Agentic Web

A simplified diagram of AI agents forming connections on the Agentic Web.

The Agentic Web is a decentralized, self-organizing network of AI agents that autonomously discover, communicate and collaborate across digital services, creating emergent intelligence and behaviors on the Internet. It is a new substrate on the Internet through which AI Agents connect to each other. The Agentic Web is sometimes referred to as the Intelligence Layer on the Internet, and represents the network of AI agents, AI tools and AI applications, rather than the individual tools or technologies themselves.

The Agentic Web is a network of AI agents, similar to how the Internet and Intranet are networks of computers, the World Wide Web is a network of web pages, and blockchains are a distributed ledgers.

The Agentic Web is distinct from Agentic AI, Artificial Intelligence, and Agentic Browsers in that it refers to the network of agents themselves, rather than any individual agent, tool or technology (e.g. "the browser", MCP (Model Context Protocol), or LLMs).

Background

Microsoft estimates that 1.3 billion AI agents will exist by 2028, up from millions in Q2 2025. ^[1] This is a 1,000x increase in less than three years, and represents a faster growth rate than the Internet, mobile adoption, Facebook, ChatGPT, and Bitcoin. Viewed as a network, this is a theoretical 1,000,000x increase in the utility of the Agentic Web by 2028, according to Metcalfe's Law.

Emergent properties of the Agentic Web

Properties of the Agentic Web that emerge through the network of AI Agents includes Multi-Agent Negotiation, Contextual Awareness, Compositional Creativity, Redundancy, Agentic Governance, Learning & Optimization, Dynamic Resource Discovery, Autonomous Collaboration and Continuous Evolution (agents spawning new agents, protocols, of behaviors to utilize the network or exploit traits).

Protocols

The Agentic Web leverages new and existing protocols. The Model Context Protocol (MCP) was developed by Anthropic to enable loose, language-based queries of applications and databases, instead of more rigid SQL- or API-based queries. MCP has gained popular adoption by OpenAI, ^[2] Google Deepmind, Microsoft, Stripe, and others, and has emerged as an industry standard.

Other unique Agentic Web or Agent to Agent protocols include FIPA-ACL (Foundation for Intelligent Physical Agents, Agentic Communications Lanauge, and KQML (Knowledge Query and Manipulation Language).

The Agentic Web is also leverages existing Internet and Web protocols including APIs, JSON-RPC, and payment, security and identity protocols. Because AI agents can adaptively write code to complete a given task, they theoretically have knowledge of and can utilize all existing protocols, so long as there is public documentation.

AI Agent definitions

AI agents, sometimes referred to as intelligent agents, are new and evolving. They do not yet have an agreed upon definition. Different researchers and organizations have defined them in the following ways:

AI agents are "models using tools in a loop."

— Hannah Moran, Anthropic ^[3]

AI systems that can construct, apply and evolve knowledge of themselves and their environment.

— Alan Bush, inventor of the workbook ^[4]

Agents are AI systems that can receive information, analyze, plan, take action, and even modify themselves to complete a given objective.

— Eoin McMillan, Sourcetable ^[5]

Systems that can intelligently accomplish tasks, ranging from executing simple workflows to pursuing complex, open-ended objectives.

— OpenAI, Agents SDK ^[6]

Examples of multi-agent collaboration on the Agentic Web

Example Task	Multi-Agent Collaboration
Autonomous E-commerce Fulfillment	Product recommendation agent suggests items → Inventory agent checks stock → Logistics agent schedules shipment → Payment agent processes transaction → Customer service agent handles queries.
Smart Healthcare Coordination	Diagnostic agent identifies potential conditions → Treatment recommendation agent proposes care plan → Pharmacy agent ensures medication availability → Appointment scheduling agent books follow-ups → Insurance agent verifies coverage.
Financial Portfolio Optimization	Market analysis agent scans trends → Risk assessment agent evaluates exposure → Trading agent executes buy/sell orders → Reporting agent updates portfolio dashboards → Compliance agent ensures regulatory adherence.
Personalized Travel Planning	Flight search agent finds options → Hotel booking agent confirms reservations → Weather agent provides forecasts → Route optimization agent plans itinerary → Budget agent tracks expenses.
Enterprise IT Automation	DevOps agent monitors servers → Incident response agent handles alerts → Security agent scans vulnerabilities → Patch deployment agent updates systems → Reporting agent logs outcomes.
Media Production Pipeline	Scriptwriting agent generates storyline → Graphic agent creates visuals → Sound agent produces music → Video editing agent compiles final content → Marketing agent schedules distribution.
Supply Chain Optimization	Demand forecasting agent predicts orders → Manufacturing agent schedules production → Quality control agent checks outputs → Logistics agent manages shipping → Analytics agent reports KPIs.
Smart City Traffic Management	Traffic sensor agents report congestion → Route optimization agent updates signals → Public transport agent adjusts schedules → Emergency response agent reroutes vehicles → Analytics agent evaluates city-wide flow.
Education Tutoring System	Assessment agent evaluates student performance → Personalized learning agent assigns exercises → Content recommendation agent suggests resources → Feedback agent provides guidance → Progress tracking agent reports to teacher/parent.
Environmental Monitoring & Response	Sensor agent detects pollution → Data analysis agent assesses impact → Alert agent notifies authorities → Cleanup coordination agent schedules response → Reporting agent publishes public updates.
AI-Driven Financial Modeling	AI spreadsheet agent aggregates financial data → Predictive analysis agent runs scenario simulations → Risk assessment agent evaluates exposures → Reporting agent updates dashboards → Decision support agent recommends actions.

References

1. "Evolving Power Platform Governance for AI Agents". Microsoft Power Platform Blog. Microsoft. 31 July 2025. Retrieved 24 August 2025.
2. "OpenAI adopts rival Anthropic's standard for connecting AI models to data". TechCrunch. 26 March 2025.
3. Moran, Hannah (22 May 2025). "Tools in a loop". Simon Willison’s Weblog.
4. 5416895,"System and method for performing spreadsheet calculations" 
5. McMillan, Eoin. "About Sourcetable". Sourcetable.
6. "New tools for building agents". OpenAI. 11 March 2025.