# Builder's Daily — Multi-Agent Interop

> Rolling 14-day signal for beat `multi-agent-interop`. Ephemeral context — not evergreen corpus.
> Author: Amit Kumar Agrawal | https://artificialcuriositylabs.ai
> Generated: 2026-06-06
> Human index: https://artificialcuriositylabs.ai/daily/multi-agent-interop/
> RSS: https://artificialcuriositylabs.ai/daily/multi-agent-interop/rss.xml

---

# Multi-Agent Interop — June 6, 2026
**URL:** https://artificialcuriositylabs.ai/daily/multi-agent-interop/2026-06-06/
**Beat:** multi-agent-interop
**Date:** 2026-06-06
**Topics:** a2a, agent-card, discovery, microsoft, delegation, hosted-agents
**Summary:** Microsoft Foundry ships incoming A2A public preview with versioned agent cards and Entra-gated discovery; Hosted agents in Foundry add A2A as fourth pro…

## The read

A2A and registries wire agents together — protocol, not moat. Value sits in delegation topology humans design and the institutional rules that govern who can act on whose behalf.

## What moved

- **Microsoft Foundry ships incoming A2A public preview with versioned agent cards and Entra-gated discovery** — [Microsoft Foundry Blog](https://devblogs.microsoft.com/foundry/from-building-agents-to-working-with-them-enterprise-agent-distribution-in-microsoft-foundry/)
  At Build 2026, Foundry Agent Service adds incoming Agent2Agent (public preview) alongside existing outbound A2A tools. Any prompt or responses-protocol hosted agent can expose an A2A base path, publish v1.0 and v0.3 agent cards at agentCard/{version}, and accept JSON-RPC delegation from external agents. Agent card fetch and invocation require Microsoft Entra ID (OBO or service identity); anonymous discovery is not supported. Foundry documents A2APreviewTool connections with custom AgentCardPath for cross-agent calls within the same project. **Builder angle:** Bidirectional A2A on Foundry lets you publish a governed agent card once and accept delegated tasks from LangGraph, ADK, or other A2A clients without rebuilding the agent per partner stack.

- **Hosted agents in Foundry add A2A as fourth protocol for cross-agent delegation on production runtimes** — [Microsoft Foundry Blog](https://devblogs.microsoft.com/foundry/hosted-agents-build26/)
  Build 2026 updates to hosted agents extend the protocol surface beyond Responses, Invocations (HTTP), and Invocations (WebSocket) with Agent2Agent (A2A) for agent delegation. A single hosted agent can expose multiple protocols simultaneously, including Teams/M365 channel bridging and A2A handoffs. The post also ships direct source-code deployment (python_3_13/3_14, dotnet_10) without container registry, built-in Content Safety guardrails, and Voice Live/WebSocket for real-time voice agents. **Builder angle:** Adding A2A beside Responses/Invocations means production hosted agents can both answer users and accept delegated sub-tasks from remote specialist agents on the same deployment.

## Also tracking

- **A2A v1.0 reaches production with Signed Agent Cards and 150+ enterprise adopters** — [source](https://www.linuxfoundation.org/press/a2a-protocol-surpasses-150-organizations-lands-in-major-cloud-platforms-and-sees-enterprise-production-use-in-first-year) — Linux Foundation A2A v1.0 (April 9) adds Signed Agent Cards, multi-tenancy, and SDKs in five languages; embedded in Azure Foundry, Copilot Studio, and Bedrock AgentCore Runtime.
- **Amazon Bedrock AgentCore Runtime adds native A2A protocol with Agent Card discovery** — [source](https://aws.amazon.com/blogs/machine-learning/introducing-agent-to-agent-protocol-support-in-amazon-bedrock-agentcore-runtime/) — AgentCore hosts stateless A2A servers on port 9000 with SigV4/OAuth inbound auth, session isolation headers, and /.well-known/agent-card.json discovery for cross-framework delegation.