Building the Agentic Web

Happy Monday!

While everyone's talking about which AI model will win the reasoning race, a quieter revolution is taking shape in the background. Major tech companies are quietly building the plumbing for something much bigger: an "agentic web" where AI agents can talk to each other, share knowledge, and coordinate actions across platforms and organizations. This research is the foundation for what could become a multi-trillion dollar ecosystem.

The Meta Trend: From Silos to Networks

For the past two years, AI development has been characterized by walled gardens. Each company built their own AI agents that worked brilliantly within their ecosystem but couldn't talk to anything else. It's like having a powerful computer that can't connect to the internet.

But this landscape is changing rapidly. In 2025, we're seeing a coordinated push toward standardized protocols that allow AI agents to communicate across platforms, companies, and industries. This work may unlock entirely new forms of value creation.

The parallel to the early internet is striking. Just as HTTP and other web standards transformed isolated computer networks into the global web we know today, these emerging protocols could transform isolated AI agents into a connected ecosystem where the whole becomes exponentially more valuable than the sum of its parts.

Pattern Recognition: The Infrastructure Race Is On

1. Big Tech Is Betting on Open Standards: The most telling sign of this trend's importance is how quickly major competitors are aligning around common standards. Google introduced A2A with support from over 50 partners including Atlassian, Box, Cohere, and PayPal. Microsoft immediately embraced it, adding A2A support to Azure AI Foundry and Copilot Studio.

   Meanwhile, Anthropic's MCP is gaining rapid adoption, with OpenAI (a direct competitor) fully embracing the standard across ChatGPT and its API. This cross-industry collaboration suggests these companies see agent interoperability as more valuable than competitive moats.

2. Enterprise Demand Is Driving Adoption: Unlike many AI trends that start in consumer applications, agent interoperability is being driven by enterprise needs. Companies are discovering that their AI agents need to work together to deliver real value.

   A procurement agent needs to coordinate with budget management agents and vendor systems. Customer service agents must access data from multiple departments. Enterprise adoption of MCP is accelerating precisely because businesses can't afford to have AI agents that can't talk to each other.

3. Infrastructure Layers Are Emerging: We're seeing the development of distinct infrastructure layers for agent communication:

   • Communication Protocols: Standards like A2A and MCP for how agents exchange messages and context

   • Discovery Services: Agent directories that help AI systems find and connect with relevant agents

   • Orchestration Frameworks: Tools like Microsoft's Semantic Kernel and LangGraph for managing multi-agent workflows

   • Security and Governance: Identity systems like Microsoft's Entra Agent ID for managing agent permissions and accountability

   Each layer represents potential investment opportunities as the market figures out which approaches will dominate.

The Contrarian Take: The Real Winner Won't Be the AI Models

While everyone focuses on which company will build the smartest AI model, the real value might accrue to those building the infrastructure that connects AI agents. History suggests that in transformative technology shifts, the companies providing the "picks and shovels" often capture more value than those digging for gold.

Consider this: If agents can easily communicate across platforms, the specific underlying model becomes less important. What matters more is the quality of the agent's specialized capabilities and its ability to integrate with other systems. This could commoditize large language models while creating new value in specialized agent development, orchestration tools, and interoperability infrastructure.

The companies building the standards, protocols, and tools that enable the agentic web might capture more long-term value than those building individual AI agents. Further, companies who are building agents should take an integration- and systems-first approach. It matters less that the agent itself can do things and matters more how easily it can integrate into existing systems and communicate with the tech stack.

Practical Implications: Where's the Opportunity?

This trend toward agent interoperability creates several compelling investment and business opportunities:

1. Protocol and Standards Development: Companies contributing to and implementing key standards like MCP and A2A position themselves as essential infrastructure providers. Early adoption and deep integration of these protocols creates competitive advantages.

2. Agent Orchestration Platforms: Tools that help enterprises build, deploy, and manage multi-agent workflows will become critical. Think of these as the "operating systems" for agent ecosystems.

3. Specialized Agent Development: As agents become more interoperable, there's growing value in building highly specialized agents that excel at specific tasks rather than general-purpose assistants.

4. Security and Governance Tools: Agent identity management, access control, and audit systems will become essential as agents operate across organizational boundaries.

5. Agent Marketplaces and Discovery: Platforms that help organizations find, evaluate, and integrate third-party agents could become significant businesses as the ecosystem matures.

6. Cross-Platform Integration Services: Companies that specialize in helping enterprises connect agents across different platforms and protocols will see growing demand.

The most exciting aspect of agent interoperability is the potential for network effects. As more agents become capable of communicating with each other, the value of being part of the network increases exponentially. This could lead to rapid consolidation around winning standards and create powerful moats for early adopters.

We're likely still 12-18 months away from seeing the full potential of networked AI agents, but the infrastructure being built now will determine which companies capture the most value when the ecosystem reaches critical mass.

The parallel to cloud computing is instructive: the companies that built the early infrastructure (Amazon, Microsoft) captured enormous value as the market matured, often more than the companies building applications on top of that infrastructure.

In motion,
Justin Wright

1. Paper: A coding agent for scientific and algorithmic discovery (Google DeepMind)

2. SWE-1: Our First Frontier Models (Windsurf)

3. The age of AI agents and building the open agentic web (Microsoft)

4. Getting AI to write good SQL (Google Cloud)

5. Introducing Codex (OpenAI)