AI Agents ☆

An agent loops a model through plan → act → observe: calling tools, browsing, writing files, correcting itself until the goal is met. Coding agents (Claude Code, Codex, Devin) proved it commercially; standards like MCP wire agents to the world. Agents multiply token consumption per task by orders of magnitude — the demand driver behind the inference buildout — and concentrate the safety questions, since acting models can do more than talking ones.