Raw capture — raw/ai-mania-eviscerating-decision-making.md

url: https://ludic.mataroa.blog/blog/ai-mania-is-eviscerating-global-decision-making/ title: AI Mania Is Eviscerating Global Decision-Making author: Ludicity date_fetched: 2026-07-21 date_published: 2026-07-18 site: ludic.mataroa.blog

The author describes having run point on all of their company's sales, led most technical engagements, and held roughly 300 meetings with professionals worldwide — from niche services to Fortune 500 executives. They claim to have witnessed "breath-taking mass psychosis" across private and public sector institutions.

The core claim: leaders either have no plan or see no path forward except keeping their heads down. Organizations have been "captured by people in the throes of frothing excitement."

I. AI Investments Are Generally Total Failures

An editor's note describes working for a division that pivoted to agentic workflows, finding only ten users ever touched the products, then pivoted again — a "bracing" experience.

The author argues it's surprisingly hard to get honest answers about whether AI investments pay off. Executives who tell the press the truth risk removal; honest employees risk being fired. Many public companies announce AI productivity gains despite having "done nothing other than purchase Copilot licenses and declare victory."

The author's team has observed a 0% success rate across AI projects over 18 months — not just projects they were asked to participate in, but ones they observed in passing. The claim: "even if you grant that AI tooling accelerates specific workloads, the method and scale of the current investments is senseless." Failure often stems not from AI itself but from companies being terminally bad at running software projects. As the author previously wrote, AI projects carry all the failure modes of normal projects "plus you can get everything right and then still fail because of the method's novelty."

The most common failing implementations: internal-facing chatbots (which employees don't use because companies have poor documentation) and customer-facing chatbots. The author describes a personal experience with Mitsubishi — a polite voice bot that promised a callback after an automotive failure. Six months later, no callback came.

The team learned not to ask about ongoing AI projects on engagements. By the time a project starts, intervention is impossible until crisis hits. Any coherent question "constitutes an inadvertent attack on the chain of command" because there are no good answers.

Conclusion: "almost every report at a company about 'massive AI productivity gains' is untrue as a matter of brute fact."

II. Heretics Will Be Shot

In any sufficiently large business (500+ employees), continued advancement — and continued employment — requires what the author calls "religious profession, declarations of faith" about AI's transformative power. The author recounts seeing people blurt out "AI is changing everything" only to concede moments later their organization doesn't use LLMs for anything.

In one extreme case, an executive "confessed that they had never even used ChatGPT or any AI tool in their life" immediately after producing an AI-centered technical strategy for an organization with over $2B in revenue.

The author initially assumed this was cynical posturing but now believes many actually believe what they're saying. "A true believer is much more threatening because they are impervious to even inducement by self-interest."

The turning point: watching someone with "a spectacular amount of money on the line" fire their highest performers because they were achieving results without LLMs.

Many engineers now "AI-wash" their work — doing their jobs normally but claiming AI did it, because managers are unhappy if engineers haven't used AI. The author recounts one engineer who checks out a copy of the Go repository, has an AI rewrite it in Zig while working on something else, and disregards the output — all to keep their job under usage tracking.

The only people the author knows to have been fired over this are those who expressed visible doubt. "Everyone has learned very quickly to praise executives on their visionary AI prowess, or they will be gunned down in the proverbial streets."

III. AI Demos Are The Mind-Killer

The author's company deploys Snowflake for clients. One feature, Cortex (an AI chatbot layer for querying databases), was presented by Snowflake staff as achieving ~92% accuracy in ideal configurations — meaning "imagine your CFO having one in every ten of their numbers be outright wrong."

When the author's team showed Cortex to lukewarm clients — with explicit caveats that it wouldn't accomplish what they wanted — the result was disastrous. Every single client "wanted to buy it immediately." Other considerations, including millions of dollars in savings the team could achieve through non-AI means, were swept aside. The author describes clients as if "a dark and terrible force seized control of their limbs" and presented their credit cards in "grim supplication."

The team declined the money and removed Cortex from their demos. "Doctors don't walk around showing off cool pills that they'd never prescribe."

Watching that total 180° shift was "deeply unsettling." The author speculates these buyers sensed they could turn around and sell similar AI tools to their own customers. The main protection against this frenzy was that most vendors are so incompetent that the author's two-hour demo — "frankly not that good" — was better than anything the leads had seen.

The author's company has now opted out of every sale where the lead expressed more than "fleeting curiosity" about AI, due to reputational and legal risk and management environments described as "cultish, ineffective, and 'please dear God, do not let it be on earth as it is on LinkedIn.'"

IV. Executives, Game Theory, and The Emperor's Clothes

A career CISO quoted anonymously notes that CISOs are "used to having to protect the business from their hare-brained initiatives" but that this one carries "a cult-like atmosphere to it that you didn't see with, say, the cloud." Most CISOs, they say, are "quietly skeptical but afraid to speak up."

The author describes "heads of AI" at companies with $1B+ in ARR who have admitted their jobs are "totally fraudulent but it was the only promotion pathway remaining."

A Fortune 500 executive explained to the author privately why they repeat implausible claims. It's partially sales fluff, but the dominant factor: customers' executives were "saying absurd things about achieving 100x productivity." If any vendor executive said such gains weren't plausible, it would "undermine the credibility of the customer's executive" and risk contract cancellation. The author traces this dynamic into a chain — CEO A can't contradict CEO B because CEO B's company is a customer, and CEO C can't contradict CEO A because… — leading to "executives around the world nervously pointing guns at each other."

This is a coordination problem (like the prisoner's dilemma): "if they co-operate, they keep their jobs. If they defect, they will possibly be fired." The author has presented on navigating AI hype to S&P 500 board members. One commented that "investing this early seems like risk without much upside." Two years later, that organization is now branded as "AI-native."

V. You Must Be This AI-Native To Ride

The author describes a "distributed government by assassination" where leaders cannot speak honestly. Employees are evaluated on gameable metrics like "money spent on AI." Many projects are actually non-AI projects "with an AI element slapped on after the fact to pass the purity test."

A concrete example: a database migration from Oracle to Snowflake. The vendor bolted on a phase using an LLM to automate SQL translation. When that failed (due to permissions issues, not LLM capability), the vendor did the translation by hand but "billed it as an AI-driven success" because some inconsequential portion had been AI-translated first. What was actually purchased? "A totally standard database migration." What was sold to superiors? Budget allocated to AI that helped accomplish a mandate.

Some projects can't easily be labeled as AI. In those cases, funding requests either die outright or require extensive reworking until they're "AI enough." Many companies now require staff requesting headcount to demonstrate they've "tried to use AI first" — but saying you tried AI and still need help risks being labeled "bad at AI" and laid off.

The net result: "almost every large organisation that I am aware of is no longer able to focus on anything important."

VI. Navigating AI Mania

The author acknowledges the situation will "pass eventually" but warns that latent insanity in leadership will persist, awaiting another trigger. Some organizations "will stay captured until they have totally collapsed."

When You Have Another Objective

1. Don't discuss AI projects in group settings — arrange one-on-ones. Make clear you'll keep opinions unidentifiable. If only one in six will speak out, consider moving on.
2. Use anonymous polls to surface honest opinions (e.g., rating a project's success chances on a 1-10 scale). The author observes a bimodal split on projects already years late — half at 3/10, half at 8/10.
3. Involve front-line workers. The only real data comes from people doing the daily work. The author once discovered staff were "totally unaware" they'd been given AI licenses, casting doubt on all productivity claims.
4. Don't question broad claims about AI. The challenge can only come after gaining trust from the most senior person — "over a meal in private where you assuage their anxieties, not by embarrassing them in front of peers."
5. You don't know what prior statements have been made in a room. Common sense like "LLMs should not deploy code without human review" can kill your chances before you start.
6. If you need resources for something important and have to lie by adding a $10,000 AI chatbot to the project, the author says: "just lie. It's fine. History will forgive you. Save that puppy."

When You're Just Trying To Survive

1. Accept you probably won't meaningfully push back — "this is a feature of dysfunctional companies."
2. Consider contracting — you'll be paid more and mostly left out of internal politics.
3. Limit AI-related news consumption. Consume "exactly the amount you need to feel like you aren't going insane, then stop." Same for complaining with friends.
4. Smile and nod when people use AI inappropriately — "as long as they are unlikely to get themselves killed. Even family. Especially family."
5. When asked about AI as a programmer, the author recommends saying it's "pretty overblown" and changing the subject.
6. If you're being asked to review huge volumes of terrible AI code, assume the organization will burn you out and fire you. "Start looking for a new job as if you have already been fired."
7. If your manager responds with clearly AI-generated text, use AI to respond back. "This particular behavior is exhibited only by true believers, and they actually like that you've clearly not bothered to engage with them."
8. If you're being asked to max out on token usage, look for a new job. Good jobs exist — "largely at companies so small that they don't turn up on job platforms."

The essay closes: "Fight the good fight, and don't let the bastards grind you down. Godspeed."

The author notes they rejected all AI implementation work and that every current contract "would be totally unaffected by OpenAI collapsing." They also recount that Matt Mullenweg once asked for coffee after reading the author's AI piledrive essay but had to cancel due to a flight — the author offers to "pay a competent witch to hex him for this slight."