Enterprise Agent Security: Why Coding Agents Belong in Remote Sandboxes

sergeykarayev · x · 2026-08-05

The author argues that coding agents should run in remote sandboxes rather than locally on a developer's machine.

Local Risks: Running locally exposes high-level access such as SSH keys, cloud credentials, browser sessions, and internal systems.
Remote Boundaries: While not magically safe, remote sandboxes allow companies to define strict boundaries: which secrets the agent receives, which network paths it can reach, and what logs are retained.

For companies standardizing agent usage, 'where does the agent run?' is a fundamental security architecture question.

More from coding & agent