Framework For Agentic AI Systems Design

Various companies are consistently adopting Artificial Intelligence-driven systems through the use of AI Agents. However, due to the fact that this technology is still new and it changes rapidly, it has made a lot of companies experience some struggles in adopting it into their business processes, and developers find themselves lacking guidance on how to design and develop Agentic AI systems, which are both scalable and maintainable. In this article, we are going to share with you the factors that you need to consider when designing Agentic AI systems. We believe that the pipeline shared in this article will act as a standardized guideline for developers in designing maintainable and scalable AI Agents. Maintainable: The ability to preserve its functionality throughout the Agent Life CycleScalable: The ability to implement more features, formation of collection intelligence through Agent Collaboration, and capabilities through an efficient use of Agent Protocols.Define What The Agent DoesThe first step of the process is to clearly define what your agent does. This will help the developer to understand the AI Agent Touch Points and be able to identify detailed Interactions. This might be considered a trivial exercise; however, your entire design approach relies on this stage, so we consider this one of the most important stages of your agent design, as it will regulate which protocols you will have to use for your agent interactions. Protocols are described as standards regulating how AI agents interact. As of today, there are six core agent protocols, namely Model Context Protocol, Agent2Agent Protocol, Universal Commerce Protocol, Agent Payment Protocol, Agent To User Interface Protocol, and Agent-User Interaction Protocol.Agent ProtocolsThere's an observed widespread use of specialized Agents being deployed across industries, and the scattered distributions create a need for developers to establish ways in which these Agents can communicate with one another. We predict that we are going to experience a new form of service that relates to Agents, which we will call Agents-as-a-Service, where companies are going to allow people to have access to their AI Agents for a certain specialized task in return for a subscription or a fee. Protocols are just standards regulating how AI agents interact. We classify protocols in the following categories.1. Context-oriented protocolsModel Context Protocol: Almost every AI Agent we've come across has this protocol implemented. It regulates the interaction of the AI Agent with data from various sources, be it your database, files, etc. According to Google.com, it provides a single standard connection pattern for hundreds of servers, allowing agents to automatically discover tools and eliminating the need for custom integration code for every specific API. It provides a JSON-RPC client-server interface for secure tool invocation and typed data exchange.2. Inter-agent protocols Agent 2 Agent Protocol: This standardizes how agents discover and communicate with one another. It allows a primary agent to route queries to remote specialist agents that may be built on different frameworks or servers. It is useful for peer-to-peer task delegation using capability-based Agent Cards.3. Domain Specific Protocols Universal commerce protocol: This protocol standardizes the shopping lifecycle (such as discovering catalogs and checking out) into modular capabilities. It allows an agent to interact with multiple suppliers using a unified pattern rather than integrating unique flows for every vendorAgent Payments Protocol (AP2): This handles payment authorization by adding a layer of typed mandates and configurable guardrails to transactions. While UCP handles the "what" and "who" of an order, AP2 manages who approved the purchase and provides a secure audit trail.Agent Payment Protocol and Universal Commerce Protocol work together by dividing the responsibilities of a transaction. UCP manages the commerce logistics, while AP2 manages the authorization and security. Specifically, AP2 plugs into UCP as an extension, adding cryptographic proof of authorization directly into the checkout flow.4. Interfacing protocolsAgent-to-User Interface Protocol (A2UI): This allows agents to dynamically compose native UI layouts (like dashboards or forms) from a fixed catalog of safe component primitives. It separates the UI structure from the data, enabling the agent to present information visually without requiring custom frontend code for every new requirement.Agent-User Interaction Protocol (AG-UI): This protocol standardizes how an agent streams results to a frontend. It acts as middleware that translates agent events—such as incremental text streaming or tool calls into a standardized stream, eliminating the need for manual parsing boilerplate on the client side.Design The Interaction LayerAfter you have identified your AI Interactions and touch points, you have to select the protocols that have to be used to achieve the desired Agent Action. Always ask yourself which protocol is the best to achieve a particular task. For instance, if you find yourself in a situation in which a particular Agent Action requires the use of another Agent's capabilities, you might need to select the Agent-to-Agent protocol to facilitate the interaction of Agent A with Agent B.ConstraintsIt is very important to regulate the behaviour of the AI Agents beyond protocols and tools definitions. OpenAI's model specs provide applicable instructions that can constrain and govern the behaviour of AI-related applications. The question a developer needs to ask themselves is the one that relates to the safety, Privacy, and Security Concerns of your artificial intelligence agents.ConclusionThe following Agentic AI system design framework is a project that we are aiming to make a standardized industry framework for designing Agentic Systems. The document will be consistently updated and reviewed; comments are welcome from the community.Referenceshttps://developers.googleblog.com/developers-guide-to-ai-agent-protocols/  A Survey of AI Agent Protocols