Software ate the world; what's AI going to do to software? (jdauriemma.com)

The piece argues that the defining shift of the AI era is from producer-published software to platform-published software: AI Agent platforms (ChatGPT-style assistants) can generate and execute code and thus have no theoretical limit on the amount of software they can publish. That capability, plus an API layer that pulls producers’ inventory/data into agents’ experiences (illustrated by a rug-buying example), means platforms can capture not just attention but the transaction and data flows that currently belong to producers. The author frames this with an “enshittification quotient” — platforms’ capacity to extract rents — and warns that hosted LLMs will likely be loss leaders used to win control of the data/transaction layer unless antitrust or market dynamics intervene. For AI/ML practitioners and product teams, the technical and economic implications are stark: models are infrastructure — “shovels” — while durable value is in data and transaction registries. Thin “agentic” wrappers around LLMs won’t survive if consumers consolidate around a few dominant agents, so producers will pivot from building full apps to exposing compliant, high-performance APIs. That risks commoditizing brands, compressing margins, and shifting surplus toward platform arbitration (ranking, bundling, pricing) in opaque algorithmic ways. The essay urges early policy and design attention because the architecture of today’s platforms will determine whether AI democratizes or vertically consolidates control over software, data, and commerce.