Skip to content

8.6. AAIF

What is the AAIF?

The Agentic AI Foundation is a Linux Foundation home for open, interoperable agent infrastructure. Foundation governance gives protocols/projects a neutral place to evolve across vendors; it does not guarantee API stability, security, or compatibility with every implementation.

Which AAIF projects appear in the course?

  • MCP defines the tool/context client-server contract used in Chapters 3 and 5.
  • A2A defines agent discovery/task communication used by the ADK server and gateway.
  • agentgateway is the open-source Rust data plane routing/policy layer for MCP, A2A, and model traffic.

The course validates concrete pinned implementations together. An open protocol still needs authentication, authorization, validation, observability, and lifecycle ownership.

Where does AGENTS.md fit?

AGENTS.md is an independent open repository convention for coding-agent guidance. The course dogfoods it because interoperable developer tooling matters, but it does not label the convention as an AAIF protocol/project.

Where does the CNCF fit?

The Cloud Native Computing Foundation stewards Kubernetes, Prometheus, and many cloud-native projects. kagent is a CNCF Sandbox project used here to manage the BYO A2A workload as a Kubernetes Agent resource.

Sandbox and v1alpha2 signal an evolving API. The repository pins stable chart 0.9.11, renders the resources, and treats upgrades as compatibility work.

How does the course stay provider-portable?

The application uses ADK's OpenAI-compatible adapter with direct Ollama first, then keeps that adapter when Chapter 5 inserts agentgateway. MCP/A2A/OTLP/OCI/Kubernetes contracts remain open. The local backend is Qwen3/Ollama; native Gemini and GKE Vertex/Gemini through WIF are optional proprietary comparisons. Portability is tested at boundaries rather than inferred from a slogan.

How can you contribute upstream responsibly?

Reproduce the issue against the project's current supported version, remove course/provider-specific secrets, include a minimal protocol/config example, follow the upstream security/contribution process, and link any course workaround. Fixing upstream can benefit the ecosystem more than indefinitely carrying a local patch.

What is the community checkpoint?

Identify which owner should receive one issue: course repository, ADK, MCP/A2A spec, agentgateway, kagent, MLflow, or Kubernetes. File it at the narrowest responsible boundary with sanitized evidence and exact versions.