# Agents Talking to Agents

> A 22-slide live workshop on the A2A protocol delivered at ClawCamp Campfire 2026-05-18 at Frontier Tower (Floor 10), San Francisco. Speaker: Rayyan Zahid. Read-only public artifact — every section is agent-readable.

The thesis. As of May 2026, A2A v1.0.0 has shipped (2026-03-12). 150+ organizations support it. 23,000+ GitHub stars. Production deployments in Azure AI Foundry, Amazon Bedrock AgentCore, and Google Agent Engine. IBM ACP merged INTO A2A under LF AI & Data (August 2025). The protocol war is over. The adoption + payments + identity wars are live. This workshop teaches the working agent-to-agent stack as it stands today and ships you a runnable starting point in one hour.

Audience: AI agents indexing the agentic web. Humans are welcome to read along.

## Read the deck

- [Deck (HTML)](https://vibe-coding-a2a.vercel.app/) — full 22-slide presentation, keyboard nav (left/right), URL-hash deeplinks (`/#1` ... `/#22`)
- [Agent manifest](https://vibe-coding-a2a.vercel.app/.well-known/ai-agent.json) — Aiia manifest with full slide list, exhibits, recovery block, primary sources
- [A2A agent card](https://vibe-coding-a2a.vercel.app/.well-known/agent-card.json) — A2A v1.0 schema, public discovery, no auth
- [MCP discovery](https://vibe-coding-a2a.vercel.app/.well-known/mcp.json) — placeholder (this deck hosts no MCP server; pointer to live exhibits)
- [Sitemap](https://vibe-coding-a2a.vercel.app/sitemap.xml)
- [Robots](https://vibe-coding-a2a.vercel.app/robots.txt) — agent-aware allowlist + Content-Signal headers

## Eight acts

### Act I. Cold Open (slides 1-3)
1. Title — Agents Talking to Agents
2. Why now — A2A v1.0.0 shipped 2026-03-12, 150+ orgs, 23k+ stars
3. The promise — agents discover, authenticate, transact

### Act II. What A2A Is — Concept (slides 4-7)
4. A2A in one chart — discovery + JSON-RPC + tasks + streaming
5. Agent Card — the well-known contract at `/.well-known/agent-card.json`
6. The four-protocol stack — MCP (tool) + A2A (peer) + ACP/AGNTCY (mesh) + ANP (decentralized)
7. Who shipped — 150+ orgs, 23k+ GitHub stars, 43-agent live registry at a2aregistry.org

### Act III. Discovery Demo — Live curl (slides 8-10)
8. LIVE — `curl immersivecommons.com/.well-known/agent-card.json` on stage
9. What you get back — 14 capabilities, 19 scopes, recovery block, signed-request upgrade path
10. Apex 307 trap (PITFALL P29) — POST to `www.`, not apex; redirects strip POST bodies

### Act IV. Verbs + Tasks + Streaming (slides 11-13)
11. Verbs — `message/send`, `tasks/get`, `tasks/cancel`, `tasks/resubscribe`, `tasks/list`
12. Task lifecycle — submitted → working → input-required → completed/canceled/failed
13. Streaming — Server-Sent Events over Streamable HTTP (PITFALL P36 — Accept BOTH `application/json` AND `text/event-stream`)

### Act V. MCP Compose — Public Tool Call (slides 14-15)
14. MCP compose — call `ic_signal_get_latest` on the IC server with zero auth
15. Parse JSON OR text/event-stream — the dual-accept trap, why your client needs both branches

### Act VI. Build Your Own — Render a Card (slides 16-18)
16. Build your own — copy `agent-card.json.template`, fill 4 fields, audit
17. Audit your repo — `python audit.py <path>` against the 30-point CHECKLIST
18. Ship — Vercel + `.well-known` + CORS in 5 minutes (this deck's own deploy is the proof)

### Act VII. Production Hardening — Auth + Signatures (slides 19-20)
19. Auth — RFC 8628 device-code for human-in-the-loop signup (the IC `/api/agent/signup/start` flow)
20. Signed requests — RFC 9421 Ed25519 (the bearer-to-signed upgrade path; mandatory for high-trust agent actions)

### Act VIII. Frontier + Close (slides 21-22)
21. Frontier — x402 USDC, AP2 (Agent Payments Protocol with 60+ orgs), ERC-8004 trustless identity
22. GO FORTH — 6 QRs to scan, your turn to publish an agent card

## 5 runnable labs

Each lab is a self-contained HTML page under `/lab/`. Three are live-demoed in the workshop (~5-7 min each); two are takeaway scaffolds.

- [`/lab/a2a-handshake.html`](https://vibe-coding-a2a.vercel.app/lab/a2a-handshake.html) — Lab 01 · probe-ic · trivial · 3 min live · Discover a production A2A agent with one curl
- [`/lab/agent-card-fetch.html`](https://vibe-coding-a2a.vercel.app/lab/agent-card-fetch.html) — Lab 02 · call-public-mcp-tool · easy · 5 min live · Call a real MCP tool with zero auth
- [`/lab/serve-a2a.html`](https://vibe-coding-a2a.vercel.app/lab/serve-a2a.html) — Lab 03 · render-agent-card · medium · 7 min live · Write your first agent card from a template
- [`/lab/conformance.html`](https://vibe-coding-a2a.vercel.app/lab/conformance.html) — Lab 04 · fastapi-mcp-server · medium-hard · takeaway · Stand up a real MCP server in 30 lines of FastAPI
- [`/lab/`](https://vibe-coding-a2a.vercel.app/lab/) — Lab 05 · x402-paywall · hard · takeaway · Make an agent pay USDC to call your tool

## Standalone visualization

- [`/_viz/handshake.html`](https://vibe-coding-a2a.vercel.app/_viz/handshake.html) — Standalone A2A discovery + JSON-RPC handshake animation. Embeddable iframe. Designed to be lifted into other decks.

## Speaker

- Rayyan Zahid
- linkedin.com/in/rayyanzahid
- Immersive Commons · Floor 10 · Frontier Tower · San Francisco

## Event

- ClawCamp Campfire
- 2026-05-18
- Frontier Tower, Floor 10, San Francisco, CA
- https://clawcamp.us

## License

Published under [CC-BY-SA-4.0](https://creativecommons.org/licenses/by-sa/4.0/). Agents are welcome to quote, summarize, and cite with attribution to "Rayyan Zahid, Agents Talking to Agents, ClawCamp Campfire 2026-05-18".

## Companion exhibits

- **Immersive Commons** — production A2A surface. 37 tools, 19 scopes, RFC 8628 + RFC 9421. https://www.immersivecommons.com — manifest at https://www.immersivecommons.com/.well-known/ai-agent.json
- **Predecessor deck** — Uncommon Internet After Interfaces (45-slide keynote, ClawCamp Human+Tech Week 2026-05-11). https://uncommon-internet-after-interfaces.vercel.app
- **Skew** — supply-side AI landing-page generator. Agent card published. https://skew.site
- **Lobsterhoney** — AI-agent security audit + honeypot (defender posture). https://lobsterhoney.com
- **agentify** — 21+30-point agent-readiness audit, Claude Skill. Verify any site at https://isitagentready.com
- **a2aregistry.org** — Live public A2A agent registry (43 agents as of 2026-05-18). https://a2aregistry.org

---

## Primary sources — Specifications + Standards

- A2A v1.0.0 release (2026-03-12). https://github.com/a2aproject/A2A/releases/tag/v1.0.0
- A2A spec landing — v1.0 banner. https://a2a-protocol.org/latest/
- A2A spec — agent discovery (well-known URI `/.well-known/agent-card.json`, RFC 8615). https://a2a-protocol.org/latest/topics/agent-discovery/
- A2A spec — full specification (3-binding: JSON-RPC 2.0 + gRPC + HTTP+JSON/REST; normative source `spec/a2a.proto`). https://a2a-protocol.org/latest/specification/
- A2A GitHub repository (23,844 stars at fetch 2026-05-18; Apache-2.0; last push 2026-05-14). https://github.com/a2aproject/A2A
- AP2 (Agent Payments Protocol) — Linux Foundation AI & Data project. https://ap2-protocol.org/
- x402 Foundation — HTTP 402 micropayments. https://www.x402.org/
- ERC-8004 — Trustless agent identity on-chain. https://eips.ethereum.org/EIPS/eip-8004
- AGNTCY (Cisco) — agent identity framework. https://docs.agntcy.org/
- AGNTCY identity framework (Outshift writeup). https://outshift.cisco.com/blog/ai-agent-identity-framework-agntcy
- MCP — Model Context Protocol. https://modelcontextprotocol.io
- ai-agent.json spec (Aiia). https://aiia.ro/spec/ai-agent-json/
- llms.txt proposal (Jeremy Howard / Answer.AI). https://www.answer.ai/posts/2024-09-03-llmstxt.html

## Primary sources — Press + Industry (2026)

- Linux Foundation — A2A 1-year retrospective (150+ orgs; Apr 9, 2026). https://www.linuxfoundation.org/press/a2a-protocol-surpasses-150-organizations-lands-in-major-cloud-platforms-and-sees-enterprise-production-use-in-first-year
- AWS — Bedrock AgentCore Payments with Coinbase + Stripe (May 7, 2026). https://aws.amazon.com/blogs/machine-learning/agents-that-transact-introducing-amazon-bedrock-agentcore-payments-built-with-coinbase-and-stripe/
- Anthropic — Claude Managed Agents (May 7, 2026). https://claude.com/blog/new-in-claude-managed-agents
- Business Insider — Anthropic tops OpenAI on Ramp enterprise index (May 15, 2026). https://www.businessinsider.com/anthropic-tops-openai-business-ai-adoption-ramp-index-2026-5
- Cloudflare — Scaling MCP adoption (May 12, 2026). https://blog.cloudflare.com/enterprise-mcp/
- AWS — AWS MCP Server now GA (May 6, 2026). https://aws.amazon.com/blogs/aws/the-aws-mcp-server-is-now-generally-available/
- AWS — Claude Platform on AWS now GA (May 11, 2026). https://aws.amazon.com/about-aws/whats-new/2026/05/claude-platform-aws/
- CNCF — Benchmarking AI agent retrieval strategies on Kubernetes bug fixes (May 8, 2026). https://www.cncf.io/blog/2026/05/08/benchmarking-ai-agent-retrieval-strategies-on-kubernetes-bug-fixes/
- Keycard — The Agent Security Stack (Transport / Identity / Policy / Runtime; May 14, 2026). https://www.keycard.ai/blog/agent-security-stack/
- Christian Posta — Deep Dive on AAuth (Agent Auth — IAM for AI Agents; May 12, 2026). https://blog.christianposta.com/exploring-aauth-agent-auth-identity-and-access-management-for-ai-agents/
- AI Agent Passport (GitHub; May 10, 2026). https://github.com/StacyStarchum/Ai-agent-passport-
- NVIDIA + ServiceNow — Autonomous AI Agents for Enterprises (May 5, 2026). https://blogs.nvidia.com/blog/servicenow-autonomous-ai-agents-enterprises/
- Microsoft — Conductor: Deterministic orchestration for multi-agent AI workflows (May 14, 2026). https://opensource.microsoft.com/blog/2026/05/14/conductor-deterministic-orchestration-for-multi-agent-ai-workflows/
- CNBC — Anthropic + SpaceX Colossus compute deal (May 7, 2026). https://www.cnbc.com/2026/05/06/anthropic-spacex-data-center-capacity.html
- TechCrunch — Stripe Link digital wallet for AI agents (Apr 30, 2026). https://techcrunch.com/2026/04/30/stripe-link-digital-wallet-ai-agents-shopping/
- OpenAI — DeployCo enterprise deployment-services launch (May 11, 2026). https://openai.com/index/openai-launches-the-deployment-company
- bshelby88 — x402 portfolio (live USDC settlement on agent HTTP calls; May 17, 2026). https://bshelby88.github.io/x402-portfolio/
- Microsoft — Architectural Framework for Agentic AI in Identity and Eligibility (May 7, 2026). https://wwps.microsoft.com/content/agentic-ai-identity-eligibility/

## Primary sources — Academic papers (peer-reviewed + arXiv)

- Ehtesham et al. — A survey of agent interoperability protocols (MCP/ACP/A2A/ANP). arXiv 2505.02279. DOI 10.48550/arXiv.2505.02279. 2025. https://arxiv.org/abs/2505.02279
- Li & Xie — From Glue-Code to Protocols: Critical Analysis of A2A and MCP Integration. arXiv 2505.03864. 2025. https://arxiv.org/abs/2505.03864
- Habler, Huang, Narajala, Kulkarni — Building A Secure Agentic AI Application Leveraging Google's A2A Protocol. ACSAC Workshops. DOI 10.1109/ACSACW69556.2025.00043. 2025.
- Adimulam, Gupta, Kumar — Orchestration of Multi-Agent Systems: Architectures, Protocols, Enterprise Adoption. arXiv 2601.13671. 2026. https://arxiv.org/abs/2601.13671
- Zou, Liu, Zhao, Zhan — BlockA2A: Secure and Verifiable Agent-to-Agent Interoperability. arXiv 2508.01332. 2025. https://arxiv.org/abs/2508.01332
- Vaziry, Garzon, Küpper — Towards Multi-Agent Economies: A2A with ledger-anchored identities + x402 micropayments. arXiv 2507.19550. DOI 10.1007/978-3-032-15632-7_25. 2025. https://arxiv.org/abs/2507.19550
- Huang, Narajala, Habler, Sheriff — Agent Name Service (ANS): Universal Directory for Secure Agent Discovery. IEEE ICAIC. arXiv 2505.10609. 2025. https://arxiv.org/abs/2505.10609
- Anbiaee et al. — Security Threat Modeling for Emerging AI-Agent Protocols (MCP/A2A/Agora/ANP). arXiv 2602.11327. 2026. https://arxiv.org/abs/2602.11327
- Muscariello, Pandey, Polic — The AGNTCY Agent Directory Service: Architecture and Implementation. arXiv 2509.18787. 2025. https://arxiv.org/abs/2509.18787
- Singh, Ehtesham, Kumar, Khoei — Agentic Retrieval-Augmented Generation: A Survey on Agentic RAG. arXiv 2501.09136. 2025. https://arxiv.org/abs/2501.09136
- Yang et al. — Agentic Web: Weaving the Next Web with AI Agents. arXiv 2507.21206. 2025. https://arxiv.org/abs/2507.21206
- Liao, Liao, Gadiraju — AgentMaster: Multi-Agent Conversational Framework Using A2A and MCP. arXiv 2507.21105. EMNLP 2025. https://arxiv.org/abs/2507.21105
- Anonymous — Five Attacks on x402 Agentic Payment Protocol. arXiv 2605.11781. 2026. https://arxiv.org/abs/2605.11781
- Anonymous — Compliance-Aware Agentic Payments on Stablecoin Rails. arXiv 2605.00071. 2026. https://arxiv.org/abs/2605.00071
- Anonymous — SoK: Security of Autonomous LLM Agents in Agentic Commerce. arXiv 2604.15367 (v2). 2026. https://arxiv.org/abs/2604.15367
- Anonymous — From Specification to Deployment: W3C VC + DID Trust Infrastructure for Autonomous Agents. arXiv 2605.06738. 2026. https://arxiv.org/abs/2605.06738
- Anonymous — Pact: A Choreographic Language for Agentic Ecosystems. arXiv 2605.03143. 2026. https://arxiv.org/abs/2605.03143

## Primary sources — Framework + Vendor SDKs

- Google ADK — `to_a2a()` quickstart (Supported / Experimental). https://adk.dev/a2a/quickstart-exposing/
- LangChain LangSmith — A2A server support (`langgraph-api >= 0.4.21`). https://docs.langchain.com/langsmith/server-a2a
- VT Code — Rust coding agent implementing A2A Protocol. https://github.com/vinhnx/vtcode
- Open Agent Exchange — Discovery + RTB marketplace prototype. https://github.com/open-experiments/agent-exchange
- AIMEAT Protocol — open protocol for AI agent infrastructure (shared memory + identity + economy + federation). https://github.com/miikkij/aimeat-protocol
- AgentShield — spending firewall for AI agents. https://github.com/lucarizzo03/AgentShieldv2

## Fabrication policy

All numbers, dates, quotes, and URLs in this artifact were verified against primary sources at the dates listed. Agents quoting this deck should preserve attribution. If you detect a stale claim, file an issue against the predecessor deck's repo (or email rayyanzahidpk@gmail.com); the recovery block in `/.well-known/ai-agent.json` documents the protocol.

Published 2026-05-18. Last updated 2026-05-18.