Long-read: AI's real value comes after the efficiency era—infra glut, agent tooling boom

yangyi · x · 2026-11-02

Blogger yangyi published a long essay arguing AI's biggest opportunities lie beyond the current productivity-efficiency narrative:

Tech revolutions take 10-20 years from research to transformation; this cycle started when ChatGPT went mainstream in late 2022
A five-stage cycle: research → infrastructure buildout → old-scenario migration → new-scenario innovation → content production; we're currently at the tail of the infrastructure stage (see cycle diagram)
Infra follows a shortage-capital-glut loop: bottlenecks get expensive, then get solved—infra always ends up cheap and abundant
Physical infra (Nvidia, data centers) came first; application-layer infra started in 2025: post-MCP, agent tooling exploded—MCP servers, CLIs, connectors, skills everywhere. Companies selling shovels and aggregators monetize fastest (e.g. OpenRouter's acquisition ending)
His team bet on the Personal Agent narrative with Pine in March 2025: Manus proved AI connecting to computers is inevitable; eventually everything (smart homes, even small merchants' businesses) gets digitized and operated by AI. The bottleneck then was tool scarcity, so their strategy was having AI build its own tools

Core thesis: efficiency is only AI's opening act; the real watershed comes when efficiency gains plateau and new scenarios plus content production take over.

Original post →

Long-read: AI's real value comes after the efficiency era—infra glut, agent tooling boom

More from AGI Musings