Gutenberg Lab — Morning Paper

AI Industry's Self-Regulation Proposal Emerges as Anthropic's State-by-State Capture Strategy and Foreign Influence Operations Threaten US AI Dominance

Participants

Jason Calacanis × Chamath Palihapitiya, David Sacks, David Friedberg

Bottom line

The AI regulation debate has shifted from "whether to regulate" to "who controls regulation," with the industry-led SRO (Self-Regulatory Organization) proposal representing a defensive retreat rather than victory, while Anthropic's state-by-state regulatory ratcheting strategy combined with foreign-funded anti-data center campaigns threatens to impose a 5+ year infrastructure delay that could hand China the AI race.

Three-Line Summary

DeepMind's Demis Hassabis proposed a FINRA-style industry SRO with major AI companies' endorsement. David Sacks conditionally supports it with five requirements but exposes Anthropic's "one-upmanship strategy" of progressively tightening state regulations (per Politico reporting). NY Governor announces nation's first statewide data center moratorium; Energy Secretary Chris Wright reveals common funding patterns between anti-data center and anti-fracking protesters, suggesting Chinese influence operations targeting US AI infrastructure.

Three Key Points

1. Anthropic's State-by-State Regulatory Ratchet Exposed

Politico article reveals "Anthropic's state-by-state plan to ratchet up AI rules." Sacks: "With each new state, they actually make the regulations more and more strict, more and more all encompassing. There's not actually a stable equilibrium." This contradicts public claims of seeking regulatory clarity and represents textbook regulatory capture — creating compliance barriers only well-funded incumbents can afford.

2. 100x Token Cost Gap Defines Competitive Landscape

Chamath's data shows Claude (Anthropic) at ~$56 per million input tokens vs. Chinese models at $0.50 — over 100x differential. Ramp customers' token spend grew 21x in one year. Ramp CEO Eric Glyman warns CFOs could miss earnings "by several dollars, not just a few pennies" due to uncontrolled token spend at $56/million when alternatives cost $0.50. Anthropic's pricing strategy only works if regulation kills cheaper alternatives.

3. Data Center Moratorium Creates 5+ Year Infrastructure Delay

NY Governor Kathy Hochul announces nation's first statewide data center moratorium. Sacks analyzes: "Democrats aren't gonna pause data centers forever — they're gonna pause them until they're in enough control that they can dictate all of the rules." Even after lifting, projects need 2 years to ramp up, meaning "probably a good five years at least" delay. PJM utility auction needed 7-8 gigawatts, received only 156 megawatts (98% shortfall) — energy crisis already manifested.

Editorial Perspective

This episode reveals the AI regulation debate has evolved from surface-level "innovation vs. safety" framing into a power struggle over who writes the rules. The industry-led SRO proposal parallels historical platform governance attempts — i-mode's official site審査 or App Store review guidelines — but with added complexity from national security entanglement. Anthropic's state-by-state regulatory ratcheting mirrors Microsoft's OEM licensing tactics that locked out competitors, now applied to AI. Notably, OpenAI itself reports "PRC-linked influence operations targeting AI debates in the US," indicating the regulatory discourse has become an information warfare battlefield. With energy infrastructure constraints materializing, regulatory delays could neutralize technological advantages. The convergence of Anthropic's capture strategy, state-level moratoriums, and foreign influence campaigns creates a perfect storm that could hand China the AI race through infrastructure strangulation rather than algorithmic superiority.

Source: All-In Podcast "Can the AI Industry Regulate Itself? Stripe Wants PayPal, China Catches Up, NY Bans Datacenters" (2026-07-18)

https://allinchamathjason.libsyn.com/can-the-ai-industry-regulate-itself-stripe-wants-paypal-china-catches-up-ny-bans-datacenters

• --

Coding Is No Longer the Bottleneck — Anthropic's 8x Productivity Era Reveals New Constraints

Guests

Lenny Rachitsky (Host, Lenny's Podcast) × Fiona Fung (Manager, Claude Code & Cowork Teams at Anthropic; former Meta engineering leader overseeing 500+ people)

Bottom line

In the AI era, the scarce resource is no longer execution capability but the ambition and judgment to decide what to build.

3-Line Summary

Anthropic engineers ship 8x more code per quarter than in 2025, but Fiona Fung reveals the constraint has shifted from "coding" to "verification, ambition, and product sense." The company requires managers to work as individual contributors first, deploys asynchronous agent fleets that generate PRs by morning, and uses a "Bad vs Sad" quality framework to sustain 8x velocity. This transformation previews the future for all software teams, but also surfaces unsolved challenges: engineer loneliness, context-switching overload, and how to train the next generation when foundational coding experience disappears.

3 Key Points

1. 8x Productivity and the Shift in Constraints: Anthropic engineers ship 8x more code per quarter vs 2025. Fiona states, "Coding is no longer the bottleneck. Everything is now possible in theory. Now it's about how ambitious can you be." Ideas once rejected as "too hard" are now implemented by asking Claude. The constraint has moved from execution to imagination.

2. Rise of Asynchronous Agent Fleets: Anthropic introduced "routines"—scheduled autonomous agents. Fiona's morning ritual used to be manually reviewing feedback channels; now an agent auto-summarizes and even generates PRs for her to review. "It's almost like I'm having an agent help me generate the prompts." The shift from synchronous prompting to async agent fleets is underway.

3. Managers Work as ICs First: At Anthropic, every new manager starts as an individual contributor (IC) for a period. Rationale: (1) learn the codebase deeply, (2) build trust by doing the work, (3) maintain product intuition. Fiona herself, who led a 500-person org at Meta, shipped production code for the first time since 2017 after joining Anthropic. "Claude Code gave me confidence."

Editorial Take

This shift echoes Marshall McLuhan's "the medium is the message"—a structural transformation where the tool redefines the work. Just as the printing press abstracted away scribal skill, AI abstracts coding and foregrounds the judgment layer: "what to build." Yet Fiona's worry—"how do we grow the next generation?"—remains unsolved. Each time abstraction layers rise, industries face the same question: what foundational knowledge must remain? History suggests the answer emerges in hindsight, not in advance.

Source: Lenny's Podcast "What happens after coding is solved? | Fiona Fung (Manager of the Claude Code and Cowork Teams)" (2026-06-21)

https://www.lennysnewsletter.com/p/building-the-most-ai-pilled-engineering

• --

Machiavelli Was Not a Schemer for Self-Interest—He Was a Patriot Who Served Only Florence

Guests

Dwarkesh Patel × Ada Palmer (historian at University of Chicago, science fiction author)

Bottom line

*The Prince* was not a manual for personal advancement but a handbook for state stability in an Italy facing papal regime overthrows and cascading city-state collapses—and its author was so patriotic he refused lucrative foreign jobs to serve only the regime that had tortured him.

Three-Line Summary

University of Chicago historian Ada Palmer explains that when Machiavelli wrote *The Prince* (1513), Italy faced a dual structural crisis: papal turnover every ~10 years destabilizing governments, and city-states losing centuries-old continuity. Machiavelli's admiration for Cesare Borgia's "betrayals" was scientific analysis of power maintenance—betrayal works *if* you're scary enough that others become *more* loyal—not blanket amorality. The biggest misunderstanding: Machiavelli himself refused all foreign court offers, stayed in punitive exile, and wrote *The Prince* only for Florence, making him "possibly one of the most selfless patriots in Earth's history."

Three Key Points

1. Papal Non-Hereditary Succession Created Unique Italian Instability

Popes reigned ~10 years on average, elected by coalitions *against* the previous pope, guaranteeing policy reversals. Each pope installed relatives as city-state rulers, creating "unpredictable monarchs with military power overthrowing governments every decade." Machiavelli's solution: Medici conquest of enough territory that even the papacy would fear them and negotiate rather than constantly interfere [00:02:23–00:05:28].

2. Patronage Was Trust Infrastructure, Not Corruption

Romans rioted demanding the pope appoint his *son* (not a competent general) because the son would never betray Rome—nepotism guaranteed loyalty. 99 of 100 trials ended in fines (not the capital sentence on the books) because 99 had patrons; the 1 execution meant no patron intervened. Multigenerational family entanglements (rise together, fall together) enabled trust in a world where soldiers swore oaths to commanders, not states [00:36:13–00:42:15].

3. Renaissance Art Was Defense Spending Cheaper Than War

Florence couldn't afford armies to stop France but could afford painting fleur-de-lis everywhere and gifts for the French king. Palmer: "If we fought, we'd lose. But we can try to win the culture victory game." Diplomacy is cheaper than war (like Fulbright Program ROI), and the Renaissance cultural explosion was a strategic response to instability, not surplus luxury [00:59:53–01:00:32].

Editorial Take

The adjective "Machiavellian" means self-serving scheming, yet the real Machiavelli refused foreign courts' high salaries to serve only Florence from exile—this paradox reveals how thinkers split into "useful characters" and "actual innovations." Hobbes and Spinoza also became "villains," but their misread personas became scaffolding for later thought experiments ("What would a Machiavellian politician do?"), supporting society separately from their original works. Palmer's insight resonates in the AI era: technologies and ideas are evaluated on two axes—utility and accuracy—and sometimes the former eclipses the latter. The structure persists.

Source: Dwarkesh Podcast "Ada Palmer – Machiavelli is the most misunderstood thinker of all time" (2026-06-16)

https://www.dwarkesh.com/p/ada-palmer-2

• --

AI Disclosure: This content was produced with AI assistance; facts and analysis reviewed by our editorial team.