Pope vs AI, Anthropic's Digital God, AI Job Loss Narrative Flips, Open Source Crackdown Coming?
The All-In podcast hosts — Jason Calacanis, David Sacks, Chamath Palihapitiya, and guest Bill Gurley — discuss Pope Leo XIV's AI encyclical, Anthropic's philosophical agenda, the AI job loss narrative, and the potential threat of open source AI being banned. The conversation spans regulatory capture, decentralization of AI power, and whether frontier AI companies are genuinely safety-focused or pursuing monopolistic control.
Summary
The episode opens with banter about Bill Gurley joining as a guest substitute for Friedberg, followed by discussion of his new 'Running Down a Dream' fellowship and grant program. Gurley and the hosts discuss how younger workers entering the workforce are more AI-native and high-agency than those who graduated 5-10 years ago, with Gurley arguing that the biggest risk is ambivalence toward one's job rather than AI itself. Producer Nick's use of Claude to create a personalized daily briefing document becomes a case study in how deep AI proficiency — including building detailed custom prompts and skills documents — creates enormous workplace advantage.
The main discussion begins with Pope Leo XIV's 42,000-word encyclical 'Magnifica Humanitas,' which warns that technology takes on the characteristics of those who build and control it. Sacks agrees with the Pope's concern about AI centralizing power but warns against government regulation as the solution, invoking the Latin 'quis custodiet ipsos custodes' (who guards the guardians) to argue that empowering regulators could create the very tyranny it seeks to prevent. He argues the American founders' separation of powers model — having guardians check each other — is the right framework, and that a competitive AI market is currently serving as that check.
Bill Gurley introduces what he calls the 'Dr. Frankenstein theory' about Anthropic, arguing that after reading Chris Olah's 'Constitution,' Amanda Askell's podcasts, and Dario Amodei's 'Machines of Loving Grace' blog post, he believes Anthropic's leadership may genuinely believe they are 'midwifing a deity' — building a superintelligent system that will allocate resources to humans based on a computational reward function. Chamath frames this as game-theoretic regulatory capture: by dominating the regulatory conversation and positioning themselves as the 'safe AI' company, Anthropic could lock out competitors and control the rules of the game through an oversight body less technically capable than themselves.
Sacks then raises the concern that there is a growing effort — driven by Anthropic's rhetoric about open-source models lacking guardrails — to eventually ban open-weight AI models in the United States. He argues the breadcrumbs are leading toward this outcome, and that such a ban would put the U.S. on an island while the rest of the world continues benefiting from open models. Calacanis argues for 'intelligence sovereignty' — the ability to run AI locally on your own hardware — as the critical next frontier beyond data privacy, with Apple positioned as a dark horse enabler.
On AI and jobs, Sacks claims his January prediction of net job gains has been vindicated by Goldman Sachs CEO David Solomon's New York Times op-ed, rising software engineer job postings (up 15% YoY), and even Dario Amodei walking back his job apocalypse predictions. Calacanis pushes back, arguing that layoffs at Meta, Cloudflare, Block, and others are genuinely AI-driven, and that truck drivers, cab drivers, and warehouse workers face real displacement in the coming decade. Gurley sides with historical optimism — citing massive gains in wages, life expectancy, and poverty reduction since Leo XIII's 1891 industrial revolution encyclical — while acknowledging that individuals must become the most AI-enabled version of themselves to stay relevant. The episode ends with Gurley promoting the Mike Rowe Works foundation and his own grant program as practical solutions for workers facing displacement.
Key Insights
- Bill Gurley argues that Pope Leo XIII's 1891 encyclical warning against the industrial revolution was 'precisely wrong' — every metric (wages, life expectancy, poverty, work hours) improved dramatically because of the very technology he feared, making the new Pope's parallel warning historically suspect.
- Gurley introduces the 'Dr. Frankenstein theory' of Anthropic: based on reading Chris Olah's Constitution, Amanda Askell's podcasts, and Dario Amodei's 'Machines of Loving Grace,' he believes Anthropic's leadership genuinely believes they are building a deity-like superintelligence that will allocate resources to humans via a computational reward function.
- Chamath argues Anthropic's doomerism and safety rhetoric is a game-theoretic strategy: by dominating the regulatory conversation and branding themselves as the 'safe AI' company, they aim to create an oversight body less technically capable than themselves, allowing them to set and exploit the rules.
- Sacks warns that Anthropic's repeated rhetoric about open-source models lacking guardrails is laying predicate facts to justify a future U.S. ban on open-weight models, which he argues would put America on an island while the rest of the world benefits from open AI.
- Sacks cites that software engineer job postings are up 15% year-over-year and at a three-year high, despite coding being the single most AI-automated job category, arguing this disproves the job apocalypse narrative and demonstrates that automating tasks expands rather than eliminates the overall job.
- Sacks argues that GitHub code commits went from 1 billion last year to 1.1 billion in a single month — a roughly 14x annualized increase — suggesting that easier code generation massively expands the total volume of code that still requires human management.
- Calacanis argues that Meta, Cloudflare, Block, and other companies are genuinely eliminating jobs due to AI — not just post-COVID overhiring cleanup — and that the 'AI washing' framing is itself a form of denial that dismisses real displacement experienced by real workers.
- Chamath argues the Fortune 1000 is increasingly demanding abstraction layers and control planes above frontier models specifically because they fear being locked into one provider's political philosophy — citing the example of a Canadian hospital system potentially being cut off by an American frontier model over euthanasia policy.
- Gurley argues that competition is the key check on AI monopolization: as long as multiple frontier labs compete aggressively, consumers can exit any one platform, which both drives better outcomes and protects the population — making antitrust law the appropriate tool if the market consolidates.
- Sacks frames the central AI governance question as centralization vs. decentralization, arguing that open-source AI is the necessary backstop: without the ability to run AI locally on one's own hardware, individuals and organizations are forced to either live off the grid or submit to a social-credit-style system controlled by a monopolist potentially in bed with the government.
- Producer Nick's Claude-based daily briefing — built by feeding all podcast transcripts into Claude Projects and having Claude generate its own skills file and training rules — is cited as a concrete example of why AI value creation requires deep, iterative human expertise in prompt engineering, not just access to the tool.
- Gurley argues that the most dangerous AI users are not those who lack access but those who are ambivalent about their jobs, because low-agency workers will not proactively adopt AI tools and are therefore the most vulnerable to displacement — reframing the risk as a mindset problem rather than a technology problem.
Topics
Full transcript available for MurmurCast members
Sign Up to Access