Composable by Default: AGNTCY and the Future of AI Agents

Originally open-sourced in early 2025 with contributions from Cisco’s Outshift, LangChain, and Galileo, AGNTCY quickly grew into a global community effort. In July 2025, the project was formally donated to the Linux Foundation, ensuring neutral governance and long-term sustainability. Today, AGNTCY counts more than 75 individual contributors and organizational members, ranging from platform builders to systems integrators.

Notable participants now include Dell Technologies, Google Cloud, Oracle, and Red Hat, alongside Cisco. Agentic pioneers like CrewAI, LlamaIndex, Komodor and Redis and service providers like Orium and others who have joined alongside the founding members to strengthen AGNTCY’s role as a shared foundation for agent-to-agent interoperability.

“Interoperability isn’t something a single company can solve. For AGNTCY to matter, it has to be open, inclusive, and shaped by a broad coalition. That’s what gives enterprises confidence this won’t just be another proprietary standard with an expiration date," says John Parello, Principal Engineer at Outshift.

Rather than reinvent workflows per use case, AGNTCY enables agents to self-describe, discover each other, and collaborate securely, regardless of how or where they were built. It complements and integrates with MCP (Model Context Protocol, for tool access) and A2A (Agent-to-Agent task delegation).

For organizations embracing composable architectures, AGNTCY signals a modular future where agents behave more like services— with trust, observability, and plug-and-play integration built in.

Despite their growing sophistication, most AI agents today are hard-coded for single roles, built into specific runtimes, and blind to their ecosystem. The result is a brittle architecture defined by tight coupling, manual integration, and limited reuse.

When enterprises try to scale multi-agent systems across domains, the friction is immediate:

• Agents can’t discover each other dynamically
• Capabilities and credentials aren’t standardized
• Orchestration logic becomes a brittle web of bespoke connectors

In composable systems, these challenges are already familiar— and the answer isn’t more custom code. It’s shared contracts and open standards. That’s precisely what AGNTCY aims to provide.

AGNTCY introduces two foundational protocols: Open Agent Schema Framework (OASF) and Agent Connect Protocol (ACP). These two protocols play a significant role in creating the composability that enables adaptable, flexible functioning at scale that agents need to succeed. Let’s take a closer look at each:

1. Open Agent Schema Framework

OASF is a standardized metadata format describing what an agent does, what it needs, and how it can be trusted. It includes:

• Capabilities (e.g., can summarize documents, extract entities)
• Credentials (e.g., OAuth scopes, tenant identifiers)
• Lifecycle info (availability, versions, deployment mode)
• Reputation signals (performance scores, audit trails)

Standardizing the format for this metadata enables registry-based discovery, where agents can be published, searched, and composed, much like components in a composable commerce stack.

2. Agent Connect Protocol

ACP is a secure, real-time messaging layer that allows agents to authenticate and establish mutual trust, share execution context and constraints, and coordinate, delegate, or terminate tasks. It supports modern transport formats (gRPC, WebSocket, SSE) and is designed to work across infrastructure boundaries: public, private, and on-prem.

Together, these foundational protocols make agents composable by design, not just by intention.

Composability isn’t an optional design choice for agent systems, with enterprises already converging on a few unavoidable realities:

1. Agents must reuse tools, context, and memory. No organization can afford bespoke wiring for every workflow. Shared schemas and execution histories ensure agents can inherit and extend capabilities rather than rebuild them.
2. Multi-agent ecosystems are inevitable. Most enterprises will deploy many agents across domains, often built on different frameworks and models. A discovery and trust layer allows these agents to collaborate productively, even when they come from diverse origins.
3. Agents are coupled to existing enterprise systems. Whether it’s order management, CRM, or fulfillment, agents rarely operate in isolation. Composability ensures they can integrate into legacy environments without fragile point-to-point connectors.

It’s easy to see why infrastructure concepts like evaluations, guardrails, and reputation systems belong as horizontal layers rather than siloed features. AGNTCY takes this principle forward by offering a shared substrate for interoperability. In doing so, it sets the stage for thriving ecosystems— within a single enterprise, across markets, and eventually across the broader Internet of Agents.

The conversation about agent interoperability doesn’t start and end with AGNTCY. Other important protocols are already emerging to solve adjacent challenges. MCP focuses on how agents access and use external tools, while A2A defines how tasks can be delegated and exchanged between agents.

What AGNTCY brings to the table is different. It provides the connective tissue— the discovery, identity, and messaging layers that make it possible for those other protocols to work together at scale. Think of MCP as defining what an agent can do, A2A as describing how agents collaborate, and AGNTCY as the network and trust layer that underpins both.

For teams building modular systems, composable agents and AGNTCY unlock new design possibilities:

• Composable agent marketplaces: agents become deployable components with metadata, not just brittle code endpoints
• Dynamic orchestration: agents can be swapped, upgraded, or reconfigured mid-flow using metadata, not just code
• Federated ecosystems: partners can expose secure, auditable agents without sharing internal infrastructure
• Observability baked in: agents gain runtime transparency through shared protocols, not afterthought instrumentation

It’s the same playbook composable systems have followed in commerce, CMS, and data: decouple the core, standardize the interface, and build for change.

AGNTCY is still early-stage, but adoption is growing and the pattern is familiar. Protocols win when they reduce friction at scale.

The existence and growth of AGNTCY also prompts business leaders and decision makers to start asking the right questions: Are your agents discoverable and introspectable? Can they trust and delegate to unknown peers securely? Is your orchestration logic resilient to change, or bound to one agent model? Do you want to compose agents from different vendors, or just control your own?

If you're aiming for a modular, cross-functional agent layer in your stack, AGNTCY is a signal worth watching.

AGNTCY offers more than just plumbing. It offers a pattern where agents are treated not as endpoints, but as composable services; where metadata replaces manual integration; and where collaboration between agents becomes a matter of shared standards, not shared infrastructure.

For leaders designing the future of AI systems, that’s the kind of interoperability worth investing in.