In the last month, the AI recruiting darling got hacked and the AI meeting note-taker got sued, and almost nobody outside of tech noticed either one.

Mercor, valued at $10 billion, had four terabytes of data walked out of its systems through a compromised open-source library. Candidate profiles, source code, API keys, interview videos. Forty thousand contractors exposed. Five class actions filed within days. Meta reportedly paused its work with them. Otter.ai has been in court since August, now consolidated into In re Otter.AI Privacy Litigation, and the allegation is simple: its transcription bot has been quietly joining meetings, recording people who never agreed to be recorded, and feeding the audio back into its training data. Twenty-five million users. More than a billion meetings. The plaintiff isn't even an Otter customer. He was just on a call.

Neither story is over. Both are early. Both are being covered as discrete incidents, a breach story and a privacy story, filed under cybersecurity and under law. That's not what they are. What they actually are is the same story, told twice in a month: the infrastructure for automated hiring and automated listening is being built faster than the infrastructure to make either of them safe, legal, or particularly good at the job they were hired for. It's being built on top of you, whether you agreed to it or not, and the people building it are not waiting for you to catch up.

This is the part where I'm supposed to say "and yet the productivity gains are undeniable." I'm not going to, because they aren't.

In February, the National Bureau of Economic Research surveyed roughly six thousand CEOs and CFOs across four countries. Ninety percent of firms reported zero measurable productivity impact from AI. PwC's 2026 CEO survey, covering 4,454 executives across 95 countries, found 56 percent got "nothing out of" their AI investments. Only 12 percent said it both grew revenue and cut costs.

Goldman Sachs estimates AI is contributing 0.1 to 0.2 percentage points to US GDP growth. Statistically indistinguishable from noise. Jason Furman ran a different cut: strip out AI capital expenditure and the rest of the US economy grew 0.1 percent in the first half of 2025. Without the AI frenzy, we're in a recession. With it, we have a $667 billion infrastructure buildout producing, in aggregate, approximately nothing. J.P. Morgan ran the math on what it would take for the spending to pencil: $650 billion in new annual revenue, in perpetuity, to deliver a ten percent return. They compared it to the late-nineties telecom fiber boom, where the cable got laid, the revenue never caught up, and the investors got wiped.

So where is the money actually going? To two places, mostly.

The first is to the systems integrators. Accenture did $69.7 billion in fiscal 2025. Deloitte did $64.9 billion. The system integration market is projected to hit $764 billion by 2030. These firms are not thriving because AI works. They are thriving because it doesn't. Someone has to clean the data, rewire the workflow, babysit the rollout, and explain to the board why the promised gains are just one more quarter away. That someone bills $300 an hour, and the check clears whether the model ever ships or not.

The second is to the people selling the promise. Courses on how to prompt. Templates for how to build. Newsletters about the future of work. Communities about shipping with AI. Look at any transparent indie-hacker revenue dashboard and count what percentage is a product versus a product about products. It's shovels all the way down. The gold rush is real. The shovel market is booming. The number of people actually pulling gold out of the ground is, on the evidence, small.

I say this as someone who has built real things with these tools. I have shipped products in months that would have taken years without this stuff, and the productivity at the tool level is genuinely new. The revenue is the lie. Or more precisely, the revenue is a thing almost nobody has, and almost everyone is performing, because performance is the price of admission to the next round, the next gig, the next launch, the next course.

The AI economy in 2026 is mostly an economy about the AI economy. The actual underlying goods and services are, per the data, more or less unchanged.

Meanwhile, inside the companies where all of this is supposedly working, people are being laid off.

They are being laid off on the expectation that the productivity revolution has arrived, while the data says it mostly hasn't. They are being priced out of a future that hasn't materialized. This is not a mistake. It is, if you zoom out, a fairly elegant arrangement for the people at the top. You cut headcount on the promise of AI gains. If the gains arrive, you look prescient. If they don't, you still cut costs. There is no version of this bet where the worker wins, and the bet is being placed in every boardroom in the country right now.

The word for where this is heading, if it keeps heading there, is technofeudalism. I did not coin it. Yanis Varoufakis did. The idea, stripped down, is that the productive economy of capitalism is being quietly replaced by something older and meaner, where a small class of people own the platforms, the infrastructure, and the models, and everyone else pays rent to live and work on top of them. Not with money, exactly. With attention, with data, with the willingness to keep generating value that accrues somewhere else. The middle hollows out not because the middle failed, but because the middle is no longer necessary to the people at the top.

Recruiting is the canary in this mine, because the tech industry has decided recruiting is pure coordination work, and pure coordination work is exactly what agents are supposed to do. In the last eighteen months I have watched hundreds of resume scanners and offer-letter generators and interview-intelligence platforms launch into the same narrow slice of the hiring funnel. Most of them do the same thing. Most of them do it badly. The most valuable company in the space just had its underwear pulled down in public, and the second-most-talked-about one is defending a consolidated privacy class action.

I am not writing this from the outside. I built one of those tools. HireScribe is mine. It's an interview transcription and intelligence layer, and it exists because I genuinely thought the recruiters I knew deserved better software. I also, at one point, started building a resume scanner, and I stopped because I realized I was trying to monetize a moment rather than build something actually useful. The line between contributing to the problem and critiquing the problem, in this industry, is thinner than anyone selling you a course would like to admit.

Here is the thing almost nobody wants to say out loud: if recruiting were actually pure coordination work, it would already be automated. It isn't. It's judgment work, context work, relationship work, reading-the-room work, reading-the-resume-that-doesn't-read-well work, figuring-out-which-candidate-is-lying-and-which-one-is-just-nervous work. The parts the tools are good at are the parts recruiters were already happy to hand off. The parts the tools are bad at are the parts that actually matter, and the failures show up as lawsuits and breaches rather than as productivity gains. Instead of replacing recruiters, AI has mostly given recruiters more tools to babysit. The ones I talk to are working more hours than they were two years ago, not fewer.

This is the productivity paradox in miniature. A function the industry has confidently declared "automatable" is being automated into a pile of lawsuits, data breaches, and more work. My whole industry is going obsolete before it notices, and the thing replacing it doesn't actually work yet. Both things are true at once, which is how most obsolescence actually goes.

Think about it as a game of Monopoly that's been going for a while.

Most of the properties are owned. The rents on the good squares are high enough that landing there wipes you out. The bank has extended everyone some credit, and the credit is what's keeping most players in the game. A small number of people have hotels on the dark blues and the greens. Most of the people still rolling the dice are one bad turn from mortgaging the last thing they own.

In a normal game, this is the endgame. Somebody wins, the table gets flipped, everyone goes home, and next time you shuffle the deeds and start over. In the actual economy, there's no reset. You don't get to shuffle the deeds. The people with hotels on Boardwalk get to keep them, and their kids get to keep them, and the AI infrastructure they're spending $667 billion on this year is, functionally, more hotels. On better squares.

The question nobody in the mainstream labor conversation is willing to ask out loud is: how long do people keep playing a game where most of the players have no realistic path to winning, accumulating, or even staying in? Not forever. The answer is not forever. People stop playing when the game stops making sense to them, and the signs of that stopping are already here, at the edges, if you know where to look.

COVID was a preview. People got shoved out of the office and a meaningful slice of them never came back in the same way. Some of them became creators. Some of them started small operations. Some of them assembled patchwork livings out of three or four streams nobody would have called a job twenty years ago. It wasn't scaled. It wasn't celebrated. But it was real, and it is still growing. The creator economy, the indie operator economy, the small-studio economy, the patronage economy. People quietly refusing to keep playing the square game, and finding each other in the margins.

This is the part of the future I actually believe in, and the part I am betting my life on. Not because I'm optimistic by temperament. I'm not. But because the other direction is technofeudalism, and the only counter-move that scales from the individual up is people deciding they would rather be part of something smaller, stranger, and more their own than a replaceable input on someone else's board.

Call it a new kind of guild. Call it patronage. Call it the sovereign node. Call it what the Romans called otium, the non-work time in which a free person actually became a person. The specific word matters less than the pattern: small operators, narrow audiences, strange specific work, time recovered from a game that was always rigged. None of this is guaranteed. The drift is toward the other thing. But the counter-drift is real, and the number of people feeling the pull is growing every quarter that this paradox continues to not resolve in their favor.

You are allowed to step off the board. The board is telling you that you aren't. The board is wrong.

I am writing this from inside the paradox. I was laid off a while ago, for the fourth time, and at some point during the job search I realized I was applying for roles that would put me back in the exact machine I just watched eat thousands of people. I stopped applying. I am building things I believe in, with tools that genuinely amplify what one person can do, and I have not yet turned any of it into a living. The runway is shorter than I'd like and longer than I feared. I chose this. I'm not asking anyone to feel sorry about it.

I'm telling you because I think a lot of people reading this are quietly running the same math, and the numbers are not reassuring, and the people on LinkedIn claiming they've cracked it are mostly selling you a course on how you could crack it too.

The honest version is this. We are in the capital expenditure phase of something whose payoff hasn't arrived, may arrive in a form none of us expected, and will accrue by default to people who are not us, unless enough of us choose something else. The productivity is real. The revenue is theater. The labor market is being restructured on the promise of a future that nobody has actually delivered, and the people restructuring it do not need your buy-in to continue.

You are not crazy. You are not behind. You are not missing the move. You are inside a paradox that most of the smart money doesn't know how to price yet, and neither do I. But I am increasingly sure that the move is not to play harder. The move is to notice what game you're in, and start building the thing that comes after it.

Welcome back to 404.

Tyler

Sources and further reading:

On the Mercor breach: Fortune's coverage and TechCrunch's follow-up.

On the Otter.ai litigation: NPR's original report and a recent HR Executive analysis of the consolidated case.

On the productivity paradox: Fortune on the NBER and PwC data, Goldman Sachs via Fortune, and a deep breakdown of the capex math from maseconomics.

On technofeudalism: Yanis Varoufakis, Technofeudalism: What Killed Capitalism (2023).