AI Mania Is Eviscerating Decision-Making — Ludicity's Hermit Tech Essay Decoded
Jul 18, 2026: Ludicity (Hermit Tech) claims 0% AI project success across 18 months of consulting — corporate psychosis, demo frenzies, and token leaderboards. explainx.ai maps the essay, HN critics, and what still works.
Update — July 20, 2026: PsyArXiv lab data on cognitive surrender — AI advice drops "I don't know" from 44% → 3% while confidence surges (Capraro et al. study). Enterprise demo psychosis and individual metacognitive collapse are different desks; both punish honest uncertainty.
On July 18, 2026, an essay titled "AI Mania Is Eviscerating Global Decision-Making" landed from Ludicity — who runs sales and technical delivery at Hermit Tech, a consulting firm focused on data infrastructure and failed-project recovery. Within hours it hit the Hacker News front page (103 points, 40+ comments) and reframed a debate that had been simmering since Mitchell Hashimoto's "AI psychosis" warning in May: not whether coding agents help individual developers, but whether organizations can still make rational decisions about AI at all.
The essay's headline claim is blunt: across 18 months and roughly 300 catchups with professionals — from niche service workers to Fortune 500 executives — Ludicity reports 0% success on AI projects observed, including projects seen only in passing. Footnotes complicate that number immediately: Hermit Tech rejects all AI implementation work, so the sample is not a random enterprise survey. Still, the narrative resonated because it names behaviors many employees recognize but cannot say aloud: faith declarations, token leaderboards, demo-induced buying frenzies, and an emperor's-clothes coordination trap where vendors cannot contradict a customer's 100× productivity claims without risking contracts.
TL;DR — what people are asking after reading the essay
Question
Direct answer
Did the author really see 0% AI project success?
Yes, as stated — but Hermit Tech rejects AI implementation work, so the sample skews toward failure and observation bias.
Is this the same as Mitchell Hashimoto's AI psychosis?
Related but broader — Hashimoto warned about engineering teams; Ludicity describes C-suite mandates, vendors, and procurement across industries.
Do HN readers buy the 0% figure?
Mixed — many validate the anecdotes; top critics cite selection bias and consulting incentives.
Does individual AI tool use contradict the essay?
Not necessarily — HN and Simon Willison emphasize a split: agents help skilled individuals; enterprise "AI projects" often fail for org reasons.
What is the Snowflake Cortex story?
A ~92% accuracy NL-query demo (per Snowflake staff) still triggered immediate buying frenzy despite caveats — Ludicity stopped demoing it.
What is "AI washing"?
Engineers do competent work without LLMs, then claim AI did it because managers require AI usage — or spin token-burn loops to satisfy leaderboards.
What should I do if my org is captured?
Anonymous polls, 1:1 truth-telling, usage metrics from daily users, avoid group challenges to AI faith — or contract to escape internal politics.
explainx.ai's read?
Treat the essay as organizational pathology, not a global success rate; pair skepticism with ROI discipline and honest measurement.
Who Ludicity is — and why the essay spread
Ludicity writes the Ludicity blog and leads sales plus technical delivery at Hermit Tech, a consulting practice known for blunt essays on enterprise dysfunction — database migrations, analytics stacks, and rescuing software projects that went off the rails. The July 18 piece was cross-posted to Hermit Tech's company blog in a more "authoritative" format for sharing with clients.
The author's stated vantage point: over the past year, Ludicity ran point on company sales, led technical components on nearly all engagements, and accumulated ~300 conversations across public and private sectors. That position — consulting without needing to sell AI implementations — is central to both the essay's credibility and its critics' rebuttal.
The opening epigraph is from Mitchell Hashimoto (HashiCorp, Ghostty):
I strongly believe there are entire companies right now under heavy AI psychosis and it's impossible to have rational conversations with them about it.
explainx.ai covered Hashimoto's May 2026 warning in depth: AI psychosis, vibe coding fatigue, and the Cursor generation. Ludicity extends the diagnosis from engineering culture to global institutional capture — banks, hospitals, government, and vendors caught in the same dynamic.
Section I: Are enterprise AI investments failing?
Ludicity's core empirical claim: every AI project observed over 18 months failed — not only engagements Hermit Tech was asked to join, but projects glimpsed during unrelated work. That includes the familiar rollout pattern: internal chatbots nobody uses because documentation is thin, and customer-facing chatbots that polish interactions while hiding broken backends.
The Mitsubishi callback anecdote is the essay's sharpest consumer example: a natural-sounding phone bot promised a swift callback after an automotive failure. Six months later, no call — yet the interaction may have counted as resolved without human intervention in whatever dashboard Mitsubishi runs. Ludicity was planning to buy a car and chose not to buy another from the brand. The metric looked fine; the customer did not.
This is Goodhart's law in a phone tree: when callback avoidance becomes the proxy for success, the proxy rots. explainx.ai's Kaiser nurses vs AI surveillance coverage shows the same pattern in healthcare — empathy scores and handle-time KPIs that optimize the dashboard, not the patient.
The selection-bias counterargument (HN)
The footnote matters: Hermit Tech rejects all AI implementation contracts and has minimized bubble exposure. Critics on Hacker News argued this creates double selection bias:
You only observe projects that are already troubled enough to need consultants.
Your business model includes recovering from failed software projects — failures are literally the lead pipeline.
Ludicity would likely reply that passive observation of unrelated projects still showed failure, and that the essay's value is pattern recognition in executive behavior, not a published success-rate study. Both can be true: the 0% number is not statistically portable, while the chatbot/metrics/demo stories match many readers' experience.
Section II: Faith declarations, mandates, and AI washing
The essay's second movement is darker than failed chatbots: heresy is punished.
In organizations with 500+ employees, Ludicity reports that advancement and sometimes continued employment require repeated professions of faith in AI's transformative power — not proposals for sensible use cases, but declarations that "AI is changing everything." Ludicity describes executives who emit these statements while admitting, moments later, that their org does not use LLMs for anything beyond personal ChatGPT — sometimes on the free tier.
The essay's most alarming anecdote: an employer fired top performers who achieved results without LLMs, because public AI innovation narratives required LLM usage as proof of modernity.
Tech execs are mandating LLM adoption. That's bad strategy.
Ptacek believes skilled developers get enormous leverage from agents; Ludicity believes mandates produce AI washing — engineers who competently ship work, then claim Claude wrote it; optimization experts who run token-burn loops so leaderboards show activity while they watch Netflix; repos where Go code is "rewritten in Zig by AI" to satisfy usage tracking while the human works elsewhere.
That is not an argument against Codex hitting 8 million users. It is an argument against measuring commitment in dollars spent on tokens instead of outcomes delivered.
Section III: Why AI demos are the "mind-killer"
The Snowflake Cortex section is the essay's vendor-side horror story. Ludicity deploys Snowflake analytics databases for clients but avoids Cortex in production. In a Snowflake staff presentation, ideal configuration reportedly reached only ~92% accuracy on natural-language queries against enterprise metadata — best-in-class for the category, but one wrong CFO number in ten.
Still, when Hermit Tech showed a lukewarm prospect a caveated Cortex demo — "this will not accomplish what you want" — the room flipped from ice-cold to buying frenzy. Millions in non-AI value Hermit Tech could deliver was swept aside. Ludicity compares it to a doctor showing pills they'd never prescribe.
Hermit Tech removed Cortex from demonstrations and eventually opts out of sales where leads show cultish AI fixation — reputational and legal risk exceeds revenue.
Readers building internal copilots should hear the parallel to explainx.ai's AI demo final 10% problem: demos compress complexity; procurement memory retains the flash, not the footnote.
Section IV: The emperor's clothes coordination problem
The essay's game-theory section may be its most durable contribution.
A Fortune 500 executive told Ludicity why vendors repeat absurd 100× productivity claims: customer executives at other firms already said it publicly. If a vendor contradicts them, it undermines the buyer's executive, reads as an attack, and risks contract cancellation. Because every large vendor is also a buyer elsewhere, executives worldwide hold guns pointed at each other — cooperate on the narrative or get replaced by someone who will.
Anonymous Heads of AI at billion-dollar ARR companies have written Ludicity saying their roles feel fraudulent but were the only promotion path left.
A career CISO quoted in the essay: most are quietly skeptical but afraid to speak up — same as cloud hype, except with a cult-like atmosphere.
This is where the essay intersects with macro AI economics discourse without agreeing on timelines. Stanford's "We Must Act Now" letter (July 13, 2026) warns AI may transform the economy faster than the Industrial Revolution — 16 Nobel laureates, Jeff Dean, Jack Clark, and 200+ signatories calling for guardrails before displacement. Ludicity's essay is the institutional complement from the inside: even if transformative AI arrives, today's decision-making apparatus may be the wrong instrument to deploy it — captured by demos, mandates, and ungameable narratives.
Those truths can coexist: long-run transformation risk does not invalidate short-run procurement psychosis.
Section V: "AI-native" purity tests block sensible work
The essay argues decision-making has ground to a halt because every proposal must pass an AI purity test:
A standard Oracle-to-Snowflake migration gained a preliminary LLM SQL-translation phase; when automation failed on permissions, humans did the work — but finance heard an "AI-driven success."
Headcount requests require proof you tried AI first; if you tried AI and still need help, you may be labeled "bad at AI" and laid off.
Sensible software purchases die unless someone slaps an AI label on them.
Ludicity's framing matches Tesler's theorem — often paraphrased as "AI is whatever hasn't been done yet." Once a capability works reliably, it stops being "AI" and becomes software. Enterprise "AI projects" therefore chase the moving frontier — chatbots, agents, NL analytics — while boring migrations that actually save license costs fail the purity test unless rebranded.
That is the opposite of explainx.ai's executive ROI framework: buy commoditized capability, build where proprietary context matters, and measure against a specific outcome — not against whether the proposal says "AI" enough times.
What Hacker News added — individual productivity vs enterprise projects
The HN thread did not treat the essay as gospel. Useful refinements:
Selection bias and incentives. Hermit Tech sells failure recovery; rejecting AI work means observing others' failed AI projects while not betting on successes. Fair critique — but many commenters still recognized their own employers in the faith-declaration passages.
Individual vs enterprise. Several engineers argued personal agent use (Claude Code, Codex, Cursor) does accelerate skilled practitioners — consistent with Ptacek and with 8M Codex users. The essay is weaker on this distinction; HN supplied it. Tesler's theorem helps: your best developers using agents is automation; your board mandating an "AI-native" transformation program is organizational theater.
Documentation and chatbots. Simon Willison highlighted Ludicity's point that internal chatbots fail when knowledge is not written down — LLMs are not psychic. That aligns with grounding vs fine-tuning decisions: retrieval quality is a content and permissions problem before it is a model problem.
Physical agents vs LLM hype. The essay focuses on knowledge-work LLM rollouts. Yann LeCun's July 2026 argument — that token predictors face hard limits on continuous physical-world planning — is a separate caution: enterprises chasing "AI-native" branding may be under-investing in embodied robotics and over-investing in chat layers that PNC data suggests most households won't pay for directly.
Ludicity's survival guide — and explainx.ai's additions
The essay closes with practical advice for people inside captured orgs.
When you have another objective
No group confrontations on AI faith — use 1:1 settings; group dynamics punish the first skeptic.
Anonymous polls (1–10 success odds) — Ludicity cites a bimodal 3/10 vs 8/10 split on a project already three years late.
Ground truth from daily users — one client discovered staff did not know they had AI licenses, undermining productivity claims.
Do not challenge "AI is changing everything" until you have senior trust — earned privately, not by embarrassing executives.
Assume prior public commitments — someone may have claimed 100× productivity and cannot walk it back.
Ludicity's darkest pragmatic tip: if you need a puppy saved from a fire, add a $10k chatbot line item and discuss only that in meetings. History may forgive you.
When you are just trying to survive
Contracting pays more and avoids most politics. Limit AI news consumption. If you are drowning in 2000-line AI PRs, job-search now — Ludicity says the ending is predictable. If your manager replies with AI-generated text, match the energy and exit.
explainx.ai adds executive-facing structure without denying the essay's emotional truth:
Acting vs responding governance — agents that write to systems need different oversight than chat
Faith declarations
Outcome metrics tied to revenue, cycle time, error rate — not license counts
Purity tests
Build vs buy on whether the capability is commoditized — ROI framework
What this essay does not resolve
Ludicity does not prove AI cannot work at enterprise scale — only that this cycle's institutional implementation pattern is failing observably in their sample. The essay underweights:
Individual and team-level gains from coding agents where skilled practitioners drive the workflow.
Paid adoption depth among the minority who subscribe — PNC's ~$31/month average among payers and seven-month retention streak.
Counter-examples in documentation, translation, and triage where narrow deployments succeed when scoped honestly.
It also does not replace policy debate. We Must Act Now addresses macro displacement risk; Ludicity addresses micro decision-making pathology. You need both lenses: prepare for transformative economics and stop funding chatbots whose metrics are Mitsubishi callbacks that never call back.
Summary
Ludicity's July 18, 2026 essay describes corporate AI psychosis: 0% observed project success (with heavy selection bias), faith declarations replacing measurement, Snowflake Cortex demos that trigger irrational buying despite ~92% accuracy caveats, token leaderboards that reward burn over outcomes, and an emperor's-clothes coordination trap among executives and vendors. Hacker News validated the anecdotes while challenging the success rate — distinguishing individual agent productivity from enterprise AI theater.
explainx.ai's read: treat the essay as a warning about organizational decision-making, not proof that AI is a fad. Pair Ludicity's anonymous polls and ground-truth interviews with executive ROI discipline, Goodhart-resistant metrics, and honest reading of who actually pays for AI. Mandates are bad strategy — Ptacek and Ludicity agree. What replaces them is measurement, not another all-hands declaration that AI is changing everything.
Essay details, HN scores, and cross-links reflect sources available on July 19, 2026. Ludicity's observations are anecdotal consulting testimony, not a published enterprise benchmark; treat the 0% figure as a rhetorical stake, not a population statistic.
