AWS re:Invent 2025 - Building agentic AI platform engineering solutions with open source (OPN303)

Building Agentic AI Platform Engineering Solutions with Open Source

Overview

• This presentation discusses how organizations can leverage open source tools and technologies to build AI-powered platform engineering solutions.
• The speakers, Neil Thompson from AWS and Hassid Kalpag from Cisco's incubation unit Outshift, cover the current state of platform engineering, the role of AI, and a real-world example of an agentic AI platform built using open source components.

The State of Platform Engineering

• Platform engineering continues to be a popular approach for organizations to operationalize on top of cloud platforms like AWS.
• 76% of organizations have at least one dedicated platform team, responsible for building centralized capabilities like:
  • Golden paths to production (standardized CI/CD)
  • Self-service developer portals (e.g. Backstage)
  • Out-of-the-box observability
  • Compute abstractions (e.g. Kubernetes, ECS, Lambda)
• However, building an effective platform engineering practice can be challenging, often leading to developer frustration and platform teams spending too much time on handholding.

The Rise of AI in Platform Engineering

• Developers are increasingly using AI, with 90% reporting using AI in their daily work and 80% saying it has made them more productive.
• As platform engineers, the challenge is to dovetail AI with platform abstractions to ensure developers are productive within the organization's platforms.
• AI can be leveraged across the developer workflow, not just for code generation, but also for tasks like:
  • Troubleshooting CI/CD pipeline issues
  • Navigating self-service platform capabilities
  • Optimizing cloud costs
  • Automating security and vulnerability remediation

A Real-World Example

• The presentation showcases a demo of an agentic AI platform built using open source components.
• Key elements include:
  • An agent framework (e.g. Strands) to power the AI agents
  • Model Context Protocol (MCP) servers to provide agents access to platform data and actions
  • Agent-to-Agent (A2A) protocol for agent collaboration
  • Integration with developer tools like CLIs, IDEs, and messaging platforms

Technical Deep Dive

• Agent frameworks like Strands, Langchain, and others provide the core functionality for building AI agents, including:
  • Model flexibility to keep up with evolving AI models
  • Session management and memory
  • Observability to understand agent behavior
• MCP servers allow agents to access platform-specific data and take actions, e.g.:
  • AWS API, Kubernetes, Backstage, GitHub, etc.
  • Enables agents to be tailored to the organization's specific platform
• A2A protocol enables agents to discover, collaborate, and delegate tasks to each other, improving specialization and efficiency.

Cisco's Outshift Platform

• Cisco's incubation unit, Outshift, has been building an agentic AI platform engineering solution using these open source components.
• Key features include:
  • Centralized AI agent that can be accessed through various developer touchpoints (CLI, Slack, Jira, etc.)
  • Automated provisioning of development resources (EC2 instances, etc.)
  • Incident management and production troubleshooting assistance
  • Significant reduction in platform engineering support desk overhead

The Open Source CAPE Project

• Cisco has open-sourced their agentic AI platform engineering solution as the CAPE (Cloud-native AI Platform Engineering) project.
• CAPE provides a scalable, open-source system that can be applied in production, including:
  • Built-in knowledge base
  • Pre-built agents for various platform capabilities
  • Integrations with developer workflows and tools
  • Extensible architecture based on MCP, A2A, and other open standards

Key Takeaways

• Open source is providing a strong foundation for building agentic AI platform engineering solutions, with frameworks, protocols, and pre-built components.
• Integrating AI into platform engineering can improve developer productivity, reduce operational overhead, and enable more autonomous platform capabilities.
• Cisco's Outshift platform and the open-source CAPE project demonstrate a real-world example of these principles in action.
• Organizations should explore how they can leverage AI and open source to enhance their platform engineering practices and better support their developers.