Who’s building agents vs

Описание: Shipping AI agents to production is hard—security, compliance, eval, and observability all bite. In this talk, Raahul Dutta (Founder, Pebbling.ai) breaks down how to make agents talk to each other securely across orgs, stacks, and clouds using the Pebbling protocol.

You’ll learn:
Why most teams can prototype agents but struggle to deploy to production
The missing layer: a secure, interoperable A2A protocol (beyond MCP/HTTP)
How peer-to-peer, mTLS, and DIDs (decentralized identifiers) create trust between agents
Negotiation + micropayments workflows between agents
Pebbling’s open-source stack:
Pebbling Protocol – secure P2P transport and message schemas
Hibiscus – federated agent registry for discovery + reputation
Sheldon – certificate authority / trust service
Imagine – multi-party orchestration for complex use cases (e.g., drug discovery, clinical trials)
Deployment model: framework-agnostic (LangChain/LangGraph, OpenAI SDKs, LlamaIndex, Rust, Python, etc.) with a “single line” cookie-cutter to register and go live
How Pebbling differs from A2A/ACP/MCP and why developer trust & openness matter

Real-world pain points addressed:
Risk & compliance (enterprise controls, auditability)
Observability & online evaluation for agent behaviors
Interoperability across languages, vendors, and clouds
Data silos and cross-org collaboration (P2P, no central snooping)

If you’re building serious AI agents for banks, pharma, or large enterprises, this is a blueprint for secure, production-grade agent ecosystems.

Chapters
00:00 Intro & speaker background
01:05 Who’s building agents vs. shipping to prod
02:15 Top blockers: security, compliance, observability, eval
03:25 Vision: one-line deploy, agents speaking securely P2P
04:30 Pebbling protocol overview (DIDs, mTLS, schemas)
06:05 Product suite: Hibiscus (registry), Sheldon (CA), Imagine (orchestration)
08:10 Enterprise use case: cross-site data silos in pharma
10:05 Pebbling vs A2A/ACP/MCP (trust, negotiation, micropayments)
12:10 Dev experience & hosting options (cloud, k8s, local)
13:40 Q&A: nefarious agents, reputation, federation & latency