Customize AI in Amazon Connect Customer: Agents, Prompts, and Guardrails

This article explains how to customize AI in Amazon Connect Customer using AI Agents, Prompts, and Guardrails to control generative AI interactions with customers and support for human agents.

• AI Agents autonomously handle customer interactions and provide agent assistance across voice and digital channels
• AI Prompts define task-specific instructions with nine template types including orchestration, answer generation, and case summarization
• Select preferred LLMs (Claude Sonnet, Nova Pro, Nova Lite, Claude Haiku) directly in the console for each prompt type
• AI Guardrails enforce safety through content filters, denied topics, word filters, PII detection, hallucination checks, and custom blocked messages
• Claude Sonnet recommended for complex tasks; Nova Lite/Haiku for speed-critical tasks; Nova Pro for self-service interactions
• Cross-region inference available to optimize performance and availability across AWS regions
• Best practices include positioning static content before variables, using 1,000+ token prefixes for prompt caching, and monitoring performance

Contact centers can now optimize AI performance for speed, accuracy, and cost by selecting the right model for each use case directly from the admin console.