Glossary

The vocabulary of AI agent management

The category is new enough that its words are still being pinned down - and several are used interchangeably when they should not be. Each entry below opens with a direct, dated definition, then draws the line between it and the terms next to it. One intent per page.

1. Agentic orchestrationRuntime coordination of autonomous agents: who runs, in what order, with what hand-offs.
2. AI agent orchestrationThe control layer that routes tasks across many agents into one coordinated system.
3. AI agent memoryDurable cross-session knowledge, and the judgment of what to surface, when, and at what depth.
4. AI agent observabilityInstrumenting agents to see what they did and why: traces, tool calls, tokens, errors.
5. AI agent governanceThe policies and controls that keep a fleet of agents inside your risk and compliance boundaries.
6. AI agent workspaceThe environment where agents run and are kept accountable, a place rather than a framework you code against.
7. AgentOpsThe operational discipline of running agents in production, and also a company name.
8. Multi-agent systemSeveral role-scoped agents dividing labor and coordinating toward one outcome.
9. AI agent audit trailThe complete, timestamped, replayable record of everything an agent actually did.
10. AI agent permissionsScoped grants bounding what each agent may read, write, run, or spend: access by role.
11. Human-in-the-loopA deliberate human gate between an agent's intention and its consequence: approval before action.
12. Agent sandboxRuntime isolation that bounds an agent's reach, making a bad run recoverable instead of catastrophic.
13. Inner loopThe cycle an agent runs on its own: investigate, implement, test, report. Accountability stays in the outer loop.
14. Outer loopThe human-owned cycle around the agent: verify, decide, own. Where accountability lives.