The Essay That’s Making Rounds
Nikhil Suresh, a consultant who has run point on sales and led technical engagements for his company over the past year, published an essay called “AI Mania Is Eviscerating Global Decision-Making” that landed like a depth charge on Hacker News and was swiftly picked up by John Gruber at Daring Fireball. The timing could not be more perfect — or more grim.
His thesis is simple and devastating: across every single AI project he and his team have observed over the past year and a half, the success rate has been zero percent. Not “most.” Not “many.” Zero. And the people who could speak up about it have learned to stay silent — or else.
I read the whole thing twice because it resonated so deeply with what I’ve seen as an ICT manager in the Philippine government sector. While we’re not a Fortune 500 company, the same dynamics play out in every organization that suddenly decides AI is the answer to questions nobody thought to ask. If you’ve been feeling like you’re the only sane person in a room full of people who’ve collectively decided to believe something that doesn’t match reality, this essay will make you feel a little less alone.
Why AI Projects Are Failing at Scale
Suresh breaks down the underlying mechanics with brutal clarity. The most common pattern, rolled out across businesses worldwide, is the internal chatbot or the customer-facing chatbot. The story is always the same: employees don’t use internal chatbots because most companies have low-quality documentation that an LLM can’t magically improve upon, and customer-facing bots leave actual problems unresolved.
He describes a personal experience with Mitsubishi’s voice bot — natural-sounding, responsive, promised a callback. Six months later, he never got one. That request likely shows up in their metrics as “resolved without human intervention,” which makes leadership look good while actual customers walk away frustrated. It’s the kind of metric-gaming that happens when you tie executive bonuses to AI adoption numbers rather than actual outcomes.
Where I work, I’ve seen similar dynamics. Teams are told to integrate AI into workflows without clear metrics for success, and suddenly everyone’s scrambling to bolt chatbots onto existing systems whether they add value or not. The project gets announced, the press release goes out, and six months later nobody can tell you whether it actually saved time or money. Sound familiar?
This connects to a broader pattern I covered in the $3 trillion AI question — when will any of this actually pay for itself? The numbers don’t add up, but asking the question in a public meeting has become career-limiting.
The Religious Fervor Problem
Here’s where Suresh’s essay gets truly unsettling. He describes a corporate environment where continued employment has started to require repeated professions of belief in the transformative power of AI. Not practical discussions about where AI might help — actual declarations of faith. Non-technical executives produce technical AI strategies for billion-dollar organizations without ever having used ChatGPT. Engineers are being evaluated on how many tokens they consume, with higher consumption being treated as better performance.
The response from actual engineers has been predictable and darkly humorous: they’re “AI-washing” their work. They do their jobs the same way they always have, then claim Claude did it. Some have even set up LLMs prompting themselves in semi-plausible loops so their token consumption looks legit while they watch Netflix. Not a single one has been caught.
I’ve written before about the quiet AI cost correction that’s already underway — companies like Microsoft are pulling back from expensive AI integrations even as the public narrative stays relentlessly bullish. The gap between what’s said in earnings calls and what’s happening on the ground is growing wider by the quarter.
The Coordination Trap Nobody Can Escape
One of the most insightful parts of the essay describes the prisoner’s dilemma that executives now find themselves in. A Fortune 500 executive confided to Suresh that their company’s customers were making absurd claims about 100x productivity gains. If the vendor’s executives said those gains weren’t plausible, they’d undermine their own customer’s credibility — which could cost them a contract. And getting a contract cancelled because you wanted to be honest is a great way to get fired.
So every executive is simultaneously terrified of being the first to tell the truth, and the collective delusion metastasizes. Board members admitted they were skeptical but felt their positions required them to demand AI investment anyway. One organization that was described as a “decade-old multi-billion dollar” company is now branding itself as “AI-native” — whatever that means.
This dynamic mirrors what I see in government IT procurement. Once a technology trend reaches a certain critical mass of hype, it becomes impossible to question a proposal that includes it. You can’t say “this AI component adds no real value” because the person who proposed it has already staked their reputation on it. The incentives are structurally broken, and nobody wants to be the one holding the truth when silence is safer.
When Even The Demos Are Dangerous
Suresh describes a revealing experience with Snowflake’s Cortex AI chatbot. His team ran a demonstration for a lukewarm client — with full caveats that the tool wasn’t production-ready — and the result was immediate and alarming. The client, who had been hesitant to buy the team’s main offerings, suddenly wanted to buy Cortex immediately. They were willing to set aside millions of dollars in achievable value from non-AI solutions just to get their hands on a flashy chatbot demo.
Suresh’s team was so disturbed by this reaction that they removed Cortex from their demonstrations entirely. His analogy is perfect: “Doctors don’t walk around showing off cool pills that they’d never prescribe.”
This explains so much of what I see in the broader tech landscape. The demo is seductive, the actual deployment is disappointing, but by the time the disappointment sets in, everyone involved has already moved on to the next flashy thing. The two AI economies — the world of billion-dollar compute deals and the world of open-weight models — only amplify this. When capital is this abundant and the stories are this compelling, critical thinking gets priced out.
What This Means For Developers And Managers
If you’re a developer or manager caught in this environment, Suresh offers practical survival advice that rang true to me:
For anonymous feedback, use polls. Ask people to rate AI project confidence on a scale of 1 to 10. He’s consistently seen bimodal splits — half at 3/10, half at 8/10 — even on projects that are years late. Bringing that data to leadership can crack the illusion.
Never question broad AI claims in group settings. When someone says “AI is changing everything,” just let it pass if your goal is to fix a real problem. The challenge can only come after you’ve built trust privately.
Start looking early. If you’re being evaluated on token consumption or asked to review 2,000-line AI-generated PRs, assume the organization will burn you out. Start the job search while you still have energy.
Lie to survive if you must. Suresh’s blunt advice: “If you work in the fire service and need money to stop a puppy from catching fire, just lie. Add a $10,000 AI chatbot to your project, exclusively discuss that part in meetings. Save that puppy.” I wouldn’t necessarily endorse dishonesty, but I recognize the impossible position people are being put in.
I talked about this dynamic in my piece on the hidden costs of premature AI integration — if you’re paying for AI twice (once in subscription costs, once in productivity lost to bad implementations), you’re not alone.
The Bubble Will Burst
The most reassuring thing about Suresh’s essay, despite its bleakness, is his conviction that this will pass. Bubbles always do. The blockchain hype faded (though not everyone escaped that trap), and this one will too. Organizations that are fully captured by AI mania will either collapse or eventually be forced to reckon with reality when the promised productivity gains fail to materialize on any balance sheet.
In the meantime, the best thing you can do is protect your own judgment. Limit your consumption of AI hype news. Find a small company where the incentives are still sane. Build real skills that don’t depend on the latest frontier model. And when someone asks you about AI, my new favorite response from the essay is worth stealing: “Oh, that stuff is pretty overblown” — and then change the subject.
Read Suresh’s full essay at Ludicity. It might make you feel saner for harboring your own doubts.