The AI Agent Glossary

Plain-language definitions for the AI agent fleet era — the terms you need to run, coordinate, and govern autonomous coding agents.

Agent Compute Unit

An agent compute unit (ACU) is a normalized measure of the computational resources consumed by an AI agent during a task, combining token usage, tool call volume, and wall-clock execution time into a single comparable metric.

Agent Event Bus

An agent event bus is a shared messaging system that lets AI agents publish structured events and subscribe to events published by other agents, enabling asynchronous coordination without direct agent-to-agent calls.

Agent Fabric

Agent fabric is a shared communication and coordination layer for a fleet of AI agents, providing event publishing, subscriptions, task claims, inbox messaging, and a persistent audit trail across all agents in the system.

Agent Handoff

An agent handoff is the structured transfer of work from one AI agent to another, including the context, artifacts, and state information the receiving agent needs to continue without reprocessing everything from scratch.

Agent Jail

An agent jail is an OS-level isolation boundary around a locally launched AI agent. On Linux that is typically user, mount, PID, and network namespaces, so the process cannot reach the host network, host filesystem, or ambient credentials except through an explicit broker.

Agent Observability

Agent observability is the ability to understand what an AI agent is doing, why it made specific decisions, and how its performance compares to expectations — derived from structured logs, metrics, and traces captured during agent execution.

Agent Orchestration

Agent orchestration is the coordination layer that sequences, routes, and monitors multiple AI agents so they collaborate on a shared workflow without conflicting or duplicating effort.

Agent Quarantine

Agent quarantine is the practice of automatically suspending an AI agent from active work when its behavior exceeds a risk threshold, isolating it from further actions until a human investigates and explicitly re-enables it.

Agent Risk Scoring

Agent risk scoring is the continuous evaluation of an AI agent's current behavior against a set of risk features to produce a numerical score that reflects the probability of the agent causing harm if allowed to continue operating.

Agent Role

An agent role is a defined responsibility boundary assigned to an AI agent — such as developer, reviewer, QA engineer, or release manager — that scopes its permitted actions, determines which events it responds to, and shapes the prompt it receives.

Agent Sprawl

Agent sprawl is the condition where an organization has deployed more AI agents than it can effectively monitor, govern, and coordinate — leading to duplicated work, conflicting outputs, uncontrolled costs, and degraded observability.

Agent Token Budget

An agent token budget is a preconfigured limit on the total number of tokens an AI agent may consume in a single session or over a given period, used to control cost and prevent runaway execution.

Agentic Engineering

Agentic engineering is the practice of designing, deploying, and operating software systems where AI agents perform substantive engineering work — writing code, running tests, managing deployments — as autonomous participants in the development process rather than as passive tools.

AI Agent Control Plane

An AI agent control plane is the centralized management layer for an organization's autonomous agents — the single surface where agents are registered, configured, started and stopped, governed, monitored, and audited, separate from the data plane where the agents actually do their work.

AI Agent Cost

AI agent cost is the total expense of running autonomous AI agents, including model API fees (charged per token), compute for tooling infrastructure, developer time for configuration and oversight, and the cost of errors the agent makes that require human remediation.

AI Agent Fleet

An AI agent fleet is a coordinated collection of autonomous AI coding agents — assigned distinct roles such as developer, reviewer, QA, and release manager — that collaborate on software work under centralized management, governance, and auditing.

AI Agent Governance

AI agent governance is the set of policies, controls, and monitoring practices that define what autonomous AI agents are permitted to do, how their actions are audited, and how humans retain meaningful oversight of agent behavior in production systems.

AI DevOps

AI DevOps is the integration of AI agents and automation into the DevOps toolchain — CI/CD pipelines, infrastructure management, incident response, and release coordination — so that routine operational tasks are handled autonomously with humans focusing on exceptions.

AI Software Engineer

An AI software engineer is an AI agent capable of performing the full range of software engineering tasks — reading requirements, writing and testing code, reviewing pull requests, debugging failures, and communicating status — at a level of autonomy comparable to a human junior-to-mid engineer.

Approval Gate

An approval gate is a checkpoint in an AI agent workflow where execution pauses and a human or authorized system must explicitly approve before the next stage begins.

Autonomous Coding Agent

An autonomous coding agent is an AI system that reads a task description and independently writes code, runs tests, fixes errors, and commits work — completing a software development task end to end with minimal human intervention.

Coding Agent

A coding agent is an AI system that uses a language model combined with code execution, file access, and version control tools to write, test, and modify software in response to natural language task descriptions.

Destination-Level Egress

Destination-level egress grants an agent connectivity to a specific host or SNI for a bounded window. Merge, delete, and install stay on other controls.

Git Worktree

A Git worktree is an additional working directory linked to an existing Git repository, allowing multiple branches to be checked out simultaneously in separate directories without cloning the repository again.

Human in the Loop

Human in the loop (HITL) is an AI system design pattern where human judgment is incorporated at defined points in an otherwise automated workflow, ensuring human oversight at consequential decision points rather than replacing human judgment entirely.

Model Context Protocol

Model Context Protocol (MCP) is an open standard, introduced by Anthropic in 2024, that defines how AI models communicate with external tools and data sources through a structured JSON-RPC interface.

Multi-Agent Coding

Multi-agent coding is a software development approach where several specialized AI agents — each assigned a distinct role such as developer, reviewer, or tester — collaborate on the same codebase, handing off work through structured events or messages.

Prompt as Job Description

Prompt as job description is a design philosophy for AI agent prompts where the system prompt is structured like a human job description — defining role, responsibilities, scope, reporting relationships, and behavioral expectations — rather than as a list of instructions for a single task.

Shadow AI

Shadow AI refers to the use of AI tools, agents, and services by employees or teams within an organization without the knowledge, approval, or oversight of IT, security, or management.

SWE-bench

SWE-bench is a benchmark for evaluating AI coding agents, consisting of real GitHub issues from open-source Python projects paired with the actual code changes needed to resolve them.

Keep your AI agents from escaping

Jail Fleet-launched Linux agents, then run saved workflows with review, approvals, and an audit trail. Prove it at /security/#containment.

See how it works Install Fleet