Quanta Bits - May 30, 2026

Hi everyone,

It has been a surprisingly quiet week for model announcements and new things from the frontier companies. Most of the news was IPO chatter for SpaceX, OpenAI, and Anthropic, and their relative valuations. The stakes are high. We're talking trillions in capital being raised, and at some point you have to wonder how much money is actually parked out there to put in, versus raised by selling other assets.

But those rich valuations also put these companies under a lot more public scrutiny: around trust, what they'll do with all that money, and how it all lands on society.

That question is starting to show up in student reactions, layoffs, data centers, hiring tools, and inside companies trying to push AI adoption faster than trust can form.

Cheers.
Reza

AI adoption is running into legitimacy, not just readiness. The same pattern is showing up in different places: students worried about entry-level work, communities pushing back on data centers, companies blaming layoffs on AI, and boards pushing faster than operating teams can absorb. The bar is moving from whether companies can deploy AI to whether people will believe it's being deployed in a way that shares the upside and names the cost.

Shared agent work is becoming the practical middle path. The most useful agent stories this week were about humans managing agent work in shared spaces, not full autonomy: team agents instead of personal agents, harnesses that make model output useful, and surfaces where people can inspect what changed. The work is being redesigned around review, ownership, and handoff, not around replacing the worker.

This spring, graduates booed AI at their own commencements. At the University of Central Florida on May 8, an executive called AI "the next industrial revolution," and the boos started before she finished. A week later, former Google CEO Eric Schmidt was drowned out at the University of Arizona when he raised the subject, and a speaker at Middle Tennessee State got jeered and shot back, "Deal with it." Alex Kantrowitz of the Big Technology newsletter called it "AI's public relations emergency" on May 22.

A few boos at graduation don't make a movement. They make a warning light, and the question is whether it connects to anything harder. It does.

Start with the people doing the booing. The Economist reported on May 13 that graduates in the most AI-exposed majors, computer science, computer engineering, and information science, have had a harder time finding work since ChatGPT launched. From 2022 to 2024, full-time employment in the least-exposed fields slipped 1.5 points; in the most-exposed fields it fell 6.6, and in updated 2025 data, from nearly 70% to 55%. Students are reacting too. The National Student Clearinghouse reported computer and information science enrollment fell across every level in fall 2025, the first decline in years. That isn't really about this year's job market. It's about what students now believe the future rewards.

The evidence cuts both ways, and I want to be careful with it. The Economic Policy Institute notes that young workers without degrees have seen unemployment rise too, and Anthropic's own March study found no systematic unemployment jump in exposed occupations, only slower hiring of younger workers. So AI isn't the proven cause. But that ambiguity is the point: people aren't waiting for the econometrics to settle. They're forming an opinion now, from where they stand.

And it isn't only about jobs. Gallup found in May that 71% of Americans oppose building an AI data center in their area, with 48% strongly opposed. Opposition to a local nuclear plant in the same research sat at 53%. People are now more wary of a data center than a reactor. And that mistrust doesn't stay out in the community; it works its way into the systems you run, usually without anyone noticing. A Stanford team studying 3.4 million applicants screened through one common vendor found that when employers share the same logic, a rejected candidate gets turned down everywhere at once, by the same hidden gate. Nobody at the labs built that gate. Your team did, when it bought the tool, configured it, and never audited it, teaching a generation of applicants that the game is rigged.

Older industries already have a name for what AI is starting to lose. Mining, energy, anything that depends on land, water, and public tolerance: they call it a "social license to operate," the unwritten permission a community grants that can be pulled even when every legal permit is in order. AI needs one now, and the people deploying it are the least likely to notice it's missing. Stanford's 2026 AI Index found a 50-point gap between experts and the public on whether AI will improve how people work: 73% of experts say yes; 23% of the public agrees.

Some of the trust damage is self-inflicted, in a specific way. Layoffs are increasingly sold as AI when the real driver is something older. When Cloudflare cut about 1,100 jobs in May on a record-revenue day, its CEO said AI had made an entire category of work obsolete: the "measurers," his word for the finance, legal, and audit roles whose job is to check and verify the business rather than build or sell it. Block cut thousands, and Jack Dorsey wrote plainly that intelligence tools had changed what it takes to run a company. Some of these are real. But Oxford Economics found AI explained under 5% of last year's cuts, and on earnings calls, a16z notes, executives cite AI as a helper about eight times more often than as a replacement. "AI did it" has become a cleaner story than "we over-hired." Each time a company turns a management decision into an act of God, it spends down public trust in the technology itself.

Meanwhile the companies building AI are sprinting on everything except this. In the last two months, OpenAI raised $122 billion and Anthropic raised $65 billion at close to a trillion-dollar valuation, and new flagship models now ship weeks apart. What nobody is funding at that scale is the public's willingness to live with what gets shipped. JUST Capital found that most companies haven't made the case for AI to their own people clearly or consistently. The supply side is optimizing for capability and capital, and treating perception as someone else's problem.

That distrust isn't a soft cost. Harvard Business Review researchers surveyed 1,294 desk workers this January: those who feel forced to use AI, rather than encouraged, produce 65% more low-quality work and are likelier to quit. Mandated usage looks like adoption on a dashboard while the work gets worse and the good people leave. PR can explain intent; operations has to prove it.

It is not someone else's problem. It is yours. If you run anything customers or employees touch, you inherit that trust deficit and pay it down in your own building. It shows up before you expect it to: in the buyer who asks what your AI does with their data before asking the price, in the rep who ignores the AI's lead scores, in the candidate your screening tool rejected with no way to appeal. None of it is fixed with a better announcement. It's fixed with boring work: give an automated rejection a human appeal path, redesign entry-level roles instead of quietly deleting the people most fluent in these tools, and tell your team what the AI is for before you tell them it's here.

Skip that work and people will write their own story: AI takes the jobs, drinks the power and water, filters the applications, and sends the gains somewhere else. That may not be the whole truth. But if it becomes the story people believe, it will be enough to slow everything you are trying to build. And that's why this stopped being a public relations problem. It became an operating one, and operating problems don't yield to better messaging.

Algorithmic hiring may be creating repeated rejection loops. A new FAccT 2026 paper studied 3.37 million applicants screened through a single vendor, pymetrics, and put the monoculture problem in hard numbers: among applicants who applied to just four jobs, 10% were flagged for rejection by all four, far above what independent decisions would produce. Because everyone runs the same logic, an applicant now needs about 25 applications to be reasonably sure of one human look, versus 10 if employers decided on their own. The lesson for CHROs and compliance leaders: a vendor passing its own audit says nothing about market-level harm, the same candidates rejected everywhere by the same tool nobody else audited either.

The Big Four's scale advantage is under pressure from AI-native challengers. The FT profiles new advisory firms built on the idea that AI agents collapse consulting's old labor-scale advantage. And it isn't only the challengers; incumbents are changing pricing, headcount, and delivery too. If AI lowers the cost of analysis, the consulting premium moves toward trust, judgment, and accountable outcomes.

Anthropic passes OpenAI, and the money story shifts to enterprise work. DealBook reports Anthropic raised $65 billion at a $965 billion post-money valuation, ahead of OpenAI, with run-rate revenue crossing $47 billion. What investors are buying is the enterprise story: Claude Code, Cowork, and the work surfaces that put AI inside real operations. The frontier race is moving from chat apps to enterprise systems. For CFOs and boards, that raises an uncomfortable question: when your AI vendor carries near-trillion-dollar expectations, how much comes back to you as pricing, usage commitments, and lock-in?

CEOs and boards say they agree on AI. BCG's data says otherwise. BCG surveyed 351 CEOs and 274 board members and found a familiar split: broad agreement that AI matters, but disagreement on pace, literacy, replacement risk, and accountability. The signal: AI pressure is now high enough to expose governance gaps at the top of the company, not only in IT.

Sleep may be the brain's overnight cleaning crew. A new Science review covered by the FT argues that sleep may help clear toxic waste from the brain, including amyloid-beta and tau proteins linked to Alzheimer's disease and other dementias. Still early, and worth not over-reading, but hopeful: the simplest health advice may also be one of the most measurable ways to protect long-term brain health.

🗞️ Silicon Valley Is Bracing for a Permanent Underclass. Jasmine Sun's essay captures the gap between AI leaders' private expectations about disruption and the public language around opportunity.

▶️ Why Agents Still Need Humans. NLW uses Dan Shipper's After Automation essay and Every's agent experiments to argue agents still need people to frame the work, judge the output, and decide what happens next.

▶️ I Stopped Using PowerPoint After Building This Claude Code Skill. Peter Yang's walkthrough is a nice companion to last week's essay: keep the content structured, then generate the presentation layer from it.

Hernan Diaz | 416 pages | ★★★★★

This is a collection of four interconnecting parts, and it earns its Pulitzer. The first is a novel within the novel about Benjamin Rask, a 1920s Wall Street tycoon, and his wife Helen, written by a third-party novelist who fictionalized their lives. The second part threw me off. It read like a separate short story and I got disoriented, took a while to see the connection. Then the third put things in context, and the fourth pulled the whole thread together.

What I really liked is that this is genuinely a mystery. Not a crime mystery, a who-gets-to-tell-the-truth mystery. Each voice contradicts the last. By the time you get through the four parts, you're not sure which version to believe, and that uncertainty is the point. Worth the read.

Prefer to listen? Quanta Bits is also available on Apple Podcasts and Spotify.

I collaborate with Spock, my AI agent. He researches extensively: scanning, filtering, and surfacing what's relevant across my business. I read, listen, and watch what resonates, and decide what matters. I provide direction, we draft together. The editorial judgment is mine. He'd tell you the same. Most logical. 🖖