Beating the AI Doom Cycle

May 18, 2026 · Episode Links & Takeaways

Author

AI Daily Brief
May 18, 2026

MAIN STORY

The AI Doom Cycle

Today's episode was one extended piece rather than the usual headlines-plus-main format — because every story worth covering this weekend turned out to be part of the same larger pattern. The AI doom cycle is a five-stage emotional and cognitive journey people go through in their relationship with AI: skepticism and disbelief, AI psychosis (peak conviction that AI changes everything), doom desperation, real-world recalibration, and finally enlightened anxietment — that portmanteau of anxiety and excitement that is the healthiest, most productive place from which to actually engage with AI's implications for policy and work.

CYCLING THROUGH

Skepticism and Disbelief

This phase is increasingly on the wane — most people who've actually used a modern model have moved past it. The ones still here are either running on outdated priors about what AI can do, or have made AI skepticism their professional brand. Still, the anger at the people building this technology is getting louder, and nowhere is that more visible than on college campuses.

AI Psychosis

Peak conviction that AI changes everything — which, once you start seeing the implications, has a way of tipping you straight into the next stage.

Ken Griffin
The prominent skeptic who went home "fairly depressed."
Back in January at Davos, Citadel CEO Ken Griffin said AI was all hype engineered to justify the infrastructure build-out — and that the AI-generated stock research his team had tested was "all garbage." That view has been updated. Speaking at Stanford Business School this month, Griffin said AI had become "profoundly more powerful" in nine months and that Citadel was seeing 15–25% productivity gains from work previously done by people with master's and PhDs in finance. But the conversion came with a gut punch: Griffin said he went home "fairly depressed" because he could see how dramatically this was going to impact society.

Doom Desperation

Part of what makes doom desperation so potent right now is that a lot of it is coming from the people building the technology — which is also why the animosity from people outside the tech bubble is so white hot. If this thing is going to be bad for us, why are you building it?

Mustafa Suleyman
18 months to automate all white-collar work — or so the headline said.
Fortune grabbed attention with a headline claiming Microsoft AI CEO Mustafa Suleyman had given white-collar work an 18-month expiry date. The actual story was more nuanced: Fortune was resurfacing quotes from a February Financial Times interview, and the article's main point was that three months later, the wholesale replacement of white-collar workers Suleyman implied isn't happening. The tools keep improving, but it's task automation, not job replacement.

Dario Amodei
Old quotes getting new traction — which is interesting in itself.
The Wall Street Journal resurfaced video clips of Anthropic CEO Dario Amodei from a WEF interview earlier this year, in which he put specific numbers on his predictions: 10% overall unemployment, 50% unemployment in entry-level white-collar roles, and software costs going to zero. What's notable is that these aren't new comments — they're from months ago — but they're landing harder now than when they were first made. Some in the tech industry are skeptical of the framing: as one commenter put it, this narrative is what allows Anthropic to raise $30 billion at a $900 billion post-money valuation.

Deedy Das
11 million people saw the SF doom post.
Menlo Ventures partner Deedy Das went viral on Friday with a portrait of the current Bay Area tech vibe: a 10,000-person cohort at the frontier labs and NVIDIA sitting on generational wealth, while everyone else watches from the outside wondering if they missed their window. The post maps four moods — career disorientation, malaise about work's future, paralysis among middle managers, and even a strange purposelessness among those who've already made it. The reaction was predictably mixed. Bucco Capital's summary of the critical response: "Life has wasted success on the people described in this post."

Commencement Speakers Are Getting Booed

One of the media's favorite themes right now is graduation speakers getting booed off stage for mentioning AI. On Friday, Eric Schmidt was loudly jeered at the University of Arizona — even as he was actively acknowledging that the tech industry had failed younger generations during the social media era. The boos peaked when he said the question isn't whether AI will shape the world, but whether graduates will shape AI. Days earlier, Gloria Caulfield — a real estate executive with no tech background — got the same treatment at UCF just for calling AI "the next industrial revolution." It would be too simple to say graduates hate AI. As Trevor Garcia put it, they're not anti-technology; they're anti being told to be excited about something that feels like it was built for everyone except them.

Real-World Recalibration

This is the step back from the loudest voices and most media-friendly extremes to look at what's actually happening — including evidence that confirms some concerns alongside evidence that complicates the doom narrative.

Meta Layoffs
About 8,000 people told Wednesday is their last day.
Meta has told around 10% of staff their roles are being eliminated — roughly 8,000 people. The official explanation is running the company more efficiently and offsetting other investments, which effectively means payroll is being converted into infrastructure spending. Wired reports that looming layoffs combined with newly installed screen-tracking software (said to be for AI training) have driven morale to record lows. One policy staffer described the vibe as "over it — lack of connection to the mission, upcoming layoffs, American employees being used to train the AI models that will replace them."

Token Budgets and the End of the AI Subsidy Era
The token spend party is ending. ROI is now required.
The tokenmaxxing era — where companies encouraged employees to experiment freely with AI through leaderboards and unlimited access — is crashing into physical reality. Back in April, Anthropic quietly shifted enterprise customers off the $200 flat-rate Claude Code subscription and onto a $20/seat rate with all usage billed separately. GitHub Copilot is making the same move to usage-based billing in June. Meanwhile, Microsoft has reportedly started canceling Claude Code licenses and pushing employees to GitHub Copilot CLI — officially framed as dogfooding their own product, but sources say it's also financial. The GitHub Copilot subreddit captures the scale of the previous subsidies vividly: one user found that their current $39/month plan would cost $5,851 under usage-based billing. Token budgets are now showing up across the industry, and the ROI mindset is returning.

Salesforce Spending $300M on Tokens
Marc Benioff has a different response to rising token costs: bigger budgets.
On All-In over the weekend, Salesforce CEO Marc Benioff said his company will likely spend $300 million on Anthropic tokens this year — and he's enthusiastic about it. His argument: coding agents are letting Salesforce implement and sell software simultaneously in ways that were previously impossible.

AI Can Cost More Than Human Workers
An expensive, capital-intensive technology demands cost-benefit analysis.
An Axios piece from a few weeks ago made the point explicitly: for some teams, compute costs already exceed personnel costs. NVIDIA VP of Applied Deep Learning Bryan Catanzaro noted that for his team, "the cost of compute is far beyond the costs of the employees." This cost reality is going to change the timeline calculus on how quickly companies can even theoretically automate their way out of headcount.

Enlightened Anxietment

The end state isn't Pollyannish — it's the place from which more specific, more interesting, and ultimately more useful conversations become possible. When AI doom is treated as inevitable, policy collapses to UBI or nothing. When the disruption is understood as real but discrete and uneven, there's room for actual policy.

Jensen Huang at Carnegie Mellon
The only AI commencement speech that didn't get booed.
Jensen's CMU speech barely got any press compared to Schmidt's booing — which is itself a story about what gets covered. But the difference in reception was stark. Jensen's message was that AI amplifies rather than replaces, and that the demand for engineers and radiologists has continued to grow even as AI writes more code and analyzes more images. Whether or not one agrees, it's a message of agency rather than inevitability — and Joanna Stern noted that she found the broader campus pushback "unsurprising and even encouraging" because it means there's actual debate.

Mark Cuban's Token Tax
A policy idea worth actually discussing.
Cuban proposed a federal tax on tokens at the provider level — less than 50 cents per million — that he argues would accomplish four things: push providers to optimize tokenization and efficiency, reduce energy use, generate up to $10 billion/year in federal revenue that could grow 30–100x over a decade, and create a funding mechanism to respond to AI's unexpected downsides. The subsequent discussion thread — including a back-and-forth with Palmer Luckey — was more substantive than most AI policy discourse.