Agent infrastructure is everything an AI agent needs around the model to do real work: grounding in memory and knowledge, tools it can call, a workspace to act in, an identity with permissions, and governance that keeps its actions safe and reversible. The model is one ingredient. The infrastructure is the product.

The model is not enough

A large language model is a reasoning engine. On its own it has no memory of you, no tools, no place to act, and no accountability. The last few years made models capable; what made them useful was surrounding them with the rest of the system — grounding, tools, memory, safeguards, and the ability to iterate until a job is done. That surrounding system is agent infrastructure.

The three layers you can tune

Every agentic system optimizes at three layers: the model (raw capability), the harness (the tools, memory, and workspace around it), and the context (the specific knowledge you ground it on). The model is rented and improving on its own. The harness and the context are the durable layers — the parts that are yours, that compound, and that you can't outsource.

The Cloud as agent infrastructure

The Cloud is agent infrastructure delivered as a shared workspace. It gives an agent a persistent place to keep memory, real tools to call over open and authenticated interfaces, an identity and permission model, and a governance floor. A super agent that runs on it is called an Ora — open, governed, grounded in your workspace, and yours.

Ora · Opus 4.8 · Cowork