Glossary

Agent Constitution

An agent constitution is the written policy that defines what an AI agent is authorized to do, what it must refuse, how it escalates, and how it speaks — enforced at runtime by a policy layer.

An agent constitution is the deployment-time counterpart to a model's training-time safety work. It covers scope (which workflows the agent owns), prohibitions (behaviours it must refuse), escalation (when and how it hands off to humans), tone (how it speaks), and change-control (who can amend it and how).

A usable constitution is short (2–5 pages for a first deployment), specific, testable, and versioned in source control. It is enforced by a policy layer at runtime — not just via system prompts, which drift and can be overridden.

Constitutions are reviewed quarterly and amended whenever scope changes, incidents occur, or red-team evaluations reveal gaps.

See also
  • Constitutional AI— Constitutional AI is an approach to training and deploying AI systems in which model behavior is guided by an explicit written set of principles — a "constitution" — rather than only by reinforcement from human feedback.
  • AI Agent Governance— AI agent governance is the set of controls, policies, and audit mechanisms that keep deployed AI agents operating inside defined boundaries.
  • AI Agent— An AI agent is a software system that uses a large language model to perceive its environment, reason about tasks, and take actions in external systems on behalf of a user.
Guide: Agent governance & constitutionGrowTK governance posture

More terms in the glossary

AI Agent

An AI agent is a software system that uses a large language model to perceive its environment, reason about tasks, and take actions in external systems on behalf of a user.

AI Voice Agent

An AI voice agent is an AI agent that interacts with users through voice — answering inbound calls, placing outbound calls, and conversing in real time.

AI Chatbot (AI Chat Agent)

An AI chatbot — more accurately, an AI chat agent — is an AI agent that interacts with users through text: web chat, WhatsApp, SMS, Slack, Teams, or social messaging.

Retrieval-Augmented Generation (RAG)

Retrieval-augmented generation (RAG) is an architecture that grounds a language model's responses in a specific knowledge base by retrieving relevant passages at inference time and conditioning the response on them.

Multi-Agent Orchestration

Multi-agent orchestration is the practice of coordinating multiple specialized AI agents to accomplish a task that is too complex or too broad for a single agent.

Constitutional AI

Constitutional AI is an approach to training and deploying AI systems in which model behavior is guided by an explicit written set of principles — a "constitution" — rather than only by reinforcement from human feedback.

See the full glossary

Related pages

Enterprise AI Agents

How GrowTK builds and ships these concepts for clients.

Guides

Long-form walkthroughs that go deeper than definitions.

Scope a deployment

Ready to talk to someone about your specific use case?