Build a serverless image editing agent with Amazon Bedrock AgentCore harness

This article demonstrates building a serverless image editing agent using Amazon Bedrock AgentCore harness, which handles orchestration, tool routing, and memory management through configuration rather than custom code.

• AgentCore harness eliminates orchestration code—define agents entirely through API parameters with no Python framework or container needed
• Per-invocation model switching routes basic chat to Claude Haiku and image edits to Claude Sonnet while preserving conversation context
• Per-invocation persona override injects domain-specific system prompts (Real Estate, Retail, Automotive) without redeployment
• AgentCore Memory stores 30-day conversation history, enabling agents to retain full context across turns within a session
• AgentCore Gateway with MCP exposes three Lambda-backed image editing tools with semantic routing for automatic tool selection
• InvokeAgentRuntimeCommand runs Python scripts directly on the microVM for post-processing like watermarking with zero token cost
• Full solution deploys with single command using AWS CDK, including React frontend, authentication, encrypted storage, and three image editing tools

AgentCore harness enables production-ready agents with straightforward tool-calling patterns by removing operational overhead and allowing behavior iteration through configuration changes.