Anthropic's Digital God, Pope vs AI, Job Loss Narrative Flips, Open Source Crackdown Coming?

2026-05-29 - 95 min - source - Read full transcript

Jason Calacanis (host)David Sacks (host)Chamath Palihapitiya (host)Bill Gurley

Key insights

Pope Leo XIV's first encyclical argues technology takes on the character of whoever builds, finances, and controls it, and calls for AI regulation.

The 235-page, 42,000-word 'Magnifica Humanitas' warns business leaders to safeguard humanity from AI, calls for worker retraining, child safety guardrails, and a ban on autonomous weapons, and was joined by Anthropic co-founder Chris Olah despite lobbying from Amazon, Google, and Meta to soften its language.

ai-regulation-centralization

Sacks agrees the core AI risk is centralization of power but argues government regulation is the more dangerous cure, not the disease.

He warns that an 'FDA for AI' would let government define and expand 'safety' the way social-media 'trust and safety' regulation expanded into disinformation and psychological harm, ultimately becoming a censorship agenda; he prefers antitrust enforcement against monopolistic AI players over a model-approval regulator.

ai-regulation-centralization

Gurley proposes a 'Dr. Frankenstein theory': Anthropic's leadership may sincerely believe they are midwifing a superior, god-like successor to humanity, not just building software.

Citing Chris Olah's 80-page 'Constitution' document and Dario Amodei's 'Machines of Loving Grace' essay (which describes a future 'cybernetic ecology... watched over by machines of loving grace' allocating resources to humans), Gurley argues this reads less like product strategy and more like genuine belief in creating a benevolent deity.

anthropic-motives

Anthropic's public doomerism plausibly functions as regulatory capture: branding itself the 'safe' AI company casts rivals as reckless and can justify monopolistic control.

Chamath frames this as simple game theory: put a few dominant labs in a room, keep external overseers less technically capable than you, and you create an exploitable asymmetry; Sacks agrees this dynamic could lead directly to bans on open models that further entrench the leaders.

anthropic-motives

Sacks predicts the AI-safety rhetoric in Washington is building toward an eventual push to ban open-source/open-weight models.

He points to Anthropic blog posts repeatedly singling out open models as unable to keep guardrails on cyber and bio-threat topics as 'predicate facts' being laid for a future ban, which he argues would hand the technology's future to China rather than eliminate risk.

open-source-vs-closed

Intelligence sovereignty (owning and running your own AI on your own hardware) is emerging as the next frontier of privacy, and Apple is positioned as a dark horse because of its on-device compute strategy.

Chamath distinguishes 'data sovereignty' (can't see my photos) from 'intelligence sovereignty' (can't tell me what to think or interpret my data for me), and argues new Apple hardware (M5, high-memory Mac Studio) makes running your own models locally increasingly viable.

open-source-vs-closed

The AI-jobs narrative flipped sharply in the week before this episode: Goldman Sachs CEO David Solomon and even Sam Altman and Dario Amodei walked back mass-job-loss predictions.

Solomon's NYT op-ed argued AI will automate 25% of work hours, not eliminate 25% of jobs, citing precedents like bank tellers increasing after ATMs; Sacks frames this as vindication of his contrarian January prediction that AI would net create jobs, not destroy them.

ai-labor-market

Sacks cites hard labor-market data against imminent mass AI job loss: unemployment near record lows and software job postings at a three-year high despite near-total code automation.

He cites 4.3% unemployment (below the 5% full-employment threshold), software engineer job postings up 15% year-over-year to a three-year high, GitHub code commits up roughly 14x year-over-year, and a Yale Budget Lab study finding no discernible AI-driven labor disruption in three years, arguing exploding code volume requires more engineers to manage it, not fewer.

ai-labor-market

Calacanis argues layoffs at Meta, Amazon, and Block should be taken at face value as AI-driven displacement, not dismissed as post-COVID overhiring correction.

He points to Meta's 8,000 cuts, Amazon's stated plan to eliminate 600,000 future positions via robotics and AI, and CEOs' own public statements attributing cuts to AI, arguing companies are rewarded by public markets for doing more with fewer people and will keep eliminating roles like drivers, warehouse workers, and middle managers even as new company formation offsets some of the loss.

ai-labor-market

A securities lawyer is warning clients that blaming layoffs on AI ('AI washing') when the real cause is prior overhiring could expose companies to shareholder lawsuits for securities fraud or puffery.

Gurley and Sacks discuss a trial lawyer (Donnie King) flagging this risk, using Jack Dorsey's 50% cut at Block (widely seen by analysts as overstaffing correction rather than genuine AI efficiency) and a same-day Wix layoff note as examples worth scrutinizing.

ai-labor-market

Claude proficiency is currently the single most marketable job skill in the economy, comparable to spreadsheet literacy in an earlier era, though the advantage is likely temporary.

Sacks argues being the only person at a firm who knows how to use Claude confers an outsized advantage today, illustrated by the show's producer building a detailed custom prompt and 'skills document' that lets Claude generate a highly contextual daily briefing from past transcripts and topics.

ai-productivity-tools

A benchmark testing frontier models as financial analysts found top models statistically indistinguishable, raising questions about ROI on incremental AI training spend.

The Rogo eval found Opus, GPT, and Sonnet separated by less than 0.3 percentage points at the top of the leaderboard ('there is no single best model anymore'), prompting Gurley to argue for more open-source connector standards (akin to Kubernetes for cloud workflows) so models become swappable commodities rather than locked-in dependencies.

ai-productivity-tools

Books referenced

Runnin' Down a Dream - Bill Gurley - Gurley's new book, the basis for his fellowship (runningdownadream.org) giving $5,000 grants to people chasing career reinventions; discussed at length re: lifetime learning and AI-era career risk.

Media referenced

Magnifica Humanitas (Magnificent Humanity) - paper - Pope Leo XIV's first encyclical, a 235-page, 42,000-word document on AI that warns technology takes on the character of those who build, finance, and control it, and calls for AI regulation.
Machines of Loving Grace - article - Dario Amodei's blog post/poem envisioning a post-labor future where AI systems allocate resources to humans; cited by Gurley as evidence for his 'Dr. Frankenstein theory' about Anthropic's worldview.
The Constitution - other - An ~80-page document by Anthropic's Chris Olah defining the model's values; Gurley cites it as evidence Anthropic sees itself as building something god-like rather than shipping software.
Vanity Fair profile of Chris Olah - article - Cited for background that Olah was raised evangelical and is now an atheist, relevant to his presence alongside the Pope on AI ethics.
NYT op-ed by David Solomon - article - Goldman Sachs CEO's op-ed arguing the AI job apocalypse is overblown, cited as evidence of a narrative shift toward AI creating rather than destroying jobs.
Rogo financial-analyst AI benchmark - paper - A published eval testing frontier models as financial analysts, quoted for the line 'there is no single best model anymore,' with Opus, GPT, and Sonnet scores separated by under 0.3 percentage points.
Yale Budget Lab labor market study - paper - Cited by Sacks as finding no discernible AI-driven labor market disruption over the past three years.
The Information article on Anthropic's growth - article - Cited showing Anthropic pulling ahead of OpenAI in growth rate, used to argue AI market share may be heading toward monopoly/duopoly.

Companies

Anthropic - Central subject of the episode: its doomerism, lobbying, Chris Olah joining the Pope's AI encyclical discussion, and its growth pulling ahead of OpenAI.
OpenAI - Compared to Anthropic on growth rate and on Sam Altman's own walkback of AI-job-loss predictions.
Meta - Cut 8,000 jobs after 20,000 prior cuts; Zuckerberg attributed cuts partly to AI and deployed employee-monitoring software to improve AI training data.
Amazon - Andy Jassy's stated plan to eliminate 600,000 future positions via AI/robotics; cited as evidence of real AI-driven job displacement.
Goldman Sachs - CEO David Solomon's op-ed arguing AI job apocalypse fears are overblown.
Cloudflare - Matthew Prince's layoffs and comments on AI consolidating roles like 'measurers' (PMs/designers) into fewer jobs.
Block (Square) - Jack Dorsey's 50% workforce cut attributed to AI, widely called AI-washing by analysts per Sacks.
Wix - CEO's layoff note flagged as a possible case of AI-washing operational problems.
Kirkland & Ellis - Reported to be spending half a billion dollars to build its own frontier model in-house.
Abacus - A Chamath-backed portfolio company selling on-prem AI hardware/platform boxes to insurance and healthcare firms wanting model independence.
Apple - Framed as a dark horse in AI due to its on-device compute (M5, Mac Studio) enabling data/intelligence sovereignty.
xAI - Elon Musk's claim of rewriting the training stack in C for order-of-magnitude speed gains on 220,000 GPUs.

Techniques and frameworks

Mega prompt / skills document workflow - Producer Nick's method of feeding transcripts into Claude and asking it to draft a custom prompt and skills/training-rules document for the show's daily briefing.
Dr. Frankenstein theory - Gurley's framework that Anthropic's leadership may sincerely believe they are creating a superior, god-like successor species rather than a product.
Quis custodiet ipsos custodes / separation of powers - Sacks's political-philosophy frame for why AI regulators need checks, drawing on the American founders' solution of dividing power among competing branches.

Summary

The episode opens with Bill Gurley sitting in for David Friedberg, and quickly turns to Pope Leo XIV's first encyclical on AI, "Magnifica Humanitas," a 235-page document warning that technology takes on the character of whoever builds, finances, and controls it, and calling for regulation, worker retraining, and a ban on autonomous weapons. Anthropic co-founder Chris Olah joined the Pope in discussing the document despite lobbying from Amazon, Google, and Meta to soften its language. Sacks says he broadly agrees with the Pope that centralization of power is AI's core risk, but argues that empowering government to regulate models (an "FDA for AI") is the more dangerous fix, since the definition of "safety" tends to expand the way social media's "trust and safety" mandate did into censorship. He prefers antitrust enforcement against monopolistic players and market competition among frontier labs as the real check on power.

That leads into an extended dissection of Anthropic's motives. Gurley lays out two competing theories: a familiar "regulatory capture" read (Anthropic's doomerism as a lobbying tool to lock in favorable rules) and a new "Dr. Frankenstein theory" - that the company's leadership may sincerely believe they are midwifing a superior, god-like successor species, pointing to Chris Olah's 80-page "Constitution" document and Dario Amodei's "Machines of Loving Grace" essay, which envisions AI systems allocating resources to a post-labor humanity. Chamath calls this narcissistic "delusions of grandeur," while Sacks notes the game-theoretic upside regardless of sincerity: branding yourself the safe AI company casts rivals as reckless and can justify monopolistic control. The group connects this to a prediction that Washington's AI-safety rhetoric is building toward an eventual ban on open-source and open-weight models, which Sacks argues would just hand the technology's future to China and the rest of the world while stranding US infrastructure. Chamath frames the counter-argument as "intelligence sovereignty," running your own models on your own hardware, with Apple's on-device compute strategy positioned as a dark horse.

The back half is dominated by a sharp, contentious debate between Calacanis and Sacks (with Gurley refereeing) over whether the AI jobs narrative flipped this week. Goldman Sachs CEO David Solomon published an NYT op-ed calling the AI job apocalypse overblown, and both Sam Altman and Dario Amodei reportedly walked back earlier mass-job-loss predictions - which Sacks treats as vindication of his contrarian January call that AI would net create jobs. He backs this with unemployment near record lows (4.3%), software job postings up 15% year-over-year to a three-year high despite GitHub code commits rising roughly 14x, and a Yale Budget Lab study finding no discernible AI-driven labor disruption. Calacanis pushes back hard, arguing that Meta's 8,000 cuts, Amazon's stated plan to eliminate 600,000 future roles, and Jack Dorsey's and Matthew Prince's own public statements attributing layoffs to AI should be taken at face value, and that the "overhiring correction" framing lets companies dodge accountability for real, painful displacement of drivers, warehouse workers, and middle managers. Gurley and Sacks separately flag a securities-fraud risk in "AI washing," where a company blames layoffs on AI to mask ordinary overstaffing corrections.

A shorter thread covers the economics of AI tooling itself: Sacks argues Claude proficiency is currently the single most marketable job skill in the economy, illustrated by the show's own producer building a detailed custom prompt and "skills document" that lets Claude generate a highly contextual daily briefing. A cited Rogo benchmark testing frontier models as financial analysts found Opus, GPT, and Sonnet scores separated by less than 0.3 percentage points, prompting Gurley to argue for more open-source connector standards so models become swappable commodities rather than lock-in dependencies, and Chamath to describe enterprise customers building "headless" control planes to hot-swap between providers. The episode closes on lighter notes about token-spend overruns (an anecdote about a company burning $500,000 in a single month on unmanaged AI usage) before a personal shout-out to Tulsi Gabbard's husband.

Notable Quotes

"I don't think they think they're writing software. I think they're midwifing a deity here." - Bill Gurley

"I very much agree with the Pope that the biggest risk of AI is a centralization of power and then its misuse against us in some Orwellian way." - David Sacks

"The best way to protect yourself from AI is to be the most AI enabled version of yourself you can be." - Bill Gurley

"Every cab driver is losing their job. Every truck driver is losing their job in the next 10 years." - Jason Calacanis

"I'm the CEO of Goldman Sachs. Period. The AI job apocalypse is overblown. Period." - David Solomon (quoted by Jason Calacanis)