A1Mobilsoft's Journey to Automating Operations Inquiries (2): Implementing Three Agents and Human vs. AI Validation | Amazon Web Services

A1Mobilsoft, MegazoneCloud, and AWS published a follow-up case on automating operational inquiries with a 3-agent AI setup.

Using Bedrock AgentCore and Strands, they built Knowledge, Code, and DB agents to handle Slack-based support questions.

They reduced wrong answers by re-evaluating vector search results with an LLM diagnosis gate and tuning confidence thresholds.

The team also documented implementation issues, including JSON parsing with Claude Haiku, and validated the system by comparing human and AI decisions across 14 parallel cases.