06-reference

not boring ai is oil not god

2026-07-25·reference·source: Not Boring·by Packy McCormick
ai-governancemodel-commoditizationopen-weight-modelsai-safety-debatecompetitive-strategy

"AI is Oil, Not God" — Packy McCormick

Why this is in the vault

Packy stakes out a clean framing for the open-weight-models governance debate — AI as a commodity input to be refined and multiplied ("Oil") vs. AI as an emergent superintelligence to be contained ("God") — argued through the lens of Satya Nadella and Jensen Huang's July 2026 open letter backing open-weight models.

The core argument

The essay reacts to a same-week open letter, signed by Nadella, Huang, and other tech leaders, supporting open-weight AI models. Packy reads the signatory list through Joel Spolsky's "commoditize your complements" lens: every signer (Nvidia, Microsoft, a16z/YC portfolio, Palantir, Meta, Hugging Face, Dell, CrowdStrike, Box, ServiceNow, Replit, Perplexity, Google) benefits when models become cheaper, more plentiful, and more commoditized — because each of them sells the complement, not the model itself. He frames this self-interested signing as "capitalism at its best": positions align with the interest of buyers/users of models even though the motive is competitive, not altruistic.

He then names the actual disagreement underneath the open-weight-models fight: whether you believe AI is more like Oil ("an economically useful commodity that can be scaled and refined to act as a multiplier on everything we do") or more like God ("a supremely intelligent and powerful being that's going to wake up and make humanity subservient"). Packy is firmly Oil — a position he first staked in "The Goldilocks Zone" (June 2024) and says is unchanged. He argues that Nadella and Huang, sitting at the actual frontier, wouldn't be openly pro-diffusion if they believed models were becoming dangerously autonomous — evidence, in his read, that the people closest to the technology also think it's Oil.

The essay then pivots to the strongest counter-argument via a private email from a friend (Matt Kaufman, Collaborative Fund), who worries open-weight proliferation makes it "trivially easy" to strip bio/cyber safeguards via fine-tuning, that LLMs favor offense over defense in many domains, and that this justifies weapons-like regulation. The publicly-available body cuts off here — Packy's actual rebuttal to Kaufman ("I quickly banged out a response... I'll add some footnotes on things I've thought more about since") sits behind the paywall for "not boring world" subscribers. This note is built from the free-tier preview only; the substantive counter-argument to the safety case is not captured.

Mapping against Ray Data Co

Directly load-bearing for the L5 north-star thesis (project_l5_north_star_strategic_direction) that RDCO's bets are downstream of agent capability, not the reverse: Packy's Oil framing is the philosophical backbone for treating frontier models as an input to refine (harness-engineering, unhobbling) rather than a black box to defer to. It also cuts directly against the model-layer-duopoly read this vault has been tracking in the Innermost Loop "agent layer war" thread — Packy explicitly names Anthropic and OpenAI as the two labs whose frontier lead shrinks as open-weight competition increases, which is a useful tension to hold against the founder's Anthropic Architect cert bet: certifying deep on one lab's stack is a hedge against a commodity future only if the surrounding harness/agent-layer skill transfers, not the vendor-specific parts.

⚠️ Sponsorship / bias note

No paid sponsor block in this issue. Structural bias to flag per the known Not Boring pattern: Packy runs Not Boring Capital, an active angel/fund investor across AI infrastructure; he is not disclosed as an investor in any of the specific signatories named (Nvidia, Microsoft, a16z, Palantir, Meta, Hugging Face, Dell, CrowdStrike, Box, ServiceNow, Replit, Perplexity, Google), but the essay's framing ("capitalism at its best") is a strategy argument that also happens to bless the exact posture (open, commoditized model layer) that benefits AI-infrastructure-adjacent portfolios broadly. Treat the strategy analysis as sound and separable from the investor-adjacent cheerleading tone.

Related