nOps speeds FinOps agent delivery by 75 percent with Amazon Bedrock AgentCore

nOps, which manages over $4 billion in cloud spend for customers, transitioned its FinOps analytics agent Clara to Amazon Bedrock AgentCore to overcome limitations of its previous API-centric infrastructure, such as latency, complexity, and high maintenance. The new architecture uses a single Strands-based agent on Amazon Bedrock AgentCore with direct tool access, a data layer built on Databricks Lakehouse Metric Views for governed analytics, and Databricks Lakebase for storing persistent state like sessions and canvas specs. Interaction is handled by a Vercel-hosted Next.js application with streaming Server-Sent Events, and async workflows use Amazon DynamoDB, Amazon SNS, Amazon SQS, and Amazon API Gateway WebSocket. This restructuring cut production time from 10-12 months to 4 months (a 75 percent reduction), increased the Correctness score to 81.7 percent and Helpfulness to 79.4 percent, reduced tool failure rate from 7.49 to 0.92 percent, and eliminated the LangGraph/LangChain orchestration layer. The company also improved manual analysis efficiency from 2 hours to 30 minutes for customer success teams, and they can now run 4-6 production agents on a shared runtime.