Traditional service-oriented designs impose stringent schemas that complicate updates whenever services amend internal data representations. As a result, clients must constantly adapt or face compatibility issues.
The advent of Agentic AI will re-shape the AI industry in more than one way. We will see the birth of specialized AI Agents built around distinct tasks, datasets, or domain capabilities. Rather than relying on a monolithic large language model for every query, users will increasingly interact with a proliferation of domain-targeted AI solutions—examples may include a ride coordinator AI, a food delivery AI, a lodging AI, or a property records AI. These specialized agents excel in their respective areas by leveraging contextualized datasets and processes that are uniquely aligned to their domain.
Inter-Agent Communication
Consider how these interactions evolve further: AI systems will increasingly communicate with one another without direct human intervention. A personal AI agent might coordinate with a service provider AI—such as a travel booking AI—which itself consults multiple upstream AI services (like an airline scheduling AI or a regional transit AI) to complete a user’s request. This network of interconnected AIs can generate new workflow patterns spanning diverse industry verticals, from logistics and transportation to healthcare and finance. For instance, imagine a personal AI assistant tasked with planning a multi-leg journey. It consults a travel booking AI for flights, which then queries a separate airline operations AI to confirm seat availability and real-time scheduling. Meanwhile, the travel booking AI also communicates with a hotel reservation AI to secure accommodations, relaying all confirmations back through the personal AI. In this scenario, neither the user nor a single monolithic platform orchestrates every detail. Instead, specialized AI services exchange information directly, dynamically negotiating and adjusting their inputs and outputs in real time. This multi-agent arrangement allows each AI to leverage its domain-specific expertise, resulting in a more robust, flexible, and efficient user experience across evolving market and industry landscapes.
The Need for a Standard
Facilitating meaningful communication among heterogeneous AI systems calls for a unifying convention or protocol. Such a standard must satisfy several interrelated criteria:
- Simplicity: The specification should be easily understood and implemented, minimizing onboarding barriers for new services.
- Ease of Implementation: AI developers and service providers must be able to integrate the protocol into existing infrastructures with minimal friction.
- Adaptability: As AI services evolve, the protocol must accommodate emerging capabilities or schema changes without requiring any rework on client or server sides.
- Future-Proofing: AI agents using this standard should not only support interactions with existing systems but also with new systems that are dynamically discovered. For example user may ask an AI agent to book a ticket in a theater that the AI agent has never talked to before. As long as the theater has it’s own AI service.
- Security: Ensuring that AI agents are properly authenticated and authorized to perform the requested actions on behalf of the user is paramount.
- Context Awareness: The protocol should permit session-based interactions, allowing requests and responses to carry forward context across multiple exchanges.
By articulating these guiding principles, the field moves closer to establishing a robust framework for AI-to-AI interactions, ensuring that specialized agents can interoperate effectively and adapt to the rapid pace of technological change.
AI2AI
AI2AI stands as an architectural paradigm that combines HTTP transport with natural language plain text interactions, facilitating a dynamically adaptive communication conduit between AI agents.
AI2AI leverages the existing, widely deployed HTTP protocol stack, maximizing interoperability within corporate and institutional firewalls. All requests are sent using HTTP POST requests to a single URL.
POST / HTTP/1.1
Host: ai.weather-info.com
Content-Type: text/plain
What is the sea temperature in Havana, Cuba today?HTTP/1.1 200 OK
The sea temperature is currently 22.5 Celsius degrees,
measured at 12:35 local time.Rather than imposing stringent, schema-bound payloads, messages are conveyed in human-readable text. This approach enables real-time negotiation of terminologies and clarifications.
Session Management
By associating consecutive requests with a shared session state via cookies, AI agents can maintain context across multiple interactions, thereby achieving a conversational mode of operation.
POST / HTTP/1.1
Host: ai.ride-coordinator.com
Content-Type: text/plain
Hello, I’d like to request a ride from 123 Main Street
to 456 Elm AvenueHTTP/1.1 200 OK
Set-Cookie: Session-Id=e52dbcb6-5b23-4313-b455-60c2f36fdf3e; HttpOnly
Certainly, I can help you with that. Can you specify the
preferred pickup time?POST / HTTP/1.1
Host: ai.ride-coordinator.com
Content-Type: text/plain
Cookie: Session-Id=e52dbcb6-5b23-4313-b455-60c2f36fdf3e
I would like the ride to arrive in 15 minutes.HTTP/1.1 200 OK
Traffic is moderate, and the estimated travel time is approximately 20 minutes.Security
AI2AI integrates seamlessly with OAuth2 for authentication and authorization. In most implementations, the Client Credentials Flow is recommended to ensure each AI service obtains a valid token from a centralized authorization server before initiating interactions. This mechanism provides robust identity verification, guarantees that only trusted AI entities can exchange data, and aligns with enterprise-level security protocols—such as token expiry, scope-based access controls, and audit trails—to protect sensitive information in complex, multi-agent ecosystems.
Not a Replacement for Traditional Communication
AI2AI serves as a powerful adjunct to rather than a complete substitute for REST, GraphQL, and traditional web interfaces. Although AI2AI excels at open-ended, conversation-driven communication, many mission-critical operations require the proven reliability and structured nature of existing protocols. For instance, verifying sensitive financial transactions or transmitting high-volume sensor data often benefits from the predictable schemas, low latency, and explicit error handling mechanisms that REST or GraphQL can offer. Likewise, certain user-facing tasks may still require conventional web UIs to ensure robust accessibility and clarity. In this sense, AI2AI complements these established methods, bridging the gap between strictly defined data exchange and more flexible, context-rich AI interactions.
Request for Comments
AI2AI addresses a pressing need at the intersection of AI innovation and service evolution. By eschewing rigid data formats in favor of flexible, plain-text negotiations and session-driven state, this model directly caters to the fluid nature of AI research and deployment. While it does not supplant all established API conventions, AI2AI introduces a forward-looking mechanism particularly well-suited to domains where perpetual refinement and conversational workflows are paramount. In tandem with the rise of agentic AI services, such as specialized on-demand or data-centric models, AI2AI provides a framework for future-proofing interactions in a world where many distinct, domain-focused agents collaboratively serve user needs.