The plain-English version
The surrounding software that runs an agent loop and manages its tools, execution environment, and working context.
Why it matters
A prompt cannot supply a missing browser, filesystem permission, or dependable mechanism for carrying project state between runs.
Let’s make it click
Two tools using a similar model may differ because one can execute tests and retain project artifacts while the other only returns text.
Vibe Bro’s heads-up
Attributing every success or failure to the model ignores the environment in which it works.
You might also hear
- coding agent runtime
- orchestration
- scaffolding
The docs behind it
Research sources reviewed .
Editorial review · October 28, 2026
Supported: Anthropic describes prompts, tools, environment/progress artifacts, and verification around long-running agents. The term's general system boundary is supported; component value remains model/task-dependent.