AWS re:Invent 2025 - Ditch your old SRE playbook: AI SRE for root cause in minutes (AIM260)

Summary of AWS re:Invent 2025 Presentation: "Ditch your old SRE playbook: AI SRE for root cause in minutes (AIM260)"

Introduction

• Presenters: Peter Kantos (Founder & CEO of Resolve AI), Jos, and Angelo
• Resolve AI builds AI agents that can use production tools like a software engineer to accelerate incident troubleshooting, remediation, and broader production management

The Challenge of Managing Modern Production Systems

• Software engineering has two key parts: coding/building and running production
• AI has significantly impacted the coding/building side (e.g. GitHub Copilot, autonomous code generation), but the production/operations side remains a major challenge
• Key issues with managing modern production systems:
  • Complexity of the full software stack, with siloed and unintegrated tools
  • Dynamic and distributed infrastructure, often across multiple clouds
  • Need for diverse expertise (application, infrastructure, platform)
  • Lack of complete documentation, with knowledge spread across tools and human minds
• This leads to high costs, frequent war rooms/escalations, and infrastructure over-provisioning, especially for brownfield applications

The Rise of AI for Site Reliability Engineering (SRE)

• Humans are no longer the sole operators of production tools and systems
• Transition from humans as the operators to agents as the operators, with humans managing the agents
• Next step is for agents to become both the operators and the "glue" that stitches together information from disparate sources
• Key requirements for effective AI SRE agents:
  • Understand the full production environment like a human
  • Learn from every interaction and improve over time
  • Operate production tools with the same capabilities as humans
  • Discover and encode human expertise into the agents

Resolve AI's Approach

• Resolve AI is a multi-agent system, with agents specialized for code, logs, metrics, infrastructure, knowledge, and documents
• Agents work together to accomplish tasks like incident troubleshooting and remediation
• Agents learn from past incidents and interactions to improve their performance over time
• Agents can operate production tools, gather evidence, and provide recommendations, allowing humans to focus on higher-level tasks

Coinbase's Experience with Resolve AI

• Coinbase, a large cryptocurrency exchange, has been using Resolve AI to address production incidents
• Key principles for Coinbase's adoption:
  • Validate the agent's performance against real-world incidents
  • Test the agent's ability to handle vague prompts and lack of domain knowledge
  • Integrate the agent with key data sources like Kubernetes and Datadog
  • Provide the agent with general and team-specific knowledge
• Resolve AI has been able to identify root causes of incidents faster than humans, with over 50% accuracy
• Future plans include further integration to allow the agent to propose and execute remediations automatically

Key Takeaways

• AI agents can be as effective in production operations as they have been in the coding/building side of software engineering
• Effective AI SRE requires agents that can understand the full production environment, learn from interactions, and operate production tools with human-level capabilities
• Coinbase's experience demonstrates the potential for AI to accelerate incident resolution and free up engineers to focus on higher-value work
• Adoption of AI SRE is expected to grow rapidly, with the potential for transformative productivity gains in the software industry