How Cornerstone OnDemand Cut Database Diagnosis Time by 78% Using Amazon Bedrock and Strands Agents

← AI for Your Work

WebCrimp analysis and practical explanation. AI-assisted; reviewed before publication.

Quick Summary

Cornerstone OnDemand built a multi‑agent AI system called Orion AI that automates database diagnostics and lifecycle tasks. By delegating work to specialized agents coordinated through a hub‑and‑spoke pattern, the team reduced manual diagnosis from 45 minutes to 10 minutes—a 78 % improvement—and streamlined cross‑team reporting and alert handling.

What Happened?

Before Orion AI, the Enterprise DataOps team spent about 45 minutes per incident manually querying SQL Server system views, correlating logs, and coordinating with SRE and data teams. The process involved more than ten manual steps for lifecycle changes and suffered from a 15‑minute reporting lag and noisy alerts. Cornerstone implemented Orion AI using Amazon Bedrock for foundation models and Strands Agents for orchestration. The system features a meta‑orchestrator that routes requests to domain‑specific agents (diagnostics, blocking analysis, real‑time SQL, lifecycle, analytics, knowledge, notification, etc.). Agents use keyword‑first routing with semantic search fallback, bypass conversational memory for live metrics, and apply custom guardrails for safety. Results showed diagnosis time dropping to 10 minutes, lifecycle steps collapsing to a single interaction, reporting lag becoming instantaneous, and redundant alerts reduced by a median of 65 %.

Why It Matters

The case demonstrates how a well‑designed multi‑agent architecture can turn reactive, time‑consuming database operations into proactive, self‑orchestrating workflows. It cuts mean time to resolution, reduces alert fatigue, eliminates manual handoffs, and frees engineers to focus on higher‑value work. The approach also illustrates practical guardrail patterns, memory scoping, and hybrid routing that other teams can adopt to improve reliability and efficiency in data‑intensive environments.

Who Can Use It?

AI/ML · Cloud · DevOps · Data Analytics · Software Development · Operations · Management

How Can I Use This?

If you work in AI/ML, Cloud, DevOps, Data Analytics, Software Development, Operations, Management

• DevOps engineers can replicate the hub‑and‑spoke pattern by defining agents for each operational domain (e.g., monitoring, diagnostics, lifecycle) and using a lightweight orchestrator to route requests. Data analysts can leverage the specialized diagnostics agents to automatically surface blocking chains, wait types, and long‑running queries, then validate the output before acting. Cloud architects can adopt the keyword‑first routing with semantic search fallback to keep latency low while preserving accuracy for ambiguous queries. All roles should implement custom guardrails (prompt‑level constraints, human‑in‑the‑loop confirmation gates, output sanitization) and scope conversational memory to sessions while bypassing it for live metric queries to avoid stale data. Workflow suggestions need verification in your own environment before production deployment.

AI Tip

Start with a narrow set of tools per agent; this keeps the model context focused and improves tool selection accuracy.

AI Trick

Use an in‑memory keyword lookup for the majority of predictable requests (≈80 %) and fall back to semantic search only when needed, which keeps routing latency under a millisecond for most queries.

Example Workflow

1. 1. Define domain‑specific agents (privacy, diagnostics, lifecycle, analytics, knowledge, notifications). 2. Equip each agent with tools via MCP or direct SDK calls, wrapped as Strands @tool functions. 3. Implement a meta‑orchestrator that uses keyword routing with semantic search fallback to select agents. 4. Scope conversational memory to session scope; bypass memory for queries that require current system state. 5. Add custom guardrails: prompt‑level safety, human‑in‑the‑loop confirmation gate, output sanitization, and rate limiting. 6. Deploy agents on a container platform (e.g., Amazon ECS) and expose them through a web UI for natural‑language interaction. 7. Monitor with CloudWatch and X‑Ray, iterating on routing confidence and latency.

Copy This Prompt

Investigate the recent performance incident on {database_name} occurring around {timestamp}. Provide: (1) current blocking chains and wait types, (2) likely root cause with prioritized remediation steps, (3) a ready‑to‑use Jira ticket summary and description assigned to the on‑call engineer for {team}.

Recommended Tools

• Amazon Bedrock
• Strands Agents
• Amazon ECS
• AWS SDK/Boto3
• Jira integration via MCP
• Portal‑Tools MCP server
• Amazon DynamoDB
• Amazon CloudWatch
• AWS X-Ray

Check official documentation, availability and pricing before using these suggestions.

Difficulty

Intermediate

Estimated Time Saved

Not measured; validate with your own workflow.

Source

AWS AI
How Cornerstone OnDemand cut database diagnosis by 78% with Amazon Bedrock
Original publication date: 2026-10-07T15:38:28+00:00