subsidizing a multi-billion-dollar television production with years of unpaid training?

The Name for What’s Happening Here Is “Patriotism Washing”

What the patriotism police of both the Super Bowl and the Olympics are doing is confusing inherently public goods—civic activities, democratic institutions—with entertainment products created by private organizations for profit. This corporate propaganda has been so effective that most people don’t even realize they’ve internalized it and are peddling it.

There’s a term for governments using sports strategically to improve their reputation, distract from controversy, or legitimize power: sportswashing. Saudi Arabia hosting major boxing events and golf tournaments. Qatar hosting the World Cup. Russia hosting the 2018 World Cup and the 2014 Winter Olympics. The playbook is familiar.

What I’m describing is the inverse. I’d call it patriotism washing: corporations strategically wrapping their products in national identity to make consumers feel a civic obligation toward what is, at bottom, a commercial transaction. They blur the line between entertainment product and public good so effectively that consumers feel they have actual national ownership over a product they merely watch. The NFL doesn’t belong to you. The Olympics don’t belong to your country’s citizens. They belong to the organizations that own and operate them for profit. Your relationship to them is as a viewer—a customer—not as a stakeholder.

The next time someone tells you the Super Bowl halftime show should be in English because football is American, or that an Olympic athlete should keep quiet and be grateful, remember what’s actually happening. A corporation sold you the idea that its product is your birthright, and you believed it enough to enforce its brand guidelines for free.

Hi, I’m Jamaal. Welcome to Jamaal’s Letter, where I share insights, predictions, and strong opinions on the future of markets, work, business, tech, and more.

The people who read this newsletter are the people who run the world. If you aren’t yet a subscriber, consider joining them by subscribing below. Please share it with your friends and colleagues if you find it helpful:

Subscribe now

Read time: 5 minutes

When I was in grad school at Harvard, I had a European classmate who loved to needle Americans about one thing in particular: our obsession with optimization. He’d watch us agonize over course selection strategies, recruit for jobs with ruthless efficiency, and map out our post-graduate lives like five-year military campaigns, and then he’d just ask, with genuine bewilderment: “Why are you always trying to optimize everything?”

I didn’t have a good answer. Not because I couldn’t articulate one, but because the question barely registered. To me, optimization wasn’t a choice. It was the default setting. Obviously you should maximize outcomes. Obviously you should pursue the highest-return path. The idea that optimization itself might be a problem — that it might reflect something broken rather than something admirable — simply didn’t compute. So our debates on the topic were never particularly sophisticated. I was arguing from a position I’d never actually examined.

I’ve thought about those conversations a lot over the years. And they came flooding back recently when I stumbled onto an article in the Stanford Review that makes a provocative claim: getting into Y Combinator—the world’s most successful startup accelerator—is a move for cowards.

The piece, written by Stanford freshman Bassel Ojjeh, argues that the culture around startups today—and YC in particular—reflects a broader decline in courage among Stanford students. What was once a daring pursuit of bold, world-changing ideas has devolved into a calculative pursuit of prestige and safety. The article hits on a few key themes: a campus culture of risk aversion, a shift from courage to optimization, the transformation of YC from contrarian incubator into a prestige-brand safety net, and broader institutional pressures like grade inflation that insulate students from the embarrassment of meaningful failure.

I’d never heard of the Stanford Review before this article surfaced. I never encountered it during my time on campus, largely because it’s a right-of-center publication staffed mostly by undergrads. (It was founded by Peter Thiel in 1987, which tracks.) Nonetheless, it’s a genuinely interesting outlet. Despite not agreeing with many of the pieces I skimmed, I’m always struck by how intelligent, thoughtful, and eloquent college students can be. The writers at the Review are no exception.

A bit about where I’m coming from: I went to Stanford for business school. I’ve known many YC founders and executives, and I’ve led investments in a number of YC-backed companies—some successful, and even more that were not. So when I read Ojjeh’s piece, a bunch of thoughts immediately popped into my head.

Here’s my reaction.

YC Is a Just A Tool.

At its core, Y Combinator is a tool to help entrepreneurs build companies. An exceptionally successful one, sure, but a tool nonetheless. The people applying to YC are starting companies. The attractiveness of the program is a testament to its track record and its branding. But ultimately, these people are building.

And building a company is an immensely difficult task. Full stop.

Now, if YC feels like a soft landing after graduation—a safe intermediate step, a credential to collect, a line on the résumé—rather than a means to the end of building an influential organization, then I can see how it might be a place for cowards. If someone isn’t motivated by the actual work of building a company, and instead sees YC as a prestigious hedge against the uncertainty of their twenties, then the critique has teeth.

But if you’re actually trying to build a company? I don’t see how using the best available resources makes you a coward.

The Myth of the Inevitable Startup

Part of why entrepreneurship feels sexy right now is a narrative problem. Successful startup stories seem inevitable and intentional in hindsight. Name a successful company and the founding story probably reads like it was always going to work out. The scrappy garage. The visionary founders. The pivot that changed everything.

The truth is that the journey is far less certain and far more messy than the legend suggests. Building companies is not glamorous. People just think it is when the story gets told in reverse, with all the chaos and doubt edited out.

It’s also far less intentional than people imagine. Plenty of successful companies started as someone building something small and figuring it out along the way. The mythology of the startup as a courageous, fully-formed act of genius is just that—mythology. If YC helps founders navigate the unglamorous middle, that’s not cowardice. That’s pragmatism.

The Job Market Elephant in the Room

It’s hard to read this article without thinking about how brutal the white-collar job market is right now—and how that context fuels this entire narrative. If some graduates are treating YC as a crutch, they’re doing so in part because their traditional employment options have narrowed considerably. When the conventional paths dry up, alternative paths start to look more appealing, regardless of motivation.

That doesn’t excuse a lack of genuine entrepreneurial drive. But it does complicate a simple framing of YC applicants as cowards. The economic backdrop matters.

The Bermuda Triangle of Talent—Plus One

Embedded in Ojjeh’s article is a critique that maps neatly onto a concept I’ve thought about for years: the “Bermuda Triangle of talent.” This describes the phenomenon where top university graduates get funneled into careers in consulting, finance, and corporate law—three industries that absorb a disproportionate share of elite human capital.

The implication of the Stanford Review piece is that YC has become a fourth leg of this talent stool. And that’s an implication worth sitting with.

I’ve reflected on this phenomenon a lot in the almost two decades since I left investment banking and found my way to elite graduate education. People go into these professional services jobs for obvious reasons: they offer tremendous optionality, high compensation, and relative stability compared to other fields.

But there’s another reason that doesn’t get discussed enough: these jobs let you see the big picture.

I’m biased, but my brain is far more engaged by big, existential questions than by narrow operational problems. Most jobs want you to hyper-focus on something small—one product feature, one customer segment, one quarterly metric. Professional services jobs in finance, law, and consulting allow you to skip that. You don’t hire McKinsey or Goldman Sachs to write copy for a marketing email or figure out why your churn rate ticked up 5% this month. You hire them to help you raise a billion dollars or acquire your largest competitor.

That ability to focus on the most important, highest-priority strategic issues is deeply attractive to a certain kind of mind. And if YC is increasingly drawing from the same pool of talent that gravitates toward consulting, banking, and law, the question isn’t just whether those founders are cowards. It’s whether the entire system—from elite education to elite career paths—has been optimized for prestige over purpose.

That’s a much bigger conversation. And it’s one worth having.

Hi, I’m Jamaal. Welcome to Jamaal’s Letter, where I share insights, predictions, and strong opinions on the future of markets, work, business, tech, and more.

The people who read this newsletter are the people who run the world. If you aren’t yet a subscriber, consider joining them by subscribing below. Please share it with your friends and colleagues if you find it helpful:

Subscribe now

Read time: 5 minutes

TLDR:

• I traded prediction markets in the earlys 2000’s on Intrade.com, back when they were offshore, sketchy, and illegal in the U.S. Ironically, I haven’t touched them now that they’re mainstream.

• Prediction markets have exploded, going from under $100M in monthly volume in early 2024 to over $13B by late 2025.

• A viral CNBC interview with Kalshi’s CEO sparked a debate after anchors pressed him on whether people with inside knowledge—like backup dancers who know the set list—can legally trade on Super Bowl markets.

• The internet is having the wrong argument. People are debating whether trading on insider information in prediction markets is illegal. It almost certainly isn’t. These aren’t securities, and halftime show intel is nothing like material non-public information under securities law.

• The real debate is about fairness, not legality. Someone will always know who’s performing at the Super Bowl before it happens. If you don’t have that information, why would you trade that market for anything other than entertainment?

• Not all prediction markets have this problem. Elections and sporting events are genuinely uncertain. Nobody knows the outcome in advance. But markets built around “knowable” events are fundamentally different, and that distinction keeps getting lost.

I’m an OG prediction market trader—back before they were popular, legal, or cool.

Starting in 2006, while working my first job on Wall Street, I traded tens of thousands of dollars on prediction markets. At the time, I used Intrade.com, an offshore platform based in Ireland, to trade mostly politics and a little sports. I was good at trading politics and bad at trading sports. I made a bunch of money betting on the 2006 midterm elections and subsequently lost it all, making in-game bets on NFL games. Because prediction markets weren’t legal in the United States back then, you had to use weird, sketchy third parties to transfer your money offshore just to participate.

Interestingly, I’ve yet to trade the current prediction markets. Which is ironic because, unlike before, they are now popular, legal, and cool.

Prediction markets grown significantly since then, with most of that growth happening in the last few years. Monthly trading volume went from under $100 million in early 2024 to over $13 billion by late 2025. Leading platforms like Polymarket and Kalshi together handled tens of billions of dollars in volume last year and are still growing at double-digit monthly rates into 2026.

As they’ve grown, they’ve also come under regulatory oversight. More money means more legal and regulatory scrutiny.

That scrutiny was the subject of a fascinating interview on CNBC yesterday. Kalshi is one of the leading prediction market platforms, and its CEO, Tarek Mansour, sat down with anchors Andrew Ross Sorkin and Becky Quick.

Mansour shared some eye-popping numbers: Super Bowl trading on Kalshi exceeded $1 billion, up roughly 2,700% year over year. A single market on Bad Bunny’s opening song alone crossed $100 million in volume.

Then things got interesting. Sorkin and Quick pressed Mansour on whether people with inside knowledge—like backup dancers who know the set list—could legally trade on those markets. He stressed that Kalshi is a CFTC-regulated exchange with Nasdaq-style surveillance and enforcement tools.

The clip — which gets interesting at 2:00 — went viral online, with people largely framing Mansour as being “uncomfortable” and “unprepared” to address the insider trading issue. That framing, unsurprisingly, misses the point. I don’t think that’s what happened at all. He actually seemed to have a pretty good understanding of where Kalshi sits within the law.

Here’s what I think is really going on: people are having a legal debate when they should be having a market fairness debate.

Take the example of someone who knew in advance about Bad Bunny’s opening song. I find it hard to believe that trading on that knowledge would ever be considered illegal. These prediction markets are not securities. And the type of information involved—what songs are being sung, which guest performers will show up, etc.—is so unrelated to the kind of material non-public information that governs securities law. The law simply doesn’t apply in the way people seem to think it does.

What people are actually debating is the fairness of trading in a market where some people have access to insider information, and where the law doesn’t make trading on that information illegal.

In other words, it’s probably never going to be illegal to know information about a Super Bowl halftime performance and trade on it. But people who do have that knowledge will always be able to use it to their advantage in these markets.

So the real question is: why would you ever participate in that type of market if you don’t have inside information?

The obvious answer is entertainment. But beyond that—if you’re trying to make real money or get a high return on your investment—I can’t see a reason why you would.

In my brief online discussions about this, some of the people I was talking with seemed overly emotional about it. I think many people are coming to this conversation with preconceived notions and biases around prediction markets. If they don’t like prediction markets, they’re trying to contort an argument for why trading on insider information should be illegal. If they love prediction markets, they’re dismissing legitimate fairness concerns entirely.

But here’s an important distinction that gets lost in the noise: this insider information problem really only applies to events that aren’t truly unknowable. What I mean is that someone always knows, with certainty, the set list of a Super Bowl halftime show.

This is fundamentally different than betting on the outcome of an election or a sporting event. Those outcomes aren’t knowable in advance. Even if you have inside information—polling data, injury reports, campaign strategy memos—you’re still just making your best guess about the future. Nobody knows the result before it happens.

Prediction markets are here to stay, and they’re growing faster than almost anyone expected. But not all prediction markets are created equal. The ones built around events where someone always knows the answer in advance—like a halftime set list—are fundamentally different from the ones built around genuinely uncertain outcomes. Until platforms and regulators figure out how to address that distinction, the smartest move for most people trading those “knowable” markets is simple: trade for fun, not for profit. Because someone on the other side of that bet already knows the answer.

Hi, I’m Jamaal. Welcome to Jamaal’s Letter, where I share insights, predictions, and strong opinions on the future of markets, work, business, tech, and more.

The people who read this newsletter are the people who run the world. If you aren’t yet a subscriber, consider joining them by subscribing below. Please share it with your friends and colleagues if you find it helpful:

Subscribe now

Read time: 3 minutes

I built and sold a digital media business over a decade ago. I’ve sat on the boards of several news organizations since. Here are some thoughts on today’s news at the Washington Post.

The Post laid off roughly one-third of its staff, hundreds of positions across newsroom, sports, books, international, audio, metro, and business operations. The standalone sports department and books section are gone. The daily podcast is canceled.

I have four immediate thoughts:

1. This Was Inevitable

This was inevitable in the modern corporate environment under private ownership. It would have happened under any private owner, whether a family whose business is inextricably tied to the Post, or a larger owner for whom it’s one asset in a portfolio. Whether that owner is Jeff Bezos, a private equity firm, or a larger media company.

It’s important not to understate the political kowtowing that the Post’s current owner has shown to the current president. But if the broader corporate dynamics here surprise you, you’re naive about how the corporate world works now.

Let me remind you: four of the eight most profitable companies on the planet engaged in mass layoffs last year. This is how the game is now played.

2. Two Larger Trends Are Converging

• The decline of traditional media. Traditional media has been declining for decades as digital media rises. Organizations like the Post have faced an innovator’s dilemma in transitioning to a digital-first world. The Washington Post still prints and delivers more than 650,000 physical newspapers per week. That tells you something. This isn’t to say traditional media can’t be profitable in the digital age. The New York Times has done it. But the economics are far less forgiving, and today’s corporate environment has no patience for outlets still figuring it out.

• The decline of institutions. This is happening across all sectors of society. The internet inherently shifts power from large institutions to small ones, and from small ones to individuals. It’s playing out not only in journalism but in entertainment, finance, and technology, albeit at different paces and in different ways.

3. New Media Cannot Replace What We’re Losing

I’ve heard people ask, "Why is this necessarily bad?" Why hold on to traditional media outlets when we have new media?

I don’t think people understand what’s at stake.

New media is growing and proliferating exponentially. But it does very little of what institutions like the Post do. Most of new media is not journalism. Some of it is, and some of it is excellent, but the percentage that qualifies as journalism is small, and even smaller when weighted by audience size.

Most of new media is commentary, analysis, and opinion that sits on top of original reporting. The major news stories that uncover new information are the foundation upon which most of new media exists.

Institutions like the Post use elite talent, journalistic practice, long-standing reputation, and rigorous process to uncover information about how our world works that otherwise wouldn’t come to light. New media then takes that newly uncovered information and adds opinion, context, and analysis on top.

When these institutions shrink, the entire ecosystem loses its foundation.

4. There Is No Such Thing as Job Stability Anymore

Today, many really talented people at The Post lost their jobs. I’ve seen reports of more than 300 people. This is a reminder that there is no such thing as job stability anymore.

I believe this is a slowly unfolding economic crisis that will eventually force larger external forces to weigh in. But until that happens, your only recourse is resilience, control, and ownership. Resilience means diversification and an adaptable career that bends but doesn’t break when things inevitably go wrong. Control means income and professional happiness derived from entities that you are in charge of. Ownership means assets that free you from overreliance on the volatility of the job market.

Hi, I’m Jamaal. Welcome to Jamaal’s Letter, where I share insights, predictions, and strong opinions on the future of markets, work, business, tech, and more.

The people who read this newsletter are the people who run the world. If you aren’t yet a subscriber, consider joining them by subscribing below. Please share it with your friends and colleagues if you find it helpful:

Subscribe now

Read time: 5 minutes

TLDR:

• Sports labor markets are radically transparent—contracts are public, performance is quantified, and hiring decisions play out in the media. Corporate labor markets are the opposite, and that asymmetry favors employers.

• Corporate employees should steal from the sports playbook to reclaim leverage in their own careers.

• Guaranteed contracts should be the norm, not the exception. Athletes can’t be fired on a whim; most corporate employees can. More guaranteed contracts would give workers time to plan their next move and financial cushion if things go sideways.

• Performance should be public. Sports has box scores; corporate America has hidden performance reviews. Without transparent metrics, hiring defaults to pedigree and networks. Until that changes, building your public profile is the closest substitute.

• Market movement should be visible. In sports, interviews and interest are covered by media, elevating candidates’ profiles even when they don’t get the job. Technology could enable this at scale for corporate workers—and employers should always worry you might leave.

The annual NFL coaching carousel just ended. It’s the process after every NFL season where a bunch of coaches are fired, vacancies arise, and everyone plays musical chairs getting new jobs. There were 10 coaching vacancies at the end of this last season. Nearly 1 in 3 teams decided the status quo wasn’t working.

What struck me this year was how transparent it all was, teams broadcasting not just their hirings but their interview processes on social media. Sports labor markets operate openly in ways that corporate labor markets don’t. Corporate employees could learn from that.

Here are three things they should steal from sports.

1. Guaranteed Money

If you’re employed in the traditional sense, you’re likely an at-will employee. Seventy-five percent of us are. That means you can be fired at any time, for almost any reason, and you can quit without notice.

That doesn’t happen in sports.

Players and coaches sign contracts that specify compensation, duration, and terms. Shohei Ohtani is in year two of a 10-year deal with the Dodgers. Jayson Tatum just signed a five-year extension with the Celtics. These contracts are guaranteed; absent egregious behavior, the employer must honor them regardless of performance.

Because employers don’t like guaranteed contracts, they’re rare in corporate America, typically reserved for senior executives, on-air talent, or roles where specialized skills command leverage. Only one in four employed people has an employment contract. Everyone else is at-will.

More guaranteed contracts would level the playing field. If you sign a two-year deal, you go into the second year knowing you’re on the precipice of renegotiation. If you don’t see an appetite to renegotiate, you have months to prepare. And if you were fired a year in, you’d be paid for the second year, a cushion to figure out your next move.

Without that certainty, people who get laid off are forced to navigate an increasingly frozen job market with no sense of when or how they’ll land.

2. Box Scores for Everything

I recently went to an NBA game for the first time in at least a decade. You could see all the major stats for each player updated in real time on jumbotrons throughout the arena, and I kept wondering why I couldn’t see even more.

Because sports play out in public, performance is highly analyzed, quantified, and available for all to see. The analytics movement has found ways to boil complex measures into single numbers. In basketball, true shooting percentage accounts for field goals, three-pointers, and free throws in one metric. Most importantly, these numbers are transparent and public.

In business, performance may be tracked, but it’s rarely transparent internally, let alone externally. Are you a good accountant? How would anyone outside your organization know?

Without that transparency, labor markets default to signals: where you worked before, where you went to school, who you know. That’s fine for an individual, but reliance on networks isn’t a systemic solution to getting people into the right jobs.

This is why public reputations matter so much, why people with big brands thrive, and why there’s such a high ROI to publishing your expertise, via social media, traditional media, books, or speaking opportunities.

3. Free Agency Headlines for Everyone

This year’s coaching carousel played out on social media in ways I hadn’t seen before. My feed kept surfacing posts from teams publicly sharing that they were interviewing candidates, not just announcing hirings, but broadcasting parts of the interview process itself.

None of the head coaches hired this year were Black. Coaching diversity has been a hot topic in the NFL since the league implemented the Rooney Rule in the early 2000s, which requires teams to interview at least two candidates from underrepresented groups for certain roles. Many lament the lack of Black hires as this year’s failure of the Rooney Rule.

But the rule’s effectiveness isn’t just about who gets hired. It elevates the profiles of candidates who might not otherwise be on teams’ radar. Because these interviews are public—announced by teams, covered by media, debated on sports shows—candidates gain notoriety. It’s like having your resume broadcast on television. They get media and the internet as digital distribution for their profile, even if they don’t land the job.

I’ve written before about how Rooney Rule-type quotas could increase diversity on corporate boards. But the logic applies to all employees. I’m not arguing for more mandates. I’m arguing for more transparency around market movement, some form of public documentation of interest, interviews, and potential moves.

We see this now in press releases and media stories about high-profile people or industries under close scrutiny. But technology enables us to do this at scale, not just for people who take new jobs, but also for people exploring jobs they haven’t yet taken.

I’ll admit, I don’t know the exact mechanism by which this plays out. But more public disclosure of real and potential labor market movement would be great for employees.

Your employer should always worry that you might leave. That belief holds them accountable, especially when job offers come with promises that never materialize.

But employers are insecure, so for this to work, you’d need to marry transparency with guaranteed contracts. If your employer knew you were interviewing, they may want to entice you to stay, or they may want to preemptively replace you. But if this were happening in the final year of a guaranteed contract, both parties would be transparently planning their next move.

The lesson from professional sports isn’t that we all need unions; that’s a separate conversation. It’s that the transparency, certainty, and leverage athletes enjoy didn’t appear out of nowhere. They were built, negotiated, and demanded over the course of decades.

Corporate employees can borrow from that playbook. Push for contracts when you can. Build your public profile. Know your market value. And never forget that the asymmetry of information in labor markets almost always favors your employer.

Unless you do something about it.

Hi, I’m Jamaal. Welcome to Jamaal’s Letter, where I share insights, predictions, and strong opinions on the future of markets, work, business, tech, and more.

The people who read this newsletter are the people who run the world. If you aren’t yet a subscriber, consider joining them by subscribing below. Please share it with your friends and colleagues if you find it helpful:

Subscribe now

Read time: 3 minutes

TLDR:

• A new study reveals a 38-percentage-point gap between how much time executives and workers think AI saves them. 19% of C-suite execs claim 12+ hours weekly, while 40% of workers say AI saves them no time at all

• Executives make AI proclamations while employees handle the messy implementation details, making AI look much better from the corner office

• Every exec feels pressure to adopt AI because it seems like everyone else is—but they’re all competing against exaggerated success stories, creating a Potemkin village of AI productivity

• Financial incentives around AI hype flow disproportionately to executives, creating powerful, motivated reasoning at the top

• AI excels at executive work (decision-making and communication) but struggles with what most employees actually do (making products and services)

A new study from AI consulting firm Section, reported by the Wall Street Journal, reveals a staggering disconnect between executives and employees regarding AI productivity.

The numbers are stark: 33% of executives claim AI saves them 4-8 hours per week, with another 19% claiming savings of more than 12 hours. Meanwhile, 40% of non-managerial workers say AI saves them no time at all. That’s a 38-percentage-point gap between the corner office and the cubicle.

Even more damning: a separate Workday survey found that 85% of employees who claimed AI saved them time actually wasted that saved time correcting AI errors.

If you’ve spent any amount of time talking to executives, board members, and employees instead of reading hyperbolic AI headlines, this isn’t the least bit surprising.

Here’s what’s actually happening.

1. Executives make proclamations; employees deal with implementation. It’s easy to feel optimistic about AI when your relationship to it is strategic rather than operational. The messy details of actually making AI work—prompt engineering, data cleanup, training it on proprietary information, error correction, the workflow integration—fall to the people doing the actual work. Things look and feel much better when you aren’t in the weeds.

2. Peer pressure is creating a Potemkin village of AI productivity. Every executive feels competitive pressure to implement AI because it seems like everyone else is doing it. But here’s the thing: every other exec is also exaggerating how much AI is actually helping. They’re all competing against carefully curated success stories that don’t reflect reality.

3. There’s a huge financial incentive in AI hype, and executives benefit most. Stock prices, board confidence, and career prospects all benefit from projecting AI transformation. This creates powerful, motivated reasoning at the top. When your compensation is tied to demonstrating innovation, you’re going to see productivity gains whether they exist or not.

4. AI is better at what executives do than what employees do. This might be the most important point. Executives spend most of their time making decisions and communicating, exactly what large language models excel at. Employees, on the other hand, make the products and services the company sells. AI is dramatically better at helping you talk, write, and choose between options than it is at actually creating things.

The bottom lineis that you can’t believe most of the hype around AI. Especially when the people hyping it loudest have the least skin in the game when it comes to making it actually work.

Hi, I’m Jamaal. Welcome to Jamaal’s Letter, where I share insights, predictions, and strong opinions on the future of markets, work, business, tech, and more.

The people who read this newsletter are the people who run the world. If you aren’t yet a subscriber, consider joining them by subscribing below. Please share it with your friends and colleagues if you find it helpful:

Subscribe now

Read time: 3 minutes

TLDR:

• Job applicants are suing Eightfold AI, arguing its screening tools should be regulated like credit bureaus by requiring that applicants receive access to their to their file

• A separate class action against Workday could include millions of rejected applicants

• The lack of transparency in hiring is a structural problem that predates AI, but AI has made it worse

• Legislative pressure is already working: Colorado’s 2021 law is why job listings now include salary ranges; Ontario just mandated that employers disclose AI screening and respond to candidates within 45 days

• The market hasn’t fixed this because employers pay for hiring software, not candidates, so candidate transparency isn’t a priority

• Expect more lawsuits and more regulation until someone builds a fundamentally different system

Lawsuits and legislation are coming for the job application process.

As reported by the New York Times, a group of job applicants is suing Eightfold AI, a company that screens candidates for hundreds of employers, arguing that its algorithmic scoring system should be subject to the same Fair Credit Reporting Act requirements as credit bureaus. The suit’s logic is that if you’re going to score over a billion people and determine their employment eligibility, you should have to tell them what’s in their file.

This lawsuit isn’t the first and it certainly won’t be the last. I predict a steep increase of legal and legislative attempts to fix a broken system. The lack of transparency in hiring has been a structural problem for years, and the longer this frozen job market persists—especially as it moves upstream into white-collar work—the more the industry will face pressure from both the courts and legislatures.

I’ve written before about why the job market feels frozen. A big piece of it is opacity:

“There’s no transparency in the job application process. When you hit ‘apply,’ you’re sending your resume into a black hole. You don’t know if a human will see it. You don’t know what part of the application matters. You don’t know if the role is still open or if they’re already in final rounds with someone else. You don’t know if the hiring manager is even looking.”

AI didn’t create this problem; it existed long before the algorithmic gatekeepers arrived. But AI has made it worse, and it’s also become a convenient scapegoat for frustrations that run deeper than any single technology.

The Eightfold lawsuit isn’t the only one. A separate case against Workday is proceeding as a potential class action that could include millions of rejected applicants. In that case, a federal judge found that evidence—including a rejection notice one applicant received at 1:50 a.m., less than an hour after applying—”plausibly supports an inference that Workday’s algorithmic tools disproportionately reject applicants based on factors other than qualifications.”

Judicial action is one pressure point. Legislative action is another.

The only reason most job listings now include salary ranges is because of a Colorado law, the Equal Pay for Equal Work Act, which since 2021 has required employers with at least one Colorado employee to include a good‑faith minimum and maximum pay range.

One state passed a law, and the entire industry was forced to shift.

Ontario recently introduced an “anti-ghosting” rule requiring employers to inform interviewed candidates of the outcome within 45 days, and to disclose when AI tools are used to screen applicants.

These are early moves, but the larger pattern is clear: it shouldn’t be this hard for well-qualified, well-credentialed people to find work. Technology has made life more efficient in countless ways, but it hasn’t materially improved the job search process, and AI has only made it worse. People are finding the combination of broken processes and declining job security increasingly unacceptable. The market hasn’t responded, so people and outside forces are stepping in.

What makes this problem so hard to solve is the misalignment I noted in that earlier piece:

“The systems that govern hiring workflows are purchased, implemented, and optimized by companies, not candidates. This means that candidate transparency is not a priority. And this problem—lack of transparency—is structural. It existed long before the pandemic, the ‘Great Resignation’ that followed, or the rise of AI. Unless the industry adopts a new norm or builds a fundamentally different kind of tool, it will persist long after the current frozen market thaws.”

Whoever cracks this—genuine transparency in the job application process—is going to be rich beyond belief. You’re coordinating between two entities with different incentives, and that’s precisely why it’s so difficult. Solving it would make things better for both job seekers and employers.

Until then, expect more lawsuits and more legislation. The black box is getting sued. And that’s just the beginning.

The people who read this newsletter are the people who run the world. If you aren’t yet a subscriber, consider joining them by subscribing below. Please share it with your friends and colleagues if you find it helpful:

Read time: 4 minutes

Subscribe now

Ben Affleck and Matt Damon went on the Joe Rogan podcast to promote their new Netflix movie, The Rip. During the conversation, Affleck launched into a nearly 3-minute diatribe on AI, which has been clipped and is going viral. AI nerds on Twitter are apoplectic because he said some uncomfortable truths.

You can watch the clip here:

Here are three things he got spot on:

AI Hype Is Corporate Marketing

“We have this sense of existential dread that it’s going to wipe everything out, but that actually runs counter to what history seems to show, which is that adoption is slow and incremental. A lot of that rhetoric comes from people trying to justify valuations around companies, saying they’re going to change everything—that in two years there’s going to be no more work. The reason they’re saying that is because they need to ascribe a valuation for investment that can warrant the CapEx spend on these data centers, with the argument that as soon as we do the next model it’s going to scale up and be three times as good—except that, actually, ChatGPT-5 is about twenty-five percent better than ChatGPT-4 and costs about four times as much in electricity and data. So those same people may say they think it’s plateauing.”

Most of what you’re hearing about AI is corporate marketing.

If you’ve followed me, you know I’ve been writing extensively about this for the past year. Prognostications that AI is going to create a technology utopia or, conversely, that AI is going to kill all jobs or kill all people are mostly designed to tell a story that justifies some sort of corporate or financial outcome. Sometimes it’s a smokescreen for layoffs. Other times, it’s a justification for fundraising valuation. Too often, it’s to farm engagement, which is itself monetized.

Almost every highly touted point of view on AI is motivated reasoning. Most of the people making these claims have a financial stake in the AI use cases they’re promoting, a reputational stake in the distraction they’re hiding behind, or a profit incentive behind the regulatory position they’re carving out.

Tech History Suggests AI Will Be Big—But Not As Big As People Are Promising

“We have this sense of existential dread that it’s going to wipe everything out, but that actually runs counter to what history seems to show, which is that adoption is slow and incremental.”

I’m settling into the view that while AI will be a society-altering, once-in-a-generation technology, like the internet, it won’t fundamentally change human behavior or economics in the way many people are predicting.

AI will not live up to the hype. Like the internet, it will be democratizing in some ways and stratifying in others. But it’s not going to eliminate all work. Human beings have an endless demand for products and services and will always look for ways to display their social status. That means there will be more jobs to replace those destroyed by AI.

I will concede that it’s important to distinguish between low adoption now and low impact later. I predict that the former will have correlation—and certainly no causal connection—to the latter.

AI’s Creativity Is Inherently Limited by How It Works

“If you try to get ChatGPT or Claude or Gemini to write you something, it’s really bad, because by its nature it goes to the mean, to the average, and it’s not reliable. I just can’t even stand to see what it writes. Now, it is a useful tool if you’re a writer and you’re trying to set something up where, say, someone sends a letter that’s delayed two days—I can give you some examples of that—but I actually don’t think it’s very likely.”

To anyone who understands how large language models work, this isn’t surprising. What is meant by the ‘intelligence’ in Generative Artificial Intelligence—specifically Large Language Models—is fundamentally a statistical prediction of which words come next, largely based on what has already been written. Anyone who uses AI writing tools deeply likely knows this to be true. What’s more important, in my opinion, is that because AI predicts what’s most likely to come next based on what has historically come before, I question whether LLMs themselves will ever be able to create anything fundamentally new.

Affleck’s commentary on technology is nothing new; he has long been open about his views. For example, he shared great insight on the creative limitations of LLMs back in 2024.

Here he is in 2003 predicting the rise of streaming in music and video:

Hi, I’m Jamaal. Welcome to Jamaal’s Letter, where I share personal reflections, insights, predictions, and strong opinions on the future of money, work, education, policy, and more.

The people who read this newsletter are the people who run the world. If you aren’t yet a subscriber, consider joining them by subscribing below. Please share it with your friends and colleagues if you find it helpful:

Subscribe now

Read time: 5 minutes

I like to play a game at dinner parties.

It usually starts after the second drink, when the conversation drifts toward something serious. I’ll lean in and ask: What do we do today that’s considered totally normal and acceptable that we’ll get canceled for in the future?

The obvious answers come quickly: The industrial-scale killing of animals for food. Our reckless treatment of the natural environment. Our addiction to overconsumption, from fast fashion to disposable everything.

But years ago, at a restaurant in Berlin, someone flipped the question: what’s unacceptable now that will be forgiven in the future?

I was surrounded by Americans and Germans. We’d spent a lot of time traveling together over the past week and had gotten well past the small talk. The conversation turned toward race. The group had varying levels of familiarity with America’s civil rights movement, its racial legacy, and its most well known protagonist, Dr. Martin Luther King Jr.

But something was different. Some of the Europeans—especially those who has spent considerable time in the U.S.—had a more sober, less laundered understanding of Dr. King. Their knowledhe of American history wasn’t only more nuanced, but they also understood America’s current racial climate without without either the defensiveness or guilt that many white Americans often carry. They knew what Dr. King actually championed in ways that many Americans I’ve met never have.

That conversation stuck with me. It made me ponder the question: What behaviors or people are unpopular and controversial today that will be completely normal in the future? Even celebrated?

Today is Martin Luther King Jr. Day. Millions of Americans will post quotes on social media. Schools, businesses, and financial markets are closed. Companies will invoke his name and legacy to sell you something.

We often forget Dr. King’s unwavering faith in progress, summarized in his famous quote: “The arc of the moral universe is long, but it bends towards justice.” To me, this belief suggests that what is considered radical, taboo, or uncomfortable in one era—like the full realization of racial and economic justice—will eventually be embraced as necessary and right by the next. It means that the uncomfortable, controversial truths being spoken today by voices we currently dismiss may well be the mainstream moral clarity of tomorrow.

I remember marching in the MLK Day parade as a kid growing up in Denver, Colorado. The parade was part of Denver’s unusually large and visible Martin Luther King Jr. celebration for a Western city. In Denver it is known as the “Marade,” a blend of march and parade. Denver was early and assertive in creating a state holiday in King’s honor. That public ritual connects back to 1956, when Dr. King visited New Hope Baptist Church—the church my family attended as a child—during the Montgomery era. And I can still remember, in 1992, when white supremacists staged a counter‑protest around the holiday, a reminder that honoring King meant walking through contested space, not just enjoying a feel‑good parade.

Looking back, even that feels performative now. I think our understanding of what Dr. King was championing was hollow. Surface-level. Most people consume Martin Luther King now through a very sanitized lens.

Now we see organizations invoke his name as part of marketing campaigns. No real conviction behind what he championed. No willingness to challenge the status quo from which they benefit. Just his image, his messages, and his legacy hollowed out for mass consumption.Before you quote Martin Luther King Jr. on the holiday that bears his name, know this: He was a pariah in his day.

During his time, he was controversial and divisive. His calls for racial equality and nonviolent protests were immensely polarizing. He was called an extremist. The U.S. government considered him a danger and actively worked to discredit and undermine him.

Only through the lens of history, moral clarity, and social progress has an overwhelming majority of Americans come to see his vision as not just mainstream but as the epitome of courage and righteousness.

Even the passage of the law that created Martin Luther King Jr. Day was as embattled as his reputation during his lifetime.

The federal holiday came only after years of political resistance and public controversy. In Congress, opponents argued that a paid federal holiday would be too expensive. They claimed that King was too controversial, attacking his opposition to the Vietnam War and smearing him with accusations of communist ties. President Ronald Reagan initially resisted the idea as well, citing cost concerns and questioning whether public support was rooted in “image, not reality.” He eventually signed the holiday bill in 1983 under strong congressional and grassroots pressure.

And even now, we are celebrating a sanitized version of Dr. King, one watered down to be palatable enough for corporate PR campaigns.

Many of Martin Luther King Jr.’s core ideas would still challenge mainstream politics today. He sharply critiqued capitalism, militarism, and U.S. foreign policy. He argued that the “giant triplets” of racism, materialism, and militarism were intertwined. He called for a “radical revolution of values” that put people above profit. These positions go far beyond most contemporary policy debates.

He opposed the Vietnam War in uncompromising moral terms. He condemned U.S. militarism abroad. He supported campaigns like the Poor People’s Campaign and proposals for guaranteed income to tackle structural poverty. These positions would still be labeled radical or far left in many national conversations today.

Which brings me back to that dinner in Berlin. And back to the question I can’t stop asking.

What controversial and divisive ideas of today will we look back on and realize were right all along?

What current voices of change are we dismissing or resisting because they challenge our comfort zones?

I think about the economic injustices that persist in America today. I think about the slow march toward authoritarianism happening under our current presidential administration. I think about how the scapegoating of immigrants is playing out across American cities as you read this.

I can think of many controversial positions and many controversial people whose positions may be vindicated by history.

The question is whether we’ll have the courage to recognize them now, or whether we’ll wait, as we did with Dr. King, until it’s safe enough to celebrate them.

Hi, I’m Jamaal. Welcome to Jamaal’s Letter, where I share personal reflections, insights, predictions, and strong opinions on the future of money, work, education, policy, and more.

The people who read this newsletter are the people who run the world. If you aren’t yet a subscriber, consider joining them by subscribing below. Please share it with your friends and colleagues if you find it helpful:

Subscribe now

Read time: 6 minutes

When I wrote about the TikTok ban earlier this year, I found myself reflecting on how much my thinking about social media has changed:

I remember taking a course in grad school around 2012 where the consensus was that social media had been overwhelmingly positive. It was the aftermath of the Arab Spring. Most of us believed platforms like Facebook and Twitter had done more good than harm.

Fast forward to today, and the consensus has changed. Mounting evidence suggests social media has been a net negative, especially for teenagers and developing brains. It has become the backbone of marketing and distribution in ways unimaginable in 2011.

In the aftermath of the Charlie Kirk murder, I logged back onto Twitter, an app I had largely abandoned. Half my feed was people celebrating his death, while the other half used it to justify some form of war. That moment convinced me that social media, writ large — with its tendency toward radicalization, polarization, and bullying — has, despite its benefits, been unequivocally bad for us.

These social-media-is-bad-for-us moments happen for different people, on different subjects, at different scales. You’ve probably had your own, some display of abominable behavior or political venom that made you think: this can’t be good for us.

However, I would argue that the most compelling case against social media isn’t political polarization, it’s what it’s doing to young people.

No one has made this case more effectively than NYU professor Jonathan Haidt. His argument is straightforward: the shift to a phone-based childhood has driven a sharp rise in teen anxiety, depression, and self-harm. These platforms disrupt sleep, foster addiction, and intensify social comparison. Girls are especially harmed by appearance-driven platforms; boys by pornography and immersive gaming.

Haidt’s proposed solutions are equally direct: delay smartphones until high school and social media until about age 16, and make schools phone-free.

But delaying social media isn’t just an individual parenting choice. It’s a collective coordination problem. One family’s restraint can socially penalize their child if all their peers are online. The kid who’s “off” social media risks missing group chats, inside jokes, and social planning. Because these platforms derive their value from network effects, meaningful protection requires society-wide rules so no single child pays the social cost for doing what adults agree is healthier.

This argument is compelling to me, partly because I’m a parent, and partly because I’ve seen how social media polarizes, distracts, and wastes time and energy. The larger takeaway: social media isn’t the consensus good we once thought it was.

I think more people will reach this conclusion in 2026. And where populations lead, governments follow.

Prediction: More than 3x as many people will be subject to age-related social media bans by the end of 2026

By the end of 2026, more than 200 million people across the globe will be subject to an age-related social media ban — more than 3x today’s number.

A growing group of countries has moved beyond basic “13+ with consent” rules toward hard age floors.

• Australia now effectively bans social media for under-16s.

• Malaysia’s Online Safety Act does the same starting this month.

• Denmark and Norway are implementing similar restrictions.

• France requires parental consent for users under 15.

• The European Union has called for a harmonized 16+ minimum.

• Here, in the United States, states like Florida, Utah, Louisiana, and Tennessee have enacted age-verification laws, creating a patchwork that I believe will eventually expand nationwide.

The momentum is real. The consensus has shifted. Now the laws are catching up.

Hi, I’m Jamaal. Welcome to Jamaal’s Letter, where I share insights, predictions, and strong opinions on the future of markets, work, business, tech, and more.

The people who read this newsletter are the people who run the world. If you aren’t yet a subscriber, consider joining them by subscribing below. Please share it with your friends and colleagues if you find it helpful:

Subscribe now

Read time: 4 minutes

I consider myself pretty good at spotting fake things on the internet.

At a couple of points in my career, I’ve done those silly icebreakers where the question is, “What’s your useless superpower?” My answer for a long time has been, “I can spot fake things on the internet.” I was always the person in the group chat pointing out the thing people were gleefully circulating as fake.

Since generative AI went mainstream in 2022, I’ve still been relatively confident in those abilities. I’m still that same person calling out AI-generated content. This mostly applies to images and videos. At first, it was easy and obvious; AI wasn’t very good at drawing fingers or hands. Later on, the tell was this hard-to-describe AI gloss, where things seemed overly polished, a little cartoonish, and overly smooth.

For a while, AI-generated humans were stuck in the uncanny valley. Mind-blowingly more realistic than anyone from history would have ever thought possible, yet oddly off-putting because of their approximate realism.

But all of that changed in the last few months.

When Google DeepMind released Nano Banana Pro in November 2025, it was undoubtedly the most realistic AI image generation tool ever created. Built on Gemini 3 Pro, it generates images with unprecedented accuracy, detail, and consistency. The model understands context, maintains character consistency across scenes, and produces output that pushes toward 4K resolution.

This viral Nano Banana Pro-generated image of a woman in a restaurant is what did it for me. The woman looks 100% indistinguishable from a real photograph. Not to some grandma on the internet who will fall for every scam, but to even the most skeptical online AI spotters. No weird hands. No AI gloss. No broken laws of physics hiding in the background. Once I saw that image, I knew we had crossed a threshold from which we would never return.

So much of the talk about AI has focused on people using it to create videos that look real. In other words, things that didn’t happen are being passed off as things that did happen.

What we don’t often talk about is the opposite: things that did happen being passed off as things that didn’t.

We’ve spent years worried about deepfakes. We’ve built detection tools. We’ve trained journalists and election officials to spot manipulated media. The entire apparatus of AI skepticism has been oriented around one question: Is this real content being faked?

But we’ve neglected the mirror image of that threat.

Prediction: The First “It Wasn’t Me, It Was AI” Scandal

In 2026, we’ll have our first societal test of this lesser-talked-about scenario.

We’ll have a cultural flashpoint moment — a video or image that goes viral and captures everyone’s attention. Think of the scale of the Coldplay concert cheating scandal that became the internet’s main obsession for a week in July of last year. We’ll see a video or image of someone doing something that gets them in trouble. Their excuse — the claim upon which their innocence, legal or otherwise, rests — will be that the image or video in question is AI-generated.

In 2026, we’ll have a scandal that isn’t something AI being passed off as non-AI.

It will be something that we all, on its face, believe to be real that will be passed off as AI-generated.

Imagine if the 2014 Ray Rice video was passed off as a state-of-the-art forgery, a maliciously rendered clip inserted into genuine surveillance footage to fit a narrative.

Or what if the Jay-Z and Solange elevator fight, a cultural touchstone born of grainy, irrefutable evidence, was dismissed as a viral “AI stunt,” a perfectly faked public curiosity designed to stir drama?

More recently, imagine the Sean “Diddy” Combs-Cassie hotel footage, which corroborated years-old accusations, became contestable as a perfectly-modeled deepfake engineered to weaponize a long-ago private moment and influence public opinion.

The accused will claim that damning footage was fabricated. Their publicists, online defenders, or lawyers will decry the maliciousness of AI scammers. Talking heads will debate whether we can ever truly verify visual evidence again. And regardless of the truth, a meaningful portion of the public will have reasonable doubt.

It is critical to note, however, that these specific high-profile examples are unlikely to be the actual testing ground for this prediction. The footage in these cases — casino surveillance, hotel elevator cameras, and institutional CCTV — has an origin that is more easily established and verifiable by a third party. The first true “It Wasn’t Me, It Was AI” scandal is far more likely to involve footage captured on a personal cell phone, doorbell cam, or from an undetermined or not entirely credible source, where the video’s origin is immediately contestable.

This is one of the darker sides of our AI moment. When AI-generated content becomes indistinguishable from reality, reality itself becomes deniable. Every piece of visual evidence becomes contestable. Every photograph becomes a Rorschach test of whom you want to believe.

The infrastructure of accountability — from courtrooms to newsrooms — was built on the assumption that seeing is believing. We’re about to find out what happens when that assumption breaks down.