AI Doomers vs. Reality: Why Humans Are More Adaptive Than AI Predictions Think

I've got a friend who's convinced that AI will make his job obsolete within two years. Every time a new research paper drops, he texts me something like "well, this is it." He's been doing this since 2023. He still has his job. He's also gotten two promotions.

That dynamic, the gap between doom predictions and what actually happens, is what I want to talk about.

Last week, a research piece from Citrini Research went viral and briefly rattled markets. The big claim: AI agents will systematically destroy the moats of companies like DoorDash, real estate brokerages, and payment networks. Margins collapse to near-zero. The S&P 500 falls 38%. Everything spirals.

It makes for a great read. It's also probably wrong.

Ben Thompson at Stratechery wrote the definitive pushback, and honestly, we think he nailed it. Let's walk through both sides.

The Citrini Argument

Credit where it's due: the core logic isn't dumb. AI agents eliminate what Citrini calls "habitual intermediation," which is basically the moat companies build on the fact that you're too lazy to comparison-shop. Today, you open DoorDash out of habit. Tomorrow, your AI agent shops every delivery platform simultaneously for the best price. Commission compression follows. Meanwhile, autonomous vehicles replace drivers, and machine-to-machine commerce routes around credit card interchange fees via stablecoins. Real estate agents get displaced as MLS data becomes universally accessible.

Barriers to entry collapse. Commoditization accelerates. Incumbents have nowhere to hide.

It's not a crazy story. But it assumes the world just... sits there and takes it. And that's not how the world works.

Thompson's Counter: The Fundamental Error

Thompson's core point is beautifully simple: the world isn't static. Every single time someone has predicted that technology would wipe out a profession or collapse a market, they've underestimated how quickly humans and institutions adapt.

Real estate agents are his best example, and it's a good one. The internet eliminated the information gap that supposedly justified their existence. Zillow, Redfin, and every MLS aggregator made property data universally available years ago. And yet: real estate agents are still here. Their share of transactions barely budged. Why? Because buyers and sellers found new reasons to want a human involved: negotiation, emotional support, local relationships, someone to blame when things go sideways.

The moat shifted. It didn't disappear.

DoorDash Specifically

Thompson makes a detailed case for why DoorDash is tougher to displace than Citrini suggests, and it's worth walking through:

Exclusive data. DoorDash has years of proprietary order history, restaurant relationships, and neighborhood delivery patterns. An AI agent searching across platforms sees public prices, but it doesn't know what DoorDash knows about optimizing a three-block radius at 7pm on a Tuesday. That stuff is gold, and it's not sitting in a public API.

Network effects are real and three-sided. DoorDash isn't just a marketplace. It's a network connecting consumers, restaurants, and drivers. Each side reinforces the others. A new entrant doesn't just compete on price; they have to recruit all three sides at the same time. Anyone who's tried to build a marketplace knows how brutal that is.

Physical world complexity. You can fork software. You can't fork a delivery network. Knowing which drivers are reliable in which neighborhoods, how restaurants handle rush hour, which buildings have nightmare lobby access. That knowledge lives in operations, not in some database an AI agent can scrape.

Payment infrastructure. The "stablecoins will route around interchange" argument sounds futuristic, but it skips over the massive regulatory and adoption hurdles that aren't solved yet. Visa and Mastercard have decades of fraud prevention, dispute resolution, and merchant relationships. You don't just... replace that.

"Agent on Agent Violence"

There's one more argument Thompson makes that I think is particularly sharp: AI-driven competition might actually benefit the incumbents who've already built trust and data advantages.

Think about it. If AI agents start aggressively comparison-shopping on your behalf, who wins? The platforms with the best data, the most reliable operations, and the strongest track record. Your AI agent optimizing for you isn't going to pick the scrappy newcomer with no reviews; it's going to pick the one that performed best historically. That's DoorDash, not some startup that launched last month.

The arms race speeds up, but the incumbents already have the bigger arsenal.

Our Take

We build AI agent workflows every day. Not to replace people, but to help them do more.

The Citrini framing is tempting because it treats AI capability as the only thing that matters. But the better question isn't "can an AI agent do what a human does?" It's "what happens when humans and AI work together on things neither could handle alone?"

Real estate agents who use AI tools to juggle 50 client relationships instead of 10 aren't going anywhere; they're just way more productive. DoorDash engineers using AI to optimize routing in real time aren't losing their jobs; they're widening the gap between DoorDash and everyone else.

The future we're building toward isn't displacement. It's partnership. Humans bring judgment, relationships, accountability, and the ability to deal with situations that weren't in any training data. AI brings speed, scale, and the ability to spot patterns across more data than any person can process.

Doom scenarios make for great thought experiments. But betting against human adaptability? Historically, that's been a losing trade.

We'll stay on Thompson's side.

---

Sources: Citrini Research: 2028 Global Intelligence Crisis · Stratechery: Another Viral AI Doomer Article · Bloomberg: Software, Payments Shares Tumble After Citrini Post