Environment
The world the agent acts on: anything outside the harness that it perceives via tool results and changes via tool calls.
The environment is everything outside the harness that the agent perceives through tool results and changes through tool calls: the filesystem, the repository, the shell, the network, the database, the issue tracker, the calendar. The harness is the agent's senses and hands; the environment is the world those hands touch.
The distinction is not academic, because the two have completely different failure characteristics. A bug in the harness is a bug you can reproduce, trace, and fix in code. A surprise in the environment is discovered by the agent mid-task, and the only defence is writing the agent to expect it: permission boundaries, sandboxing, and the assumption that any file content is hostile input.
The environment is also where state survives. Sessions end, the harness goes away, and the environment is still there with the files the agent wrote. That is what makes it the natural home for memory systems and handoff artifacts: they are state the agent leaves behind on purpose, in a form the next session can read.
Related terms
Filesystem
A tree of files and directories the agent reads from, writes to, and executes within.
Sandbox
An isolated environment the agent runs inside: container, VM, or restricted shell. It limits the blast radius.
Tool
A function the model is allowed to call, like reading a file or running a command.
Harness
Everything around the model that turns it into an agent: tools, system prompt, permissions, hooks.
Memory System
A system that makes an agent stateful across sessions by persisting to the environment and reloading at session start.
Want this working inside your team?
It starts with one conversation about your workflow.