An AI support agent inside Intercom that resolves customer conversations from your help content and hands off to people.
Full evaluationGlossary · Updated Sep 2, 2026
Guardrails
Controls that constrain what an AI system will do or say, such as topic limits, content filters, action permissions, and approvals.
Definition
Guardrails include input and output filters, allowed-tool lists, rate limits, refusal policies, and human approval steps for risky actions. In agents they define which actions can run automatically and which need a person. Good guardrails are testable and logged, and they are tuned to the application's risk rather than applied generically.
Why it matters when choosing a tool
A tool's guardrails determine how safely it can face customers or act on your systems; ask how they are configured, tested, and audited.
Where you will meet it
AI Customer Support, AI Agent & Chatbot Builders, AI Legal & Contracts
Related terms
AI agent · System prompt · Evaluation (evals) · Hallucination
Tools where this matters
Reviewed products in the related categories
AI agents, agent copilots, and automation built into the Zendesk support suite.
Full evaluation
A visual platform for designing, testing, and deploying AI agents for support and conversational products.
Full evaluation
Unified TypeScript library for streaming, structured outputs, and agent loops across AI providers
Full evaluation
Prototyping environment and developer API platform for Google's Gemini models
Full evaluation
A source-available workflow automation platform with native AI agent nodes, self-hostable or cloud.
Full evaluation