Introduction

Since publishing "The Agentic AI Stack for Enterprises" in February 2025, we've witnessed a breath-taking acceleration in the agentic AI landscape.

Just consider Google's announcement last week of their Agent Development Kit, allowing developers to build sophisticated and enterprise-compliant multi-agent systems saving the need for thousands of lines of custom code.

This rapid evolution makes our framework updates even more timely.

Organizations that implement these capabilities now aren't just preparing for the future – they're gaining immediate competitive advantages that may prove decisive in their industries.

Today, we want to share key enhancement areas we're incorporating into our framework based on recent research and industry developments. These enhancements will make the stack more future-proof and comprehensive as the agentic revolution accelerates faster than anyone predicted.

The original Agentic AI Stack for Enterprises, February 2025

1. The Agentic AI Accelerator: From Strategy to Implementation

In our original framework, we emphasized that a robust Agentic AI strategy is the essential foundation for any successful implementation. That's why we launched the Agentic AI Accelerator - a structured approach that guides organizations from initial strategic planning through to practical deployment of agentic capabilities.

The Accelerator helps enterprises break free from "AI pilot purgatory" by taking a business-first approach that identifies high-value operations, creates a clear path to production, and addresses not just technology but also process redesign and organizational change.

Through its five-phase methodology (Executive Illumination, Strategic Alignment, Rapid Prototyping, Scale-Ready Strategy, and Enterprise Activation), we're helping forward-thinking organizations deploy autonomous digital workers that can both operate independently and collaborate intelligently with human teams.

More details here: www.ai-risk.co/agentic-accelerator

2. Agent-to-Agent Economics (B2A)

Our original framework touched on "Business-to-AI Agent" (B2A) capabilities, but this area deserves more attention. We're expanding the concept to include protocols for economic transactions between agents themselves.

Imagine your enterprise's procurement agent autonomously negotiating with a vendor's pricing agent, exchanging data access privileges for preferred pricing, all within governance guardrails you define. We're developing practical guidelines for implementing these agent marketplaces within controlled enterprise environments.

Google’s agent-to-agent (‘A2A’) protocol, for example, flags a fundamental change in how Alphabet envisions businesses reaching their audiences in the future: a shift from users searching for a business and businesses paying for each click to a new world in which users employ their AI assistant to solve a problem by hiring a third party business agent which then pays a fee to Google.

3. Enhanced Discovery and Communication Protocols

Effective agent orchestration requires robust mechanisms for agents to find each other and communicate efficiently. We're incorporating standardized protocols for agent discovery, profiling, and metadata exchange into the framework.

This infrastructure will allow your agents to automatically locate the right specialist agents for specific tasks - whether those agents exist within your organization or are provided by trusted partners. Think of it as creating an internal search engine or ‘yellow pages’ for your AI workforce.

4. Strategic Negotiation and Collaboration

We're currently adding to the Stack frameworks for negotiation protocols and coalition formation. These mechanisms will enable more dynamic collaboration between agents with competing priorities or complementary capabilities.

For example, your customer service agents might form temporary coalitions with product specialists and logistics agents to resolve complex customer issues without predefined workflows. Getting this right requires balancing autonomy with appropriate oversight.

5. Cross-Enterprise Integration

While our original stack focused primarily on deployment within a single organization, enterprise reality increasingly involves collaboration across organizational boundaries. We're researching secure frameworks for agent interactions across different enterprises - including standards for managing agent reputation, trust verification, and secure credential management.

This area requires significant additional research, particularly around data governance, intellectual property management, and regulatory compliance across jurisdictions. We'll be publishing detailed guidelines on this specific area in Q3 2025.

6. Practical Implementation Roadmap

Many organizations struggle with where to begin. We're developing a phased adoption approach with a maturity model for agentic capabilities, moving from simple individual agents through multi-agent systems to fully collaborative agent teams.

This will include assessment tools for enterprises to evaluate their readiness for different aspects of the stack and identify the most valuable entry points based on their industry, current AI maturity, and strategic objectives.

7. The Emergence of Industry Standards: Model Context Protocol

We're particularly excited about Anthropic's Model Context Protocol (MCP), which is rapidly emerging as a potential standard for agent interoperability. Released initially in November 2024, MCP has gained remarkable traction with Google, OpenAI, and Microsoft all announcing support for the protocol.

MCP provides a universal, open standard for connecting AI systems with data sources, replacing fragmented integrations with a single protocol. This convergence around a single protocol by major AI companies signals a shift toward greater standardization in the agentic AI ecosystem.

For enterprises, MCP addresses several critical challenges:

Secure access to enterprise data sources
Reduced integration complexity through standardized protocols
Cross-platform compatibility for agent communication
Enhanced security features for managing sensitive data

Notwithstanding significant security concerns that will need to be ironed-out, we're integrating MCP into our stack as a foundational element for agent interoperability.

8. Distributed Architecture for Massive Scale: Agentic Mesh

As we look ahead to enterprises deploying not just dozens but potentially thousands or millions of agents, traditional approaches to agent implementation face fundamental scalability limitations. We're expanding the relevance of a new architectural framework to our stack called "Agentic Mesh."

This approach applies proven distributed systems principles to enable agent ecosystems to scale massively:

Distributed architectures that run agents across multiple computing nodes
Microservices and container designs for deployment flexibility
Standardized collaboration protocols for agent communication
Dynamic task planning and execution for true autonomy
Enterprise-grade monitoring and management capabilities

The Agentic Mesh complements our stack by providing a blueprint for how to implement agent systems at scale, addressing execution, development, and operational scalability challenges. (We've written about it before, here)

9. Specialized Agent Categories Emerging

We're tracking the rapid emergence of specialized agent categories that are showing particular promise for enterprise adoption:

Security Agents: For threat detection, vulnerability management, and incident response
Coding Agents: Specialized for software development, code review, and testing
Customer Service Agents: Designed for support ticket triage, resolution, and escalation
Workflow Automation Agents: Built for business process automation across multiple systems
Research Agents: Optimized for information gathering, analysis, and synthesis

This specialization indicates a maturing market where one-size-fits-all approaches are giving way to purpose-built solutions for specific enterprise needs.

We think these vertical agents and their ability to interact will lead to the emergence of a flourishing agentic AI ecosystem.

10. Accelerating Industry Momentum: Recent Major Announcements

The pace of innovation we're seeing is remarkable. In just the past few weeks, major AI companies have made significant announcements that validate our framework's direction:

Google introduced their Agent2Agent (A2A) Protocol at Cloud Next '25, enabling AI agents from different vendors to communicate with each other. They're also focusing on vertical integration from custom silicon (Ironwood TPUs) to application-level agents, optimizing the entire stack for inference rather than just training.
Microsoft has made autonomous agents generally available in Copilot Studio, with deep reasoning capabilities for handling complex business processes and enhanced agent flows for structured workflows. Also, new developer tools were announced for Azure Foundry.
Anthropic has positioned Claude 3.7 Sonnet specifically for agent scenarios and open-sourced the Model Context Protocol (MCP), which is rapidly becoming a standard for agent interoperability.
OpenAI unveiled their Responses API designed specifically for building AI agents, along with enhancements to their Agents SDK for orchestrating multiple agents.

These developments confirm that the agentic revolution is accelerating even faster than we predicted when we published our original framework in February. Organizations that delay implementation may find themselves at a significant competitive disadvantage as early adopters build and deploy increasingly sophisticated agent ecosystems.

What's Next?

Over the coming weeks and months, we'll release a series of detailed papers on each enhancement area, starting with our expanded view on agent-to-agent economics and discovery protocols. These will include specific implementation guidelines, vendor assessments, and case studies from early adopters.

The strategic implications are clear: organizations that begin building these capabilities now will develop significant competitive advantages as AI transitions from a tool-based paradigm to a true workforce transformation. The window for developing a thoughtful approach is narrowing rapidly as the technology accelerates.

What aspects of agentic AI are you most excited about implementing in your organization? I'd love to hear your thoughts in the comments below.

Please share this edition of AI Risk & Strategy with others and encourage them to subscribe, for free, at this link. You can also follow us on LinkedIn at AI Risk. And, to learn more about how our Agentic AI Accelerator can help your organization build and implement a robust agentic strategy, visit our website or contact us directly: www.ai-risk.co.

The ‘Agentic Stack’ for Enterprises