We Live in an Emergent Compute World

Source: Nate’s Newsletter Date: 2025-05-08 URL: https://natesnewsletter.substack.com/p/we-live-in-an-emergent-compute-world

Summary

Nate argues computing has shifted from deterministic (scripted, predictable outcomes) to emergent (capabilities discovered by running the system). The evidence: transformer models learning unexpected patterns like semantic meaning in payment sequences — behavior that wasn’t programmed. Society remains unprepared for systems whose capabilities are discovered rather than designed.

Implications

Agent product strategy thread. Emergent capabilities mean agent systems can do things their builders didn’t anticipate — in both beneficial and failure directions. Products built on deterministic assumptions will encounter emergent behavior they didn’t design for. This is an architectural argument for testing over specification.

AI economics thread. If more real-world phenomena are learnable than anticipated (Nate’s implication), the economic potential for AI is larger than current projections — but so is the risk of unexpected failures at scale.

Vendor positioning thread. Labs that ship emergent systems and discover capabilities post-deployment are operating in a different risk regime than those that exhaustively characterize behavior before release. Anthropic’s interpretability work is partly an attempt to recover deterministic understanding from emergent systems.

Watch: Whether the emergent-compute framing becomes the mainstream lens for understanding AI system behavior by 2026, displacing the “AI as tool” mental model in enterprise contexts.