A2A and the Silent Revolution A2A, MCP and T

Introduction: The Threshold Moment

Remember when you first held an iPhone in 2007? It wasn’t just a better phone—it was a complete reimagining of what a mobile device could be. That same transformative shift is happening right now in enterprise technology, only this time it’s happening with breathtaking speed.

In the span of mere months, we’ve witnessed the emergence of two protocols that will fundamentally alter how enterprise technology operates: Anthropic’s Model Context Protocol (MCP) in late 2024, quickly followed by Google’s Agent2Agent (A2A) in April 2025.

MCP provides a standardised way for AI agents to access enterprise data and systems—think of it as a “USB-C port for AI.” A2A creates a framework for AI agents to communicate with each other across platforms and vendors. Individually, they’re impressive technical achievements. Together, they represent something far more profound.

This isn’t normal enterprise software evolution. This is something faster, more organic, more alive. It reminds me of when Jobs pulled that first iPhone from his pocket—the moment when you realise what you’re seeing isn’t just a better version of what came before, but a complete reimagining of what’s possible.

We are witnessing the birth of collaborative software intelligence—systems that don’t just process information but work together toward common goals. And much like the early days of the smartphone revolution, those who grasp the implications first will redefine their industries.

I. The Speed is the Signal: A New Kind of Protocol Emergence

Traditional enterprise protocol development unfolds at a glacial pace. Think HTTP/2 (5+ years of development) or SOAP (years of consortium building). These protocols emerge through committee-driven deliberation, industry consultation, and multiple draft revisions before reaching general availability.

But MCP and A2A have compressed this timeline dramatically. MCP went from announcement to widespread adoption in months, not years. A2A emerged shortly after with support from over 50 technology partners including ServiceNow, Salesforce, and SAP. This acceleration isn’t accidental—it’s a feature of how AI development patterns are reshaping enterprise technology.

“The unprecedented speed of adoption signals something profound,” notes one industry analysis. “These protocols didn’t emerge through top-down planning but through collective recognition of a shared problem.”

What fascinates me most is how this speed reflects a fundamental shift in how we’re approaching software architecture. These protocols aren’t being imposed—they’re being embraced because they encode truths about how intelligence actually works: through collaboration, context-sharing, and specialization. The timeline compression isn’t just impressive—it’s diagnostic. It tells us we’ve hit on something fundamental about how systems should work together.

I’ve spent years building self-healing systems for cloud infrastructure, and I’ve observed firsthand that the most resilient architectures mirror how high-performing teams operate: specialized roles with clear interfaces and continuous feedback. A2A and MCP have captured this pattern and encoded it into infrastructure, and the market is responding with unprecedented enthusiasm.

How quickly could your organisation adapt if protocols that fundamentally change how software works together were to emerge in your industry with similar speed?

II. Understanding the Protocols: What They Actually Do

Model Context Protocol (MCP): The Universal Adapter

MCP solves a critical limitation of AI assistants: their isolation from enterprise data and systems. Think of it as a “USB-C port for AI”—a universal connector that standardises how AI models receive information from and interact with external systems.

At its core, MCP provides three primary interfaces:

• Tools: Functions the model can call to perform actions like searching data or updating records
• Resources: File-like data the model can read, such as documents or databases
• Prompts: Reusable templates that help standardise common interactions

MCP’s client-server architecture transforms what would be an M×N integration problem (M AI applications connecting to N tools, requiring M×N custom connectors) into an M+N problem. Tool providers build MCP servers, AI developers build MCP clients, and they connect through one standard protocol.

For ServiceNow customers, community developers have already created a ServiceNow MCP server that exposes the platform’s API through the MCP protocol, allowing any AI agent supporting MCP to directly query and manipulate ServiceNow data.

Agent2Agent (A2A): The Communication Bridge

While MCP connects agents to data, A2A connects agents to each other. It’s an open interoperability protocol designed to allow independent AI agents to communicate and coordinate with each other seamlessly, regardless of which vendor built them or which framework they use.

A2A introduces several key capabilities:

• Agent Cards: JSON descriptors that advertise what an agent can do, enabling dynamic matchmaking of tasks
• Task Lifecycle Management: Tracking of task status, results, and progress for long-running tasks
• Multimodal & UX Negotiation: Support for exchanging different content types, from text to images to UI elements
• Security Framework: Enterprise-grade authentication and authorization for agent interactions

When an agent needs assistance, it can discover and communicate with other agents via A2A, delegating tasks and receiving results through a secure, standardised channel.

The brilliance of these protocols lies in their complementary nature. A2A handles agent-to-agent dialogue, while MCP ensures each agent has access to the enterprise data and tools it needs. Together, they create a complete architecture for collaborative intelligence.

I’ve witnessed firsthand how siloed data can hamstring even the most sophisticated AI implementations. The typical enterprise has information scattered across dozens of systems—ServiceNow, Salesforce, Microsoft 365, custom databases. Before MCP, connecting an AI agent to all these sources required custom development for each integration. Now, as long as these systems expose MCP servers, any agent can access them through one consistent interface.

Similarly, without A2A, getting multiple specialised agents to work together was a developer’s nightmare of custom APIs and orchestration code. The beauty of A2A is that it standardises not just data exchange but task delegation and collaborative workflows—essentially encoding the patterns of effective teamwork into a protocol.

Consider your organisation’s most complex workflows—the ones that span multiple systems and require specialised expertise. How might they transform if all your digital systems could seamlessly collaborate like a high-performing human team?

III. From Solo Performance to Symphony: The Coordination Layer

The fundamental shift A2A enables isn’t technical—it’s conceptual. It moves us from systems that interact as API endpoints to systems that collaborate as colleagues.

Agent Cards are the perfect embodiment of this shift. They’re effectively “résumés” for AI agents, advertising capabilities in a standardised format. This social architecture allows for dynamic discovery—agents can find the right collaborator for a specific task without being pre-programmed with that knowledge.

This mirrors how high-performing human teams operate. They succeed not through perfect planning but through coordination, role clarity, and effective delegation. When I have a complex cloud operations issue, I don’t follow a predefined workflow—I assemble the right experts, explain the context, and coordinate their efforts. This human pattern of collaboration is precisely what A2A enables for AI agents.

Consider a practical example from the world of Eclipse AI’s agents, Umbra and Nebula. Umbra specialises in IT operations and incident remediation, while Nebula focuses on customer support and knowledge management. When a user reports that “the email service is slow,” this triggers a complex cross-domain workflow:

1. Nebula recognises this is an operational issue and uses A2A to delegate the investigation to Umbra
2. Umbra uses MCP to gather monitoring data and system metrics
3. Umbra diagnoses a stuck process causing the slowdown and proposes a restart
4. Nebula receives the diagnosis and resolution from Umbra via A2A
5. Nebula communicates the fix to the user in natural language

Without A2A, this kind of agent-to-agent coordination would require custom integration between every pair of agents an organisation might use. With A2A, any conforming agent can collaborate with any other, regardless of vendor or implementation details.

The power of this approach is how it handles complexity. Real-world tasks rarely map neatly to single-agent capabilities—they cross domains and require multiple types of expertise. A2A acknowledges this reality and provides the coordination layer needed to address it.

“What’s striking about A2A is how it encodes fundamental truths about how good teams actually work,” noted one analysis. “It’s not about workflow or APIs—it’s about goal-oriented collaboration.”

The cognitive shift required here is substantial. Most organisations still think about tasks in terms of workflows—sequential steps with predefined handoffs. A2A invites us to think in terms of capabilities and coordination—what expertise do we need to solve this problem, and how can those experts work together effectively?

How might your organisation’s approach to problem-solving change if you thought less about workflows and more about assembling the right team of experts (human or digital) for each challenge?

IV. The Integration Foundation: MCP and Enterprise Systems

While A2A creates a communication bridge between agents, MCP builds the integration foundation that connects those agents to enterprise systems. Without access to real-time, accurate data, even the most sophisticated AI agent is effectively blind.

The ServiceNow MCP server implementation exemplifies how this works in practice. Community developers have created an MCP server that exposes ServiceNow’s rich API through the MCP protocol. This allows any AI agent supporting MCP to:

• Search and read incident records, CMDB entries, and knowledge articles
• Create or update tickets and catalog items
• Run ServiceNow scripted actions
• Trigger workflows

In a real-world scenario, an agent like Umbra might use MCP to query relevant data about an email service issue—pulling incident data, checking server metrics from a monitoring tool via another MCP server, and retrieving configuration details from ServiceNow’s CMDB. All these MCP calls give the agent a comprehensive situational awareness that would be impossible without standardised data access.

The architectural elegance of MCP becomes apparent when you consider the alternative. Without it, each agent would need custom connectors for every system it accesses—a maintenance nightmare that scales poorly. With MCP, adding a new data source just means deploying another MCP server, with minimal changes to the agents themselves.

Security and governance are built into the MCP design. Connections can be scoped to read-only or specific actions as needed, and access controls ensure that agents only see data they’re authorized to access. This addresses a critical concern for enterprise adoption: how to give AI agents the data they need without compromising security or compliance.

What I find most powerful about MCP is how it transforms static AI into contextually aware intelligence. I’ve built systems where the difference between an AI that has real-time access to enterprise data versus one limited to its training data is night and day. MCP makes this contextual awareness standardised and accessible, rather than requiring custom development for each integration.

The approach reminds me of the revolution we saw with APIs and microservices—moving from monolithic architectures to composable, interconnected systems. MCP takes this principle and applies it to the AI layer, creating a standard way for intelligence to access and act upon enterprise data. It’s not just an integration standard; it’s a fundamental rethinking of how AI relates to enterprise systems.

Consider your organisation’s data landscape. How much more effective would your AI initiatives be if every agent could securely access all relevant information across your systems through a consistent interface?

V. The Cultural Shift: How Software Learns to Work Together

The emergence of A2A encodes a profound truth about how effective teams operate. It’s not just a technical specification—it’s a cultural template for collaborative intelligence.

Traditional software development emphasises building comprehensive, self-contained systems. The resulting applications might integrate with other systems through APIs, but they fundamentally operate as islands of functionality. A2A and MCP challenge this paradigm by prioritising specialisation and collaboration over comprehensive capability.

This mirrors how human organisations evolve. Early-stage companies often have generalists handling multiple functions. As they mature, they develop specialised teams with clear interfaces between them. The most successful organisations are those that master this transition—maintaining the benefits of specialisation while enabling effective cross-functional collaboration.

What we’re witnessing is a similar evolution in software architecture. Rather than building ever-larger models that attempt to handle everything, A2A encourages a shift to specialised agents that excel at specific tasks and collaborate to address complex problems.

My work with vertical agents has convinced me this is the only viable path forward. The task length barrier—where AI performance degrades exponentially as tasks grow in complexity—creates a fundamental ceiling for horizontal, general-purpose approaches. By contrast, specialised agents with clear boundaries consistently outperform their generalist counterparts on complex, domain-specific tasks.

This isn’t just a technical observation—it’s a strategic one. The organisations that thrive in this new paradigm will be those that master the art of composing specialised capabilities into cohesive solutions. They’ll build fewer monoliths and more networks of expertise. They’ll focus less on comprehensive planning and more on effective coordination.

ServiceNow’s existing AI agent strategy already emphasises specialised virtual agents for different functions. A2A enhances this approach by providing a standardised way for these agents to collaborate not just with each other but with agents from other vendors and platforms.

The most fascinating aspect of this cultural shift is how it accelerates itself through use. As agents collaborate, they learn not just about their specific domains but about the patterns of effective collaboration. This creates a multiplicative effect where the system as a whole becomes more capable than the sum of its parts—just like high-performing human teams.

How might your organisation’s approach to software development change if you thought less about building comprehensive systems and more about creating networks of specialised, collaborative capabilities?

VI. The Strategic Imperative: Why This Matters to Enterprise

The rapid adoption of A2A and MCP by enterprise software leaders signals a recognition that collaborative AI represents the next competitive frontier. Companies including ServiceNow, Salesforce, SAP, Atlassian, and others have already signed on as partners, recognising that these protocols address critical integration challenges in enterprise AI.

For organisations dependent on ServiceNow and similar platforms, these protocols offer concrete benefits across multiple domains:

IT Service Management (ITSM)

• Faster incident resolution: AI agents can collaborate to diagnose and resolve issues without human intervention
• More accurate knowledge retrieval: Agents can access real-time CMDB and knowledge base data via MCP
• Enhanced self-service: Virtual agents can delegate complex questions to specialist agents via A2A
• Reduced escalations: Multi-agent workflows can handle more complex requests before requiring human intervention

IT Operations Management (ITOM)

• Proactive issue prevention: Monitoring agents can collaborate with remediation agents via A2A
• Automated remediation: Agents can access operational data via MCP and execute fixes
• Cross-system coordination: Agents can work together to address issues spanning multiple platforms
• Continuous improvement: Learning agents can analyze patterns and update knowledge bases

Customer Service Management (CSM)

• Enhanced first-contact resolution: Customer-facing agents can collaborate with specialist agents
• Contextual customer interactions: Agents can access customer history and product data via MCP
• Cross-channel consistency: Agents can coordinate responses across different communication channels
• Knowledge gap identification: Agents can identify and fill gaps in service knowledge

The strategic advantage lies not just in automating individual tasks but in orchestrating intelligence across functions and systems. This breaks down the silos that typically hamper enterprise operations and creates a more responsive, adaptable organisation.

The economic case is compelling. Deloitte’s analysis of agentic AI solutions suggests at least a 30% improvement in resolution times and significant reductions in operational costs. By adopting these protocols, organisations can accelerate these benefits while avoiding the fragmentation and technical debt that typically accompany custom integration approaches.

What I find most striking about A2A and MCP is how they flip the traditional automation narrative. Instead of replacing humans, these protocols enable a new kind of human-AI partnership where agents handle routine coordination and information retrieval while humans focus on judgment, creativity, and oversight. This creates a multiplier effect where each human becomes more effective through collaboration with an ecosystem of specialised agents.

The early adopters of these protocols will have a significant advantage in building institutional knowledge about effective agent collaboration. As with any technological shift, there’s a learning curve to mastering this new paradigm—and those who start climbing it now will maintain a lead as the technology matures.

How prepared is your organisation to capitalize on the strategic advantages of collaborative AI? What high-value workflows in your enterprise would benefit most from orchestrated intelligence?

VII. Real-World Applications: Seeing the Protocols in Action

To make these concepts concrete, let’s examine how A2A and MCP transform real-world workflows through collaborative intelligence.

Eclipse AI’s Umbra and Nebula in ServiceNow

Eclipse AI has developed two specialised ServiceNow-connected agents: Umbra (focused on IT operations and incident remediation) and Nebula (focused on customer support and knowledge management).

Imagine an employee reporting that “the email service is slow” through a conversation with Nebula. This triggers a complex cross-domain workflow:

1. Cross-agent delegation via A2A: Nebula recognises that this complaint involves an IT operations issue. Using A2A, Nebula sends a task to Umbra: “Please investigate the email server latency for user X.” Nebula knows to delegate to Umbra because Umbra’s Agent Card indicates it specialises in diagnosing email system issues.
2. Data gathering via MCP: Upon receiving the request, Umbra uses the ServiceNow MCP server to query relevant data: incident history, server metrics (perhaps via another MCP server connected to a monitoring tool), and configuration details from ServiceNow’s CMDB. These MCP calls provide Umbra with comprehensive situational awareness.
3. Diagnosis and action: Based on the gathered data, Umbra diagnoses that a particular process is stuck and causing the slowdown. Before acting, Umbra sends an A2A message back to Nebula with a summary of the issue and proposed action. Umbra then uses MCP to execute the fix—perhaps invoking a ServiceNow workflow to restart the service.
4. User communication: With confirmation from Umbra, Nebula updates the employee: “Our systems detected an issue with a stuck process on the email server and have applied a fix. Your email performance should improve shortly. We’ve logged an incident for tracking: INC12345.”

This entire process could happen in minutes, with minimal or no human intervention, while maintaining full transparency through ServiceNow’s incident tracking.

The key insight here is how A2A and MCP enable specialisation without fragmentation. Nebula handles user interaction, Umbra handles technical remediation, and the protocols tie their efforts together seamlessly.

ServiceNow and Microsoft Copilot Integration

While not yet implemented, a hypothetical integration between ServiceNow and Microsoft Copilot illustrates the potential for cross-ecosystem collaboration.

Consider a change management workflow that spans ServiceNow for IT Change Management and Microsoft Teams for collaboration:

1. A ServiceNow Change Manager AI agent drafts a proposed change plan for a software deployment. It needs peer review on the risk assessment.
2. Through A2A, the Change Manager agent sends the draft plan to a “DevOps Copilot” agent (Microsoft Copilot integrated with GitHub/VS Code) that can analyze code changes and anticipate issues.
3. DevOps Copilot reviews the change, perhaps using MCP to pull recent commit history or test results. It responds via A2A with feedback about potential conflicts and suggests modifications to the deployment timing.
4. The Change Manager agent uses MCP to check ServiceNow’s schedule (CAB meetings, blackout windows) and incorporates DevOps Copilot’s feedback into a revised plan.
5. To finalise, the Change Manager agent uses A2A to request that a Teams Copilot agent draft an announcement message for stakeholders, which is then posted to the appropriate Teams channel.

This scenario demonstrates how A2A can bridge traditionally siloed ecosystems, creating a unified workflow across platforms. For organisations using both ServiceNow and Microsoft products (a common scenario), this reduces friction in cross-platform processes.

In my experience building agent architectures, this kind of cross-ecosystem collaboration represents the most significant untapped value in enterprise AI. Most organisations have attempted point solutions for specific functions, but the real transformation comes when these solutions can work together seamlessly.

I’ve seen how the Task Length Barrier limits what single agents can achieve. But when specialised agents collaborate—each operating within its domain of expertise—they can collectively handle workflows of arbitrary complexity. This mirrors how human teams tackle complex projects: not through individual heroics but through effective collaboration.

What cross-platform workflows in your organisation currently suffer from friction and manual handoffs? How might they be transformed if AI agents across different systems could collaborate seamlessly?

VIII. New Development Patterns: Building for Collaboration

The emergence of A2A and MCP doesn’t just change how agents interact—it transforms how we approach development. Here are the key shifts in mindset and practice:

1. Start Small, Start Specialised

Rather than building general-purpose agents that attempt to handle everything, focus on specialised agents with well-defined capabilities. The most successful implementations start with narrow, high-value use cases and expand from there.

I’ve consistently observed that vertical agents—focused on specific domains with deep contextual understanding—outperform horizontal approaches. For example, a finance-specialised agent might achieve 92% accuracy on domain-specific tasks compared to 47% for a general-purpose agent.

2. Design for Collaboration and Discovery

Implement Agent Cards that clearly communicate what your agent can do. Think of this as defining your agent’s professional résumé—what tasks it specialises in, what inputs it requires, and what outputs it provides.

Good Agent Cards follow these principles:

• They describe capabilities in task-oriented terms
• They specify input requirements and output formats
• They include any constraints or limitations
• They use standardised terminology for common functions

3. Build for Adaptation and Knowledge Sharing

Implement external memory architecture that allows agents to store and retrieve information outside their context windows. This creates institutional knowledge that persists beyond individual interactions.

As I’ve written previously: “For every problem solved, you store it as memory in a store outside the LLM… Don’t put it in a workflow, tell the agent it can store its memories in vector stores, graphs, cache tools.”

4. Architect with Interoperability in Mind

Design systems that can evolve as the protocols mature. This means:

• Implementing MCP servers for your critical data sources
• Adding A2A support to your existing agents
• Creating an agent registry or directory aligned to Agent Cards
• Planning for an internal “agent interoperability layer”

5. Focus on Coordination Patterns

The most valuable development effort isn’t building better individual agents—it’s creating effective patterns for agent coordination. This includes:

• Defining clear task delegation protocols
• Establishing feedback mechanisms between agents
• Creating oversight and approval workflows for critical actions
• Implementing conflict resolution when agents disagree

These shifts represent a fundamental change in the developer mindset—from building standalone applications to creating collaborative participants in an intelligent ecosystem.

The “build vs. buy” question also evolves in an A2A/MCP world. With standardised protocols, organisations can more easily mix proprietary, open-source, and custom agents to create composite solutions. This reduces vendor lock-in and allows for more flexible, best-of-breed approaches.

For ServiceNow developers specifically, this means:

• Building MCP servers that expose ServiceNow data and functions
• Adding A2A capabilities to existing ServiceNow virtual agents
• Creating integration points between ServiceNow agents and external agents
• Developing governance frameworks for agent interactions

I’ve seen the painful reality of building without these principles. I once consulted for an organisation that had invested heavily in custom agent integrations, only to find them brittle and unmaintainable as requirements evolved. Had they built on standardised protocols like A2A and MCP, they could have adapted much more easily to changing needs.

The radial approach to agent architecture—placing the agent at the center with connections (spokes) to various tools, knowledge bases, and other agents—becomes even more powerful when those connections follow standardised protocols. It creates systems that can evolve organically as capabilities and requirements change.

How might your development approach change if you focused less on building comprehensive agents and more on creating specialised components that collaborate effectively?

IX. New Challenges: The Failure Modes of Collaboration

As we embrace collaborative intelligence, we must also grapple with its unique challenges. Like any complex system, multi-agent architectures introduce failure modes that don’t exist in simpler, more isolated approaches.

The New Complexities

Coordination Overhead: When multiple agents collaborate, they spend time and tokens on coordination. This creates overhead that doesn’t exist in single-agent solutions.

Unexpected Loops: Agents can get caught in circular patterns, where Agent A asks Agent B for information, which prompts Agent B to ask Agent A, creating an infinite loop.

State Management: Tracking the status of multi-step, multi-agent processes becomes challenging, especially when agents operate asynchronously.

Emergent Behaviors: Multiple agents working together can produce unexpected interactions that weren’t designed or anticipated.

These challenges aren’t bugs—they’re inherent features of any collaborative environment, human or digital. High-performing human teams face similar issues of coordination, communication overhead, and occasional misalignment.

Observability Over Control

The key insight in managing these complexities is prioritizing observability over control. Rather than attempting to dictate every interaction, create systems that can monitor, understand, and adjust collaborative behaviors.

Effective approaches include:

• Agent Interaction Logging: Capturing the full context of inter-agent communications
• Task Tracing: Following the lifecycle of tasks as they move between agents
• Outcome Monitoring: Measuring the results of collaborative workflows against expected outcomes
• Feedback Mechanisms: Creating channels for human feedback on agent interactions

These observability practices allow for adaptive governance rather than rigid control—guiding the system toward better outcomes without stifling its collaborative potential.

Security and Governance Considerations

Collaborative AI introduces new security and governance challenges that must be addressed thoughtfully:

Authentication and Authorization: Implement strong authentication for agents and only allow approved agents to connect. Use role-based permissions to control what actions each agent can perform.

Human-in-the-Loop Checkpoints: Design critical workflows with approval gates where humans review and authorize agent-generated plans before execution.

Compliance and Audit: Ensure all agent actions are logged with appropriate detail for audit and compliance purposes. This is particularly important in regulated industries.

Security Testing: Simulate scenarios of malicious or errant agents to verify that your governance frameworks protect against potential abuse.

I’ve seen both extremes in enterprise AI governance: over-constraining agents until they become glorified chatbots with no ability to act, or granting excessive permissions without adequate oversight. The balanced approach is to implement what I call “Granular Trust Architecture”—contextual, compartmentalized, and dynamically adjustable trust levels rather than binary permissions.

In one healthcare implementation, we created a hierarchical approval system where agents could make routine decisions autonomously but required graduated levels of human approval for increasingly consequential actions. This balanced efficiency with appropriate controls, particularly for actions affecting patient data.

The most effective governance approach mirrors how organizations manage human teams: clear roles and responsibilities, transparent communication, appropriate oversight, and continuous improvement based on outcomes. The goal isn’t to eliminate all risk—it’s to manage risk while enabling the benefits of collaborative intelligence.

How would your governance frameworks need to evolve to accommodate collaborative AI while maintaining appropriate security and compliance controls?

X. Implementation Strategy: The Path Forward

For organisations looking to leverage A2A and MCP, a phased, pragmatic approach yields the best results. Here’s a strategic roadmap for implementation:

Phase 1: Pilot Projects with Clear Scope

Start with targeted pilots that demonstrate value while limiting complexity:

1. Select a well-defined use case where agents could collaborate to solve a common problem. Good candidates include:
   • Incident triage and initial diagnosis
   • Knowledge article creation and maintenance
   • Common service request fulfillment
2. Implement MCP for key data sources to provide agents with the context they need:
   • Deploy the ServiceNow MCP server for your instance
   • Connect to monitoring tools or other critical systems
   • Start with read-only access to minimize risk
3. Introduce A2A for specific agent interactions:
   • Enable A2A on a ServiceNow virtual agent
   • Create or connect with a complementary agent with specialized skills
   • Define clear protocols for task delegation

Phase 2: Creating an “Agent Interoperability Layer”

As pilot projects demonstrate value, build the architectural foundation for broader adoption:

1. Develop an internal agent registry or directory:
   • Catalog all available agents and their capabilities
   • Implement Agent Cards for dynamic discovery
   • Create governance rules for agent interactions
2. Expand MCP coverage to additional systems:
   • Move from read-only to action-oriented MCP usage
   • Connect to additional enterprise systems
   • Implement security and governance controls
3. Establish monitoring and observability:
   • Track agent interactions and outcomes
   • Identify patterns and bottlenecks
   • Refine based on performance data

Phase 3: Scaling and Organization-Wide Adoption

Once the foundation is established, expand to organization-wide implementation:

1. Develop governance frameworks for multi-agent workflows:
   • Define approval processes for critical actions
   • Establish audit and compliance controls
   • Create incident response plans for agent issues
2. Foster an ecosystem of specialized agents:
   • Build or acquire agents for specific domains
   • Create incentives for internal teams to develop agents
   • Establish standards for agent quality and security
3. Integrate with strategic partners:
   • Collaborate with technology partners to expand agent capabilities
   • Participate in industry initiatives around A2A and MCP
   • Share best practices and lessons learned

Measuring Success

Effective implementation requires clear metrics for success:

• Resolution time improvements: Measure reductions in mean time to resolution for incidents handled by collaborative agents
• Automation rates: Track the percentage of requests handled without human intervention
• User satisfaction: Gather feedback on agent interactions and outcomes
• Development efficiency: Measure the time saved by using standardized protocols versus custom integrations

I’ve found that the most successful implementations follow what I call the “Darwin Test”—evaluating whether systems are designed to evolve with rapidly changing capabilities. A2A and MCP inherently support this evolutionary approach by providing standardized interfaces that can accommodate new agents and data sources as they emerge.

The technical preparation for both protocols requires careful planning:

• For A2A: Implement the protocol specification, create Agent Cards for your agents, establish security controls, and set up monitoring
• For MCP: Deploy MCP servers for your key systems, configure access controls, define the resources and tools exposed, and test integration with target agents

By following this phased approach, organizations can realize the benefits of collaborative AI while managing the associated risks and complexities.

What pilot project would provide the most immediate value for your organization while helping you develop expertise with these new protocols?

XI. The A2A + MCP Synthesis: Social Computing Emerges

The true power of A2A and MCP emerges not when they’re used individually but when they work in concert. A2A connects agents to each other, while MCP connects them to enterprise data and tools. Together, they create a foundation for what we might call “social computing”—systems that collaborate rather than merely process.

In this synthesis:

• Agents use MCP to gather context and perform actions
• They use A2A to coordinate, delegate, and share insights
• The resulting collective intelligence exceeds what any individual agent could achieve

This architecture enables new possibilities:

Agent Marketplaces: Organizations can create internal marketplaces of specialized agents, each with clear capabilities advertised through Agent Cards. New agents can be added to the ecosystem without disrupting existing workflows.

Dynamic Workflows: Rather than following predefined steps, processes can adapt in real-time based on the specific context and available agent capabilities. This creates more resilient, flexible operations.

Collaborative Learning: Agents can share insights and patterns, creating a collective intelligence that evolves over time. The knowledge gained by one agent becomes available to others through structured sharing.

This approach connects deeply to what Andrej Karpathy has described as “AI as a vibe”—the pattern where transformer-based models create a distinct working style focused on iteration, feedback, and continuous improvement rather than rigid planning.

The next evolution of these protocols will likely include:

• Standardized ways for agents to share learned patterns
• Reputation systems for agent reliability and quality
• Capability-based discovery beyond simple descriptions
• Integration with human feedback mechanisms

For ServiceNow specifically, this synthesis could transform the platform from a workflow engine to an orchestrator of intelligent agents—coordinating collaborative problem-solving across the enterprise. The Now Platform’s existing strengths in workflow, integration, and governance make it an ideal foundation for this evolution.

In my work with enterprise AI, I’ve consistently observed that the most transformative implementations are those that break down silos between specialized systems. A medical device manufacturer we consulted for increased maintenance efficiency by 46% not by building a better maintenance AI, but by enabling collaboration between specialized agents for diagnostics, parts inventory, scheduling, and technical documentation. The A2A+MCP synthesis makes this kind of collaboration standardized and accessible rather than requiring custom integration for each use case.

The metaphor that best captures this shift is the transition from a group of musicians playing the same piece separately to an orchestra performing in concert. The difference isn’t in the individual capabilities but in the coordination and harmonization of diverse specialties.

How might your organization evolve if your digital systems could collaborate like an orchestra rather than performing as isolated soloists?

XII. Market Implications: Who Wins, Who Adapts, Who Falls Behind

The emergence of A2A and MCP will reshape the enterprise software landscape, creating winners, challengers, and laggards as the market adjusts to this new paradigm.

The Well-Positioned

Platform Providers with Open Strategies: Companies like ServiceNow that have both contributed to A2A development and embraced open standards are ideally positioned. They can leverage their existing platform strengths while participating in broader agent ecosystems.

AI-Native Startups with Specialized Focus: Vertical-focused AI companies like (the fictional) Eclipse AI can thrive by providing specialized agents that plug into enterprise platforms through standardized protocols, focusing on deep domain expertise rather than breadth.

Integration Specialists: Firms with expertise in complex system integration will find new opportunities helping enterprises create cohesive agent ecosystems across their technology landscape.

The Adapters

Traditional Enterprise Software Vendors: Companies with established enterprise applications will need to expose their functionality through MCP servers and add A2A capabilities to their existing agents to remain competitive.

Horizontal AI Providers: General-purpose AI companies will need to embrace agent specialization and interoperability or risk being outperformed by more focused solutions in specific domains.

Professional Services Firms: Consultancies will need to develop expertise in multi-agent architecture and governance to guide clients through this transition.

The Vulnerable

Closed Ecosystem Players: Vendors with proprietary agent systems that don’t embrace interoperability standards risk being isolated as the market moves toward open collaboration.

Point Solution Providers: Companies offering narrow AI capabilities without clear integration strategies may find themselves commoditized as standardized protocols make it easier to switch between providers.

Late Adopters: Organizations that delay embracing these protocols will face increasing technical debt and integration challenges as collaborative AI becomes the norm.

Looking at sector-specific implications:

Enterprise SaaS: The lines between applications will blur as agents work across traditional boundaries. Value will shift from user interfaces to effective orchestration and specialized intelligence.

ITSM/ITOM Platforms: ServiceNow

XII. Market Implications: Who Wins, Who Adapts, Who Falls Behind (continued)

ITSM/ITOM Platforms: ServiceNow and similar providers will evolve from workflow platforms to orchestrators of intelligent agents. Their value proposition will shift from process automation to intelligent coordination.

Specialized Services: Independent software vendors focusing on specific functions will need to expose their capabilities through A2A and MCP to remain relevant in an increasingly connected ecosystem.

Over the next 12-24 months, we’ll likely see:

1. Rapid protocol adoption by forward-thinking enterprises and vendors
2. Emergence of integration specialists focusing on agent orchestration
3. New marketplaces for specialized agents and MCP connectors
4. Standards evolution as real-world usage shapes protocol refinement
5. Governance frameworks specific to multi-agent architectures

ServiceNow’s position as both a contributor to A2A and a potential adopter of open standards puts it at a fascinating intersection. It can simultaneously shape protocol development while benefiting from the broader ecosystem that emerges.

From my vantage point building vertical agents for enterprise clients, this represents a profound strategic choice for technology leaders. Those who grasp the implications of collaborative AI—and act swiftly to embrace it—will gain significant advantages in operational efficiency, customer experience, and innovation capacity.

The pattern feels remarkably familiar to me. I remember watching organisations struggle with the cloud transition—some moved quickly to embrace new possibilities, while others tried to preserve existing approaches by merely lifting and shifting. The winners weren’t those with the biggest IT budgets or the most sophisticated legacy systems; they were the ones who recognized that cloud wasn’t just a new deployment model but a fundamentally different approach to computing.

Similarly, A2A and MCP aren’t just new integration standards—they’re the foundation for a new computing paradigm where collaboration, not processing power, is the critical capability.

What strategic position does your organisation want to occupy in this emerging landscape—pioneer, fast follower, or wait-and-see observer? And what are the implications of that choice for your competitive position?

XIII. Action Plan for Forward-Thinking Teams

For teams ready to embrace collaborative AI through A2A and MCP, here’s a concrete action plan:

1. Start Small, But Start Now

Next 30 Days:

• Identify one high-value use case where agent collaboration could deliver immediate benefits
• Deploy an MCP server for a key data source (e.g., ServiceNow)
• Experiment with connecting an AI agent to your MCP server

Success Indicators:

• Successful data retrieval via MCP
• Completion of a simple task using MCP-provided context
• Documentation of lessons learned and next steps

2. Think in Teams, Not Tools

Next 60 Days:

• Map your critical workflows to identify where specialized agents could collaborate
• Define clear boundaries and responsibilities for each agent type
• Create a simple A2A prototype with two collaborating agents

Success Indicators:

• Documented agent responsibility matrix
• Successful task delegation between two agents
• Measurable improvement in task completion quality or speed

3. Design for Discovery

Next 90 Days:

• Implement Agent Cards for your existing or planned agents
• Create an internal registry of available agents and their capabilities
• Establish governance rules for agent discovery and interaction

Success Indicators:

• Standardized Agent Card format adopted
• Functional agent discovery mechanism
• Clear policies for agent collaboration

4. Build for Adaptation

Next 6 Months:

• Implement external memory architecture for your agents
• Create feedback mechanisms to improve agent performance
• Develop metrics for measuring collaborative effectiveness

Success Indicators:

• Persistent knowledge storage and retrieval
• Measurable improvement in agent performance over time
• Dashboard for tracking multi-agent workflow effectiveness

5. Watch for Emerging Patterns

Ongoing:

• Participate in A2A and MCP community discussions
• Share learnings and best practices with peers
• Adapt your approach based on real-world performance

Success Indicators:

• Active engagement with protocol communities
• Regular reviews and adjustments to your strategy
• Documentation of patterns that work in your environment

For resource allocation, I recommend the following distribution:

• 20% on technical implementation (protocols, APIs, integration)
• 30% on agent development and specialization
• 25% on governance and security
• 25% on measurement, learning, and adaptation

The most common risk in early adoption is scope creep—attempting to tackle too many use cases or integrate too many systems at once. Mitigate this by:

1. Focusing on clear, measurable outcomes for each phase
2. Starting with read-only MCP usage before implementing actions
3. Implementing strong governance from the beginning
4. Creating clear feedback loops to identify and address issues early

Throughout my career implementing AI systems, I’ve consistently found that success depends less on the sophistication of the technology and more on the clarity of purpose and quality of execution. The most successful pilots are those with well-defined scope, clear success criteria, and strong executive sponsorship.

The organizations that thrive with A2A and MCP will be those that approach adoption as a learning journey rather than a one-time implementation. Each phase builds not just technical capabilities but organizational knowledge about how to effectively orchestrate collaborative intelligence.

What is the most strategic first step your organization could take to begin this journey? Who needs to be involved to ensure success?

Conclusion: The Real Shift - From Building to Orchestrating

As we reach the end of our exploration, the fundamental shift becomes clear: we’re moving from a world where value comes from building better software to one where it comes from orchestrating more effective collaboration.

Just as the iPhone redefined mobile by integrating previously separate functions into a unified experience, A2A and MCP are redefining enterprise software by enabling seamless collaboration between specialized intelligences. And just as the true power of smartphones emerged not from their technical specifications but from the ecosystem they created, the transformative potential of these protocols lies not in their technical details but in the collaborative patterns they enable.

The benefits for enterprises adopting A2A and MCP are substantial:

• Enhanced multi-agent collaboration: AI systems working together across vendors and platforms
• Seamless integration with enterprise data: Real-time access to the context needed for intelligent action
• Reduced integration costs: Standardized protocols replacing custom connectors
• Future-proofing against ecosystem fragmentation: Interoperability regardless of vendor choices

But perhaps the most profound benefit is the shift in mindset these protocols encourage—from thinking about AI as isolated tools to envisioning it as a collaborative ecosystem that augments human capabilities.

The irony is delicious: to build better software, we need to make it more collaborative—just like human teams. The machines are learning from us not just through training data but through the very architecture of how they work together. The most successful organizations won’t be those with the smartest individual agents but those that master the art of orchestrating collaborative intelligence.

As you consider your organization’s approach to these emerging standards, remember that the technical implementation is only part of the journey. The cultural and organizational shift—from thinking in workflows to thinking in capabilities, from control to coordination—is equally important.

The pioneers of collaborative AI will be those who recognize that the future of enterprise software isn’t about building bigger, more comprehensive systems—it’s about creating environments where specialized intelligence can flourish and collaborate effectively. It’s about shifting from monolithic thinking to ecosystem thinking, from rigid processes to adaptive collaboration.

In the words of Steve Jobs, “Great things in business are never done by one person; they’re done by a team of people.” A2A and MCP are bringing this human truth to our digital systems, and the results promise to be revolutionary.

How will you help your organisation make the shift from building to orchestrating?

Appendix: Technical References

For those looking to dive deeper into A2A and MCP, here are key resources:

A2A (Agent2Agent Protocol)

• Google Developers Blog: Announcing the Agent2Agent Protocol (A2A)
• A2A GitHub Repository - Open-source implementation and examples
• Agent2Agent Protocol Specification - Technical details and implementation guidance

MCP (Model Context Protocol)

• Anthropic: Introducing the Model Context Protocol
• MCP GitHub Repository - Reference implementations and documentation
• ServiceNow MCP Server - Community implementation for ServiceNow

Implementation Resources for ServiceNow

• ServiceNow Developer Portal: AI Integration Guide
• ServiceNow Community: AI Agent Forum
• Now Learning: AI Agent Development Courses

Community and Ecosystem Support

• A2A Community Forum - Discussion, best practices, and implementation help
• MCP Developer Community - Resources, examples, and support
• AI Agent Interoperability Alliance - Industry group advancing standards

About the Author

Chris Jones is Co-founder & CTO of Eclipse AI, where he leads development of enterprise-grade vertical AI agent solutions. With a multidisciplinary background spanning software development, system architecture, and AI strategy, Chris brings a uniquely practical perspective to the evolving landscape of enterprise AI.

His approach combines technical depth with strategic vision, helping organizations navigate the shift from traditional automation to intelligent agent orchestration. Chris is recognized for pioneering the “radial agent architecture” and developing frameworks for effective multi-agent collaboration in enterprise environments.

Connect with Chris on LinkedIn to continue the conversation about collaborative AI and enterprise transformation.