As autonomous agents proliferate across enterprise stacks, a familiar problem is re-emerging: fragmentation. Each team builds its own orchestration. Each vendor defines its own protocol. And each framework handles identity, discovery, and messaging differently — if at all. While agents are supposed to collaborate, today they often operate in silos.

As organizations deploy agents from different vendors and frameworks, the lack of standardized communication protocols is becoming a significant barrier to scaling AI initiatives. Enter AGNTCY, an open-source infrastructure project that recently moved under Linux Foundation governance, promising to solve interoperability challenges that could define the future of enterprise AI.

From Cisco Lab to Linux Foundation

AGNTCY originated from a practical problem at Cisco’s Outshift incubator 18 months ago. “When we started integrating agentic software into our environment, we realized you can’t have a singular agent solve for larger goals,” explains Vijoy Pandey, SVP and GM at Outshift by Cisco. “We’re moving from deterministic computing to probabilistic computing, and that requires subject matter expert agents that collaborate.”

The project, initially launched in March with partners LangChain and Galileo, has rapidly expanded to include over 75 contributing organizations. The move to Linux Foundation governance brings heavyweight backing from five founding members: Cisco, Dell Technologies, Google Cloud, Oracle, and Red Hat.

“We picked Linux Foundation because they have the neutrality of governance and the resources to nurture and grow communities,” Pandey notes. “It’s similar to how Google donated Kubernetes to CNCF, creating an ecosystem that spawned hundreds of projects and startups.”

Four Pillars of Agent Infrastructure

AGNTCY addresses four critical areas of agent communication: discovery, identity, messaging, and observability. AGNTCY isn’t an agent runtime, LLM framework, or orchestration tool. It’s infrastructure — designed to serve as the communication substrate for distributed, probabilistic systems.

At its core, AGNTCY  defines a modular stack comprising:

• OSF (Open Agent Schema Framework): A flexible schema layer that can describe individual agents, multi-agent apps, and even agent-to-tool protocols like MCP.
• Decentralized Agent Directory: Think DNS for agents — but built on distributed hash tables and P2P principles to ensure no single point of control.
• Decentralized Identity Service: Designed for task-based authorization across dynamic, non-deterministic agent interactions. Identity is extensible, privacy-aware, and not bound to a central authority.
• SLIM (Secure, Low-latency Interactive Messaging): A gRPC-based communication layer that supports many-to-many interactions, voice/video support, real-time guarantees, and post-quantum cryptography.

What binds all these components together is a deep focus on interoperability at scale — across clouds, frameworks, and vendors.

“It’s not enough to define agent protocols,” says Pandey. “We need discoverability, real-time messaging, authorization models, and observability — all designed for agents that behave more like humans than scripts.”

Why Now?

The agentic shift is already underway. Enterprises are deploying LLM-powered systems that delegate tasks to specialized agents. Frameworks like LangChain, CrewAI, and Autogen are becoming developer staples.

But today’s agent ecosystems are fragile. Each implementation is bespoke. There’s no standard way to discover agents, assign identity, or define multi-agent workflows. Communication protocols vary wildly. And few systems can monitor or debug agent behavior at runtime.

AGNTCY  addresses these gaps with a comprehensive architectural approach — and early traction suggests it’s resonating.

Use Cases Emerging — Fast

Despite being newly open sourced, AGNTCY  is already in production across several high-impact domains.

• Voice AI: SoftServe, in partnership with Cisco’s Webex unit, built a multi-agent voice platform using AGNTCY ’s identity and SLIM messaging layers.
• SRE Automation: Outshift open sourced a multi-agent SRE assistant internally known as JARVIS — now rebranded as the AI Platform Engineer — automating 30% of their SRE team’s workflows.
• Network Configuration: Telecom providers (including Swisscom) are using AGNTCY  in proprietary tools to automate and validate network migrations and architecture changes.
• Reference Architecture: A new app called coffeeAGNTCY mirrors CNCF’s Sock Shop — providing a multi-agent, multi-organization, vendor-neutral reference system that developers can fork and build upon. “It integrates A2A, AGNTCY, MCP, and in the future will integrate other open-source projects,” Pandey explains. “It’s the most important piece of code we want to open source because it gives developers a working example to experiment with.”

“We’ve also worked with telecom providers like Swisscom to build network configuration automation and validation tools using AGNTCY,” Pandey reveals. “The entire pipeline for building new networks, migrations, repairs—all of that runs on multi-agent workflows.”

Challenges Ahead: Identity and Scale

While the components are live and the code is open, Pandey emphasized that identity and messaging remain the biggest focus areas going forward. “We used to think discovery comes first, then identity. But for enterprise adoption, identity, authorization, and security are becoming the key problems to solve.”’

The scale requirements are staggering. “We’re not talking millions of agents—we’re headed toward trillions because every human and enterprise will have multiple agents representing them,” Pandey notes. “The architecture needs to be scalable from day one.”

The team is doubling down on:

• Task-based authorization for agents with ambiguous or emergent behavior
• Real-time, many-to-many communication for voice, video, and multimodal interaction
• Secure messaging with PQC baked into every layer

“If agents are going to negotiate, delegate, and learn in real time,” Pandey says, “we need messaging that behaves more like a secure conversation than an API call.”The broader ecosystem will likely define where AGNTCY  goes next — but the foundations are in place.

Positioning for the Future

AGNTCY’s approach of building around existing protocols rather than replacing them could prove crucial for adoption. The platform integrates with Google’s A2A (Agent-to-Agent) protocol and Anthropic’s MCP (Model Context Protocol), positioning itself as complementary infrastructure rather than competitive technology.

“We’re not about specific protocols,” Pandey clarifies. “We provide the missing pieces around protocols to create a working stack for multi-agent architectures.”

As AI agents become increasingly central to enterprise operations, the standardization battle reminds many of the early internet’s evolution. Whether AGNTCY becomes the TCP/IP of agent communication remains to be seen, but with Linux Foundation backing and growing enterprise adoption, it’s positioned to play a significant role in shaping how AI agents communicate in the enterprise.

For developers interested in contributing or exploring AGNTCY, the project is available at AGNTCY.org, with comprehensive documentation, code access, and the coffeeAGNTCY reference application ready for experimentation.

Get Involved

Developers, operators, and contributors can explore the project at AGNTCY.org. There you’ll find docs, Slack links, and reference apps like coffeeAGNTCY.

The code is open. The architecture is modular. And the roadmap is increasingly community-driven.

In Pandey’s words: “We believe in an open, interoperable internet of agents — and that starts with shared infrastructure.”