AWS re:Invent 2025 - Observability for AI Agents and Traditional Workloads (COP335)

Observability for AI Agents and Traditional Workloads

Key Challenges Addressed

• Application Complexity: Increased use of microservices and AI agents creating new dependencies and behaviors that are difficult to understand.
• Standardization: Lack of consistency in metrics, logs, and transaction representation across services and teams.
• Prioritization: Difficulty distinguishing customer-impacting incidents from less critical anomalies.
• Disjoint Experiences: Multiple disconnected tools for logs, traces, and metrics, making it hard to correlate issues.
• Increased Entropy: AI agents can introduce new dependencies and behaviors that further increase system complexity and unpredictability.

Application Signals: AWS's Observability Solution

• Provides discovery of services, dependencies, and standardized operational practices.
• Allows defining digital journeys (e.g. mobile user logins, payments) to prioritize monitoring.
• Correlates metrics, logs, and traces in a unified experience for faster root cause analysis.
• Collects telemetry using OpenTelemetry for managed services and customer applications.

Key Features and Enhancements

1. Complete Application Picture:

   • Automatically discovers dependencies and topology even for uninstrumented services.
   • Reduces manual work to understand application structure.

2. Organized Observability:

   • Automatically detects and groups applications based on connectivity and attributes.
   • Allows organizing observability by business units, teams, etc. to match how users work.

3. Faster Issue Resolution:

   • Provides automatic operational audits and change impact analysis.
   • Enables quick identification of important, customer-impacting issues.

4. AI Productivity Integration:

   • Connects observability data to users' own AI/ML tools and agents.
   • Provides operational insights and audits to enhance AI agent productivity.

AI Agent Observability

• Instruments AI agents using OpenTelemetry to capture their interactions and behaviors.
• Allows visibility into how AI agents are using databases, responding to users, etc.

Mobile Observability

• Expanded Real User Monitoring to cover iOS and Android mobile apps.
• Provides visibility into mobile app crashes, user impact analysis, and source-level debugging.

Uninstrumented Service Observability

• Automatically discovers and maps dependencies for services without instrumentation.
• Supports automatic instrumentation for EKS clusters and other managed services.

Cross-Account Observability

• Provides a unified view of applications and dependencies across multiple AWS accounts.
• Allows tracking shared resources and cross-account interactions.

AI Ops Integration

• Provides MCP (Model Correctness Principle) servers to enhance AI agent productivity.
• Integrates with GitHub Actions to leverage production telemetry for developer workflows.

Business Impact and Customer Results

• CCC (a leading auto claims provider) saw:
  • 50% reduction in mean time to resolution (MTR)
  • 40% cost savings
  • Improved developer, SRE, and testing team productivity
  • Proactive issue detection and resolution, leading to better system performance

Key Takeaways

• AWS Application Signals provides comprehensive observability for complex, AI-powered applications.
• Features like automatic discovery, organized observability, and fast issue resolution help teams be more productive.
• Integrating observability data with AI agents and developer workflows further enhances productivity.
• Customers like CCC have seen significant improvements in cost, efficiency, and system performance.