Outsource the Thinking, Keep the Doing

The model reasons in the cloud; the code runs on your infrastructure. You outsource the intelligence without outsourcing the risk.

Contents
    Hand-drawn diagram titled Outsource the thinking, keep the doing. On the left, the thinking happens at a cloud model provider, where the model reasons. A dashed red arrow labeled tool calls goes to the right, and a gray arrow returns results. On the right, the doing stays in your infrastructure: a highlighted sandbox runs the code, the agent holds no credentials, and unauthorized connections are denied and logged.

    Use it when Using a hosted frontier model for an agent that must work on data, files or internal services that cannot leave your network.

    Until now you had two options. Run everything in the cloud and accept that your data, code and execution live outside your perimeter. Or build the whole orchestration stack yourself and lose the intelligence that makes managed agents valuable.

    Split the thinking from the doing.

    • The thinking: the model reasons in the cloud, handles conversation and tool routing, and sends tool calls to your worker.
    • The doing: the code runs on your infrastructure, inside a sandbox per session, and only sanitized results go back.

    Where code runs is not the same as how safely it runs. The execution layer needs its own defense:

    • The agent holds no credentials. They are injected at the network boundary, so a compromised agent has nothing to exfiltrate.
    • Deny-all by default. You grant only what the agent needs.
    • Unauthorized connections are denied and logged, not silently dropped, so security teams see what agents attempt.

    One architecture, everywhere: the driver changes, the security posture does not. The best AI reasoning in the world should not require you to give up control of your execution environment.

    Topics

    Agents in Production 10Building with AI 9AI Product Strategy 24

    Start here

    How Agents Run in ProductionBeyond LLMs: The Whole Product Framework for AIThe Verification Bottleneck: Why AI's Real Cost Is Human Attention

    ↑ ↓ to move, Enter for all results