Background & Context§
The essay "AI Mania Is Eviscerating Global Decision-Making" by a consultant (who runs point on sales and technical engagements for a data infrastructure firm) details a brutal first-hand account of how AI hype has corrupted strategic decision-making across Fortune 500 companies, government institutions, and mid-market firms. The author, drawing on ~300 professional conversations and direct involvement in dozens of deployments, argues that the current AI fervor has created a "mass psychosis" where rational discourse is impossible, honest metrics are suppressed, and employees are forced to feign enthusiasm for tools that deliver zero real value. The piece is particularly damning because it comes from someone who is neither a pure AI skeptic nor an advocate—rather, a practitioner who has watched the entire ecosystem devolve into a theater of lies.
The News: What Happened Exactly§
The consultant reports that every AI project their team has observed in the past 18 months has failed—a 0% success rate across both projects they participated in and those they merely observed. This includes internally-facing chatbots (which employees never use due to poor documentation) and customer-facing chatbots (which either frustrate users or quietly drop requests into a void). One emblematic example: a Mitsubishi voice bot promised the author a call-back after reporting a car failure—six months later, no call came. The author suspects the request was counted as "resolved" by the bot, artificially inflating success metrics.
Executives are openly lying about results. The consultant describes a Fortune 500 executive who, in a private meeting, admitted that the company's public claims of "100x productivity gains" were pure sales fluff. However, the exec dared not correct the record because doing so would embarrass their customers' executives—who had made equally absurd claims—and risk losing enterprise contracts. This creates a "mutual hostage" situation: no one can admit the truth without triggering a chain reaction of firings and contract cancellations. The essay notes that "heads of AI" at billion-dollar companies have privately confessed their roles are "totally fraudulent," taken only as the sole available promotion path.
Employees are gaming AI usage metrics to avoid termination. Engineers report "AI-washing" their work—doing tasks manually but claiming Claude or ChatGPT did them, because managers reward high AI token consumption. One developer shared that they spin up a parallel repository and ask an LLM to rewrite code in a different language (e.g., Go to Zig) just to burn tokens, while actually doing the real work by hand. Token leaderboards—where higher consumption is considered better—have become a perverse incentive, with employees setting LLMs to prompt themselves in endless loops. Not a single such employee has been caught, even when the output was obviously unusable.
Sales and consulting engagements are poisoned by AI demand. The author's firm removed Snowflake's Cortex AI demo from their sales pipeline after every lukewarm client who saw it suddenly demanded to buy it immediately—despite the firm explicitly warning that the tool was not production-ready (Snowflake itself reports ~92% best-case accuracy, which is unacceptable for CFO-level decisions). The consultant compares it to "a dark and terrible force seizing control of their limbs" and forcing them to hand over credit cards. They now decline all sales where a client expresses more than fleeting curiosity about AI, because such clients inevitably become irrational, cult-like, and legally risky to work with.
Historical Parallels & Similar Incidents§
This pattern eerily echoes the dot-com bubble of 1999–2000, where companies appended ".com" to their names and saw stock prices soar despite having no viable business model. Then-CEOs were lionized for "internet strategies" that boiled down to burning cash on unprofitable customer acquisition. Skeptics were mocked as Luddites, and analysts who questioned valuations were silenced by investment banks that profited from hype. When the bubble burst, trillions of dollars evaporated, and it became clear that most of the proclaimed "productivity gains" were fabricated by companies that had simply moved existing processes online without adding real value. Today, we see the same dynamics: AI strategies are mandated from the top, metrics are gerrymandered to show success, and dissidents are purged.
A more recent parallel is the crypto/NFT mania of 2021–2022. During that period, companies like GameStop, MicroStrategy, and even Tesla adopted Bitcoin as a core treasury asset, while countless startups pivoted to "blockchain" without a clear use case. Employees were pressured to promote cryptocurrencies internally and externally; those who questioned the technology's environmental impact or speculative nature were ostracized. Executives at firms like Coinbase and OpenSea made promises about "decentralizing everything" that never materialized. When the market crashed, many organizations that had bet the farm on crypto (e.g., Three Arrows Capital, FTX) collapsed, and the broader industry was exposed as a house of cards. The AI mania mirrors this: the underlying technology (LLMs) is genuinely powerful, but the way it is being adopted—without rigorous validation, amid mass delusion—is almost identical to crypto's irrational exuberance.
A third, more technical parallel is the enterprise NoSQL boom of the early 2010s. Companies like MongoDB and Cassandra promised to replace relational databases with "schemaless" alternatives that could scale infinitely. Many organizations migrated their core transactional systems to these new databases, only to discover that they traded ACID guarantees for eventual consistency and operational complexity. Projects failed spectacularly, but vendor marketing and executive credulity kept the hype alive for years. The lesson was that adopting a novel technology without understanding its failure modes is a recipe for disaster—a lesson the AI industry seems determined to repeat. Today's AI failures often stem not from the models themselves, but from organizations' inability to run software projects effectively, compounded by the extra risk that LLMs introduce (e.g., hallucination, prompt injection, unpredictable behavior). As the essay notes, "very few companies are so good at shipping software that they can afford the extra risk profile."