The most repeated claim in corporate AI right now (“we’ve 100x’d our productivity”) isn’t a lie told by liars. It’s a lie nobody in the room is allowed to correct. That’s the uncomfortable core of an essay by Nikhil Suresh making the rounds on Hacker News this week, where it climbed to 172 points and set off a long debate about why smart people keep saying absurd things about generative AI.
The piece opens with a Fortune 500 executive who agreed to talk only if kept anonymous, described as fearing “execution by firing squad by their board.” This executive was sharp and technically fluent, yet ran a company committed to the usual battery of inflated AI claims. So Suresh asked the obvious question off the record: why repeat this stuff with no pushback? Was it just sales fluff?
📉 The real reason dissent stays silent
The answer is more interesting than “marketing.” Sales hype was part of it, but not the main constraint. According to the account in Hacker News, the vendor’s own customers were the ones claiming 100x gains. And here’s the trap that follows:
- If a vendor executive says those gains aren’t plausible, they contradict the customer executive.
- Contradicting the customer reads as an attack, or worse, heresy.
- That perception can get an enterprise contract cancelled.
- Losing a contract over an opinion that doesn’t touch your mission is a fast way to get fired.
So everyone stays quiet. The doubt exists. It just never reaches the meeting. Suresh’s larger argument is that the whole corporate world, not only tech, is caught in an AI mania that “brooks no dissent.” He frames it as religious fervor, where the heretics are excommunicated even though the heretics are mostly right.
🔍 What stands out here
This is significant because it explains a pattern that numbers alone don’t. Survey after survey shows executives reporting massive AI productivity gains while measured output barely moves. The usual explanation is hype or self-delusion. Suresh adds a structural one: the incentive to agree is stronger than the incentive to be accurate. When your revenue depends on not correcting your customer’s fantasy, honesty becomes a career risk.
There’s a second layer worth naming. Suresh points out that computers already changed the world, but most people never learned to make them do anything creative. For them, computers are communication tools, not creative ones. Generative AI flips that. A manager with zero coding aptitude can suddenly produce something that felt impossible last year. To that person, this feels like the Big Bang. The tools are genuinely useful and impressive. The problem is the gap between “useful” and the “orders of magnitude” they imagine.
🧭 What practitioners and businesses should actually do
You don’t fix an incentive problem with better slides. A few practical moves:
- Measure the boring way. Track cycle time, defect rates, and real output, not self-reported “productivity.” If the 100x is real, it shows up in the numbers.
- Reward the dissenter. If pointing out that a pilot failed gets someone sidelined, you’ve built the exact silence Suresh describes. Make skepticism safe and cheap.
- Separate the tool from the theology. Adopt AI where it clearly helps. Refuse to sign onto claims you can’t defend with data.
- Watch your vendor relationships. If your supplier won’t tell you when something won’t work, you’re not getting advice. You’re getting flattery you paid for.
Suresh doesn’t pretend to know when this breaks. His closing bet is blunt: there’s no way to talk these believers down “until the bubble bursts.” That’s a prediction, not a fact, and he’s a skeptic writing to skeptics, so weigh it accordingly.
Still, the mechanism he describes is real and testable. The next time a productivity claim goes unchallenged in your organization, ask who in the room disagrees and why they’re not saying so. The answer tells you more about your culture than about the technology. Full essay is worth reading at the original source.