Software's next evolution is orchestrated by interoperable agents, shifting from traditional development to Agentic Engineering.

Made with Rinto — analyse your own content free

Introduction

The paradigm shift from traditional software development to Agentic Engineering introduces the Factory Model.

Agent Core Architecture

The core architecture of an agent is defined as Model + Harness.

Industry Standards for Agentic Engineering

MCP, A2A, A2UI, AP2, and UCP are industry standards enabling machinery to safely interact with the world.

Why This Paper, Why Now

This paper focuses on how 'vibe coders' can use protocols to build virtual data and execution teams.

Who This Paper Is For

This paper is for software engineers, engineering managers, architects, and technical leaders embracing Agentic Engineering.

The Vibe Coder's View of MCP

This section covers consuming MCP servers efficiently, bypassing the NxM integration crisis, and debugging transport layers.

MCP Enterprise Architecture

A previous whitepaper outlined MCP enterprise architecture, including host-client-server topologies and security governance.

Vibe Coder Priority: Consumption

Vibe coders prioritize consumption over creation, hooking into existing registries for 'plug-and-play' agent superpowers.

MCP Server Onboarding Steps

Setting up your coding agent with MCP involves three simple steps: Discovery, Configuration, and Connection.

1Discovery→2Configuration→3Connection

Discovery

Vibe coders locate pre-built MCP servers from public, third-party remote, or internal registries.

Configuration

Before connecting, specify standard parameters for the server's scope and permissions, including API credentials and read/write permissions.

Connection

Once configured, the client establishes connection with the server and verifies it through a handshake request.

Bypassing the NxM Prototyping Problem

MCP is crucial for rapid prototyping by reducing integration complexity.

Traditional Integration Complexity

Traditional ad-hoc development requires custom integration code for every model-tool intersection.

Traditional vs. MCP Integration Effort

MCP significantly reduces the integration effort compared to traditional methods.

Integration Type	Effort
Traditional Integration	O(N x M)
MCP Interoperability	O(N + M)

Why MCP's Linear Scale Matters

MCP reduces integration complexity to a linear scale of O(N + M).

Debugging Issues with MCP Servers

When agents malfunction, debug transport pipes directly instead of blindly modifying system instructions.

Vibe Coder Toolkit: Best Practices for MCP Consumption

To maintain speed and stability, vibe coders should adhere to guidelines when consuming MCP servers.

Do's (Best Practices)

Follow these best practices for consuming MCP servers to ensure stability and efficiency.

1Audit Public Servers→2Use RAG for Tools→3Leverage Internal API Gateways→4Use MCP Inspector for Hallucinations→5Include HITL→6Log Tool Usage

Don'ts (Common Pitfalls)

Avoid these common pitfalls when consuming MCP servers to prevent issues.

1Don't Build Custom Wrappers→2Don't Use Unverified Public MCPs in Production→3Don't Hardcode Credentials→4Don't Connect to Production Data→5Don't Use for Data Updates→6Don't Provide Wide Access

Agent-to-Agent (A2A) Interoperability

A2A protocol is the foundational layer for resolving ecosystem fragmentation, enabling a globally interoperable virtual workforce.

Standardized Communication for Scaling

As AI systems transition to distributed networks of specialists, standardized communication becomes essential for scaling.

The Evolution of Agentic Architectures

Agentic architectures follow technology's pattern of shifting from low-level configuration to high-level orchestration.

Specialization as Scaling Mechanism

Specialization of components is a fundamental law of system design to overcome monolithic bottlenecks and scale.

Specialization Process

Specialization breaks black boxes into observable, maintainable stages, like data versioning and drift detection.

Internal Specialization Benefits

Logically partitioning a monolithic agent into distinct sub-agents allows modularization while maintaining low-latency execution.

Shift to Distributed Multi-Agent Architectures

Agentic architectures are maturing with a significant ecosystem shift toward distributed multi-agent architectures.

Prioritize 'Build vs. Buy'

To reach the next tier of scale, technical and business leaders must prioritize their focus through a modern 'Build vs. Buy' lens.

Leverage Official Specialist Agents

A better architectural strategy is to leverage official specialist agents maintained by domain expert teams for reliability.

Fragmentation Bottleneck

Orchestrating a virtual team of distributed AI specialists introduces a new bottleneck: fragmentation.

A2A Protocol for Standardization

The Agent-to-Agent (A2A) protocol is designed to standardize communication for agentic systems, resolving fragmentation.

Bounded vs. Unbounded Domains

Treating a specialist AI agent as a simple tool does not scale due to the nature of their problem-solving domains.

The GOTO Problem in Agentic Architecture

Forcing an agent with an unbounded domain into a standard tool wrapper introduces a 'GOTO statement' into the orchestrator.

Building the Virtual Workforce

A2A protocol and specialization solve technical bottlenecks and create new marketplaces for expertise.

Implementing A2A Protocols

The practical implementation of A2A architecture enables exposing AI agents and orchestrating remote agents.

Exposing A2A Agent Steps

Packaging an agent for the A2A ecosystem involves three core steps.

1Define Agent Card→2Implement Agent Executor→3Establish A2A Endpoint

Connecting Remote A2A Agents

An application acts as an Orchestrator, delegating tasks to specialized, remote A2A agents.

The Extensibility Layer: A2A for UI and Commerce

A2A protocol establishes a unified communication layer, enabling developers to build rich, transactional applications.

A2A Extensions for Functionality

A2A Extensions provide a standardized pattern for agents to advertise, negotiate, and execute optional, higher-order functionalities.

Foundational Frameworks as Extensions

Three foundational frameworks operate as native extensions atop the core A2A foundation.

Monetizing A2A Agents

Ensuring long-term commercial sustainability is crucial for agent work.

Agent-to-UI (A2UI) Interoperability

A2UI interoperability enables agents to generate complete, interactive user interfaces as outputs, resolving the communication gap.

The Communication Gap

Humans share insights visually and interactively, unlike agents that typically return raw JSON.

Generative UI & A2UI

Generative UI enables LLMs to dynamically create user interfaces at runtime based on user intent and context.

A2UI: A Secure Implementation

A2UI is a framework-agnostic standard for declaring UI intent, a Google open-source way for agents to describe interfaces.

The Basic Catalog and Bringing Your Own

A2UI ships a basic catalog of 18 ready-to-use components for prototyping and demos.

A2UI Basic Catalog Components

The A2UI basic catalog provides ready-to-use components across different categories.

Category	Components
Layout	Row, Column, List
Display	Text, Image, Icon, Divider
Containers	Card, Modal, Tabs
Media	Video, AudioPlayer
Interactive	Button, TextField, CheckBox, Slider, DateTimeInput, ChoicePicker

Production Frontends Use Own Catalog

Production frontends should bring their own catalog, mapping existing components to A2UI types.

Example A2UI Message (v0.9)

An A2UI message (v0.9 format) uses a flat adjacency list of components referenced by ID for easy generation and updates.

Generating A2UI: Two Patterns

There are two ways to produce A2UI, with the choice depending on where the layout decision resides.

Default Pattern: LLM Generates A2UI

The default pattern lets the LLM directly emit A2UI, with the model owning the layout and adapting to user intent.

Specialization Pattern: Tool Returns Fixed A2UI

The specialization pattern involves a tool returning a fixed A2UI structure, without LLM tokens spent on UI generation.

Data Flow in A2UI

Data values flow via `updateDataModel` messages, allowing clients to re-render on updates without re-sending the structure.

Query to Output Mapping

The choice of A2UI generation pattern depends on the query type and desired output.

Query	Return
"What's the average?"	Data (text)
"Compare these regions"	UI generated by the LLM (§4.4)
"Show me my dashboard"	UI built by a tool (this section)
API-to-API	Data (JSON)

Interactive Artifacts & The Canvas

Traditional chat interfaces are linear and static, but the canvas approach creates a persistent, editable workspace.

Canvas + A2UI

Combining canvas persistence with A2UI interactivity creates useful workflows for agent-user interaction.

A2UI Best Practices

Adhere to best practices for generating and handling A2UI to ensure efficient and robust interfaces.

Agents and Commerce (AP2 and UCP)

This section explores Agent Payments Protocol (AP2) and Universal Commerce Protocol (UCP) for autonomous procurement.

Autonomous Procurement Workflow

Vibe coders learn checkout, catalog, and order capabilities, and mandates with their agents for autonomous procurement.

Key Characteristics and Benefits of Protocols

Protocols offer typed schemas, security, and open source benefits, reducing integration debt and promoting vendor neutrality.

Transition to Action-Oriented Agents

Agents must transition from 'read' operations to executing 'actions' with real-world financial implications.

Ensuring Transaction Standards

Prioritizing commerce protocols and robust operational harnesses ensures agents adhere strictly to industry standards for transactions.

What is AP2 and UCP?

AP2 and UCP split the workload for autonomous commerce, demonstrated with a food delivery example.

Food Delivery Scenario

An AI Smart Assistant uses UCP and AP2 to get food for an apartment, splitting the workload.

UCP (Universal Commerce Protocol)

UCP acts like a universal translator, enabling AI to interact with merchant systems to build orders.

e.g.AI uses UCP to ask about store hours, vegetarian options, and build specific orders like 'One chicken burrito, no onions'.

AP2 (Agent Payments Protocol)

AP2 is an open, shared protocol providing a common language for secure, compliant transactions between agents and merchants.

Comparison for UCP and AP2

UCP and AP2 serve complementary roles in agentic commerce.

Metric	UCP	AP2
Function	Brain that decides what to buy, handles the menu, and puts the food in the cart.	Is the wallet that securely handles how to pay for it without you getting scammed.
Integration	Integrates with any business provider	Integrates with payment
Benefits	Unified integration: Shared language: Extensible architecture Security-first approach	Authorization & Auditability: Authenticity of Intent: Agent Error and Hallucination: Accountability

Conclusion

Adopting foundational standards eliminates technical debt, elevating developers to architects of an autonomous workforce.

▸ 8 Expand

APEX

Agent Tools and Interoperability

Software's next evolution is orchestrated by interoperable agents, shifting from traditional development to Agentic Engineering.

Made with Rinto — analyse your own content free

▸ 2 Expand

SECT

Introduction

The paradigm shift from traditional software development to Agentic Engineering introduces the Factory Model.

FRMW

Agent Core Architecture

The core architecture of an agent is defined as Model + Harness.

▸ 3 Expand

SUP

Industry Standards for Agentic Engineering

MCP, A2A, A2UI, AP2, and UCP are industry standards enabling machinery to safely interact with the world.

JUST

Need for Open Protocols

Without open protocols, agents are isolated 'custom machines' requiring fragile, bespoke wrappers for every API connection.

INSG

Standardized Interoperability Benefits

Adopting standardized interoperability layers transforms an agent's Harness into a modular, plug-and-play platform, reducing debugging time.

▸ 6 Expand

EVID

Key Protocols Defined

Several protocols serve distinct functions within agentic engineering to enable interoperability and functionality.

DATA

OpenResponses & Interactions API

These are 'Power Plugs,' modern API approaches supporting long-running tasks for LLM inference.

DATA

MCP (Model Context Protocol)

MCP acts as the 'USB-C,' connecting models to databases, filesystems, and web APIs within an agent's harness.

DATA

Skills as Playbooks

Skills are 'Playbooks,' simple markdown instructions and scripts or tools usable in a sandbox environment.

DATA

A2A (Agent-to-Agent)

A2A serves as the 'Factory Radio,' enabling specialized agents to negotiate, brainstorm, and delegate tasks.

DATA

A2UI (Agent-to-User Interface)

A2UI behaves like a 'Generative Display Window,' turning raw JSON outputs into safe, interactive visual components.

DATA

AP2 and UCP

AP2 and UCP act as the 'Global Supply Chain & Transaction Network,' enabling secure, autonomous commercial transactions.

▸ 2 Expand

SECT

Why This Paper, Why Now

This paper focuses on how 'vibe coders' can use protocols to build virtual data and execution teams.

JUST

Building Trust in Agent Development

Harnesses and protocols help build trust in agent development by adding structure to vibecoding.

INSG

Scaling Complex Goals

Standardized protocols transform isolated 'custom machines' into modular, interoperable platforms, enabling expansion.

▸ 2 Expand

SECT

Who This Paper Is For

This paper is for software engineers, engineering managers, architects, and technical leaders embracing Agentic Engineering.

JUST

Maintaining Fidelity and Reliability

Strict adherence to protocols is required to maintain fidelity and reliability of outcomes in Agentic Engineering.

▸ 5 Expand

TIP

Applied Tip for Coding Agents

A practical guide for 'vibe coders' to construct a virtual data and execution team.

STEP

Standard Guidance for Coding Agents

Use an AGENTS.MD file for standard guidance for coding agents.

STEP

Deep Thinking Before Code

Always think deeply before coding: state assumptions, surface tradeoffs, and clarify ambiguities.

STEP

Minimum Viable Code

Write only the absolute minimum code required to solve the immediate problem, avoiding speculative features.

STEP

Surgical Code Changes

When editing existing code, make highly surgical changes, restricting updates to necessary lines, preserving style.

STEP

Goal-Driven Execution

Approach every task via goal-driven execution with a clear plan, strong success criteria, and independent verification.

▸ 7 Expand

SECT

The Vibe Coder's View of MCP

This section covers consuming MCP servers efficiently, bypassing the NxM integration crisis, and debugging transport layers.

SUP

MCP Enterprise Architecture

A previous whitepaper outlined MCP enterprise architecture, including host-client-server topologies and security governance.

INSG

Vibe Coder Priority: Consumption

Vibe coders prioritize consumption over creation, hooking into existing registries for 'plug-and-play' agent superpowers.

▸ 3 Expand

FRMW

MCP Server Onboarding Steps

Setting up your coding agent with MCP involves three simple steps: Discovery, Configuration, and Connection.

1Discovery→2Configuration→3Connection

▸ 2 Expand

STEP

Discovery

Vibe coders locate pre-built MCP servers from public, third-party remote, or internal registries.

▸ 3 Expand

EVID

Primary MCP Server Sources

Pre-built MCP servers are found from Public MCP Registries, Third-Party Remote MCP Servers, and Internal Registries.

DATA

Public MCP Registries

Publicly available registries host hundreds of pre-built servers, excellent for rapid prototyping but unvetted.

DATA

Third-Party Remote MCP Servers

Vetted platforms expose hosted endpoints, offering managed pre-vetted MCP servers from official sources.

DATA

Internal Registries

Internal tools are exposed as MCP servers and cataloged in a secure registry within an organization.

▸ 2 Expand

TIP

Security Considerations for Discovery

Always consider security as a first priority when making decisions about MCP server discovery.

STEP

Seek Official Instructions

Look for official instructions where possible when choosing MCP server sources.

STEP

Avoid Credentials on Public Servers

Do not pass credentials if using public community servers; consider services like Model Armor to avoid security issues.

▸ 1 Expand

STEP

Configuration

Before connecting, specify standard parameters for the server's scope and permissions, including API credentials and read/write permissions.

▸ 4 Expand

TIP

Configuration Applied Tip

Adhere to best practices for MCP server configuration.

STEP

Check Prerequisites

Verify all necessary prerequisites are met before configuration.

STEP

Identify Scope and Access Criteria

Clearly define the scope and access criteria for the MCP server.

STEP

Include Specifications in Agent

Embed the server specifications directly into the coding agent.

STEP

Ensure Authentication

Implement robust authentication mechanisms for server access.

STEP

Connection

Once configured, the client establishes connection with the server and verifies it through a handshake request.

▸ 2 Expand

SECT

Bypassing the NxM Prototyping Problem

MCP is crucial for rapid prototyping by reducing integration complexity.

▸ 3 Expand

SUP

Traditional Integration Complexity

Traditional ad-hoc development requires custom integration code for every model-tool intersection.

DATA

Integration Complexity Formula

Integration complexity for N LLMs and M external tools is O(N x M).

EXMP

Example Integration Points

With 5 models and 10 tools, 50 bespoke integration points must be maintained.

INSG

Fragility of Ad-Hoc Integration

A single tool API change can break multiple parser loops without a protocol in place.

CMPR

Traditional vs. MCP Integration Effort

MCP significantly reduces the integration effort compared to traditional methods.

Integration Type	Effort
Traditional Integration	O(N x M)
MCP Interoperability	O(N + M)

▸ 3 Expand

SECT

Why MCP's Linear Scale Matters

MCP reduces integration complexity to a linear scale of O(N + M).

JUST

Benefits of Standardized Tool Definitions

Standardizing tool definitions allows direct connection to an agent's Harness without writing custom integration code.

▸ 2 Expand

EVID

MCP Standard Transports

Two standard transports facilitate MCP communication.

DATA

stdio (Standard Input/Output)

Used for local and prototyping efforts, treating the tool as a local subprocess.

DATA

SSE (Server-Sent Events) over HTTP

Connects to remote MCP endpoints over web protocols, streaming data in real-time.

INSG

Platform Adoption Without Custom Layers

Both transport options enable vibe coders to adopt the platform without multiple custom integration layers.

▸ 1 Expand

SECT

Debugging Issues with MCP Servers

When agents malfunction, debug transport pipes directly instead of blindly modifying system instructions.

▸ 2 Expand

EVID

Debugging Tools

Specific tools are available for debugging MCP server issues.

DATA

MCP Inspector

A native developer tool running a local web panel to query servers, view schemas, test inputs, and inspect JSON-RPC 2.0 packets.

DATA

Chrome DevTools

Ideal for web-based development environments, debugging SSE connections, tracing web streams, and checking server latencies.

▸ 2 Expand

SECT

Vibe Coder Toolkit: Best Practices for MCP Consumption

To maintain speed and stability, vibe coders should adhere to guidelines when consuming MCP servers.

▸ 6 Expand

CHKL

Do's (Best Practices)

Follow these best practices for consuming MCP servers to ensure stability and efficiency.

1Audit Public Servers→2Use RAG for Tools→3Leverage Internal API Gateways→4Use MCP Inspector for Hallucinations→5Include HITL→6Log Tool Usage

STEP

Audit Public Servers

Always review the code of publicly available, open-source MCP servers before connecting them to an agent.

STEP

Use RAG for Tools

Keep the agent's context window clean by dynamically loading tools only when needed and dropping them.

STEP

Leverage Internal API Gateways

Rely on internal tool registries to consume approved, governed data schemas, avoiding unvetted code.

STEP

Use MCP Inspector for Hallucinations

Use the MCP Inspector or Chrome DevTools to examine raw transport data when agents hallucinate tool calls.

STEP

Include HITL

Show tool inputs to the user before calling the server to prevent malicious or accidental data exfiltration.

STEP

Log Tool Usage

Log tool usage for audit purposes to maintain accountability and transparency.

▸ 6 Expand

CHKL

Don'ts (Common Pitfalls)

Avoid these common pitfalls when consuming MCP servers to prevent issues.

STEP

Don't Build Custom Wrappers

Resist writing custom REST API wrappers; search for existing MCP servers following the 'universal outlet' philosophy.

STEP

Don't Use Unverified Public MCPs in Production

Public, unverified MCP servers are for prototypes only, as they carry security and reliability risks in production.

STEP

Don't Hardcode Credentials

Never paste API keys or auth tokens directly into agents or scripts; rely on environment variables.

STEP

Don't Connect to Production Data

Use MCP servers with development projects and non-production or obfuscated data in a safe environment.

STEP

Don't Use for Data Updates

If connecting to real data, set the server to read-only mode, executing all queries as read-only.

STEP

Don't Provide Wide Access

Scope your MCP server to a specific project, limiting access only to necessary resources.

▸ 12 Expand

SECT

Agent-to-Agent (A2A) Interoperability

A2A protocol is the foundational layer for resolving ecosystem fragmentation, enabling a globally interoperable virtual workforce.

SUP

Standardized Communication for Scaling

As AI systems transition to distributed networks of specialists, standardized communication becomes essential for scaling.

▸ 2 Expand

SECT

The Evolution of Agentic Architectures

Agentic architectures follow technology's pattern of shifting from low-level configuration to high-level orchestration.

▸ 1 Expand

SUP

Parallel IT Trends

IT infrastructure saw a shift from manual configurations to 'Infrastructure as Code' and AutoML democratized AI.

EVID

AutoML Vision

Google's AutoML aimed to democratize AI by automating the 'drudge work' of neural network design.

▸ 2 Expand

SUP

Monolithic to Microservices Shift

Agentic architectures used to 'vibe code' applications are mirroring software history's transition from Monolithic to Microservices.

EVID

Single Agent Monoliths

Early vibe coding projects relied on 'Swiss Army knife' single agent monoliths, tightly coupling functions in massive codebases.

▸ 3 Expand

RISK

Monolithic Ceiling Limitations

Single agent monoliths face limitations that plagued early web applications.

DATA

Scaling Friction

Optimizing specific logic is difficult without confusing other logic, and many tools worsen decision-making.

DATA

Contextual Overload

Overarching system instructions, complex tool schemas, and conversation history quickly max out the model's working memory.

DATA

Single Point of Failure

A bug in one tool or instruction can cause the entire agent to hallucinate or crash, breaking future reasoning.

▸ 3 Expand

DCSN

Specialization as Scaling Mechanism

Specialization of components is a fundamental law of system design to overcome monolithic bottlenecks and scale.

EVID

Specialization Process

Specialization breaks black boxes into observable, maintainable stages, like data versioning and drift detection.

▸ 2 Expand

SUP

Internal Specialization Benefits

Logically partitioning a monolithic agent into distinct sub-agents allows modularization while maintaining low-latency execution.

JUST

Monolithic Sub-Agent Communication

These internal sub-agents share runtime and underlying memory, not communicating across network boundaries.

▸ 3 Expand

EVID

Technical Improvements from Modularity

Modularity addresses single-agent limitations through several key technical improvements.

DATA

Reduction of Search Space

Restricting a sub-agent's tools and skills drastically reduces tool-call errors and hallucinations.

DATA

Mitigation of Attention Dilution

Specialization allows the LLM to focus on a single domain prompt, leading to sharper reasoning.

DATA

Optimization of Contextual Load

Orchestrator routes tasks, enabling sub-agents to maintain a high signal-to-noise ratio in context windows.

▸ 1 Expand

SUP

Shift to Distributed Multi-Agent Architectures

Agentic architectures are maturing with a significant ecosystem shift toward distributed multi-agent architectures.

EVID

Industry Leaders' Adoption

Industry leaders like Google, Salesforce, ServiceNow, and Workday deploy domain-specific AI agents.

▸ 2 Expand

DCSN

Prioritize 'Build vs. Buy'

To reach the next tier of scale, technical and business leaders must prioritize their focus through a modern 'Build vs. Buy' lens.

RISK

Maintenance Tax of Custom Sub-agents

Constructing multi-agent systems with custom-built sub-agents for third-party platforms introduces significant Maintenance Tax.

JUST

Responsibility of Bespoke Development

Bespoke development means assuming full responsibility for updating prompt logic, tool definitions, and API schema changes.

▸ 1 Expand

DCSN

Leverage Official Specialist Agents

A better architectural strategy is to leverage official specialist agents maintained by domain expert teams for reliability.

JUST

Orchestrator Focus

Offloading specialist domain logic allows the Orchestrator and developers to focus on unique user value and core innovation.

▸ 2 Expand

RISK

Fragmentation Bottleneck

Orchestrating a virtual team of distributed AI specialists introduces a new bottleneck: fragmentation.

JUST

Diverse Agent Implementations

Specialist agents can be built with different technologies, languages, payload structures, conversational states, and transport layers.

JUST

Maintenance Tax of Custom Integrations

Writing custom integration code and error-correction loops for every specialist consumes entire projects.

▸ 2 Expand

FRMW

A2A Protocol for Standardization

The Agent-to-Agent (A2A) protocol is designed to standardize communication for agentic systems, resolving fragmentation.

JUST

Universal Communication Layer

A2A acts as the lingua franca for the AI ecosystem, abstracting away networking nuances, frameworks, languages, and payload disparities.

INSG

Standardizing the Virtual Workforce

A2A ensures the central Orchestrator can collaborate with any specialist agent, standardizing the virtual workforce like HTTP standardized the web.

▸ 4 Expand

SECT

Bounded vs. Unbounded Domains

Treating a specialist AI agent as a simple tool does not scale due to the nature of their problem-solving domains.

SUP

Specialists vs. Tools

The nature of engagement with specialists and tools is fundamentally different; tools are passive, specialists are collaborative.

JUST

Tool Limitations

Standard tools/APIs operate on a 'fire-and-forget' mechanism, expecting a single, perfectly formatted request.

RISK

Real-world Software Environment Challenges

Real-world software environments contain ambiguous data structures, misleading requirements, and conflicting user preferences.

SUP

Agent's Unbounded Problem-Solving Space

An agent operates in an unbounded problem-solving space, requiring multi-turn clarification for complex workflows.

▸ 3 Expand

RISK

The GOTO Problem in Agentic Architecture

Forcing an agent with an unbounded domain into a standard tool wrapper introduces a 'GOTO statement' into the orchestrator.

JUST

Interrupted Agent States

Collaborative agents can hit interrupted states, request more information, or never return expected output if the user changes their mind.

DCSN

Need for Multi-Turn Protocol

A paradigm is needed to isolate messy, multi-turn state, allowing agents to pause, negotiate, and resume work.

INSG

A2A Resolves GOTO Problem

The A2A protocol fills this gap by isolating collaborative routing, keeping the tool layer (MCP) clean and structured.

▸ 2 Expand

SECT

Building the Virtual Workforce

A2A protocol and specialization solve technical bottlenecks and create new marketplaces for expertise.

▸ 1 Expand

FRMW

The Agent Card

The Agent Card is a standardized 'CV' for AI agents, facilitating discovery.

▸ 3 Expand

SUP

Agent Card Identity Components

The Agent Card provides a machine-readable identity outlining capabilities, security, compliance, and interaction schemas.

DATA

Capabilities

Describes what tasks the agent can perform.

DATA

Security & Compliance

Details data handling policies and permission requirements.

DATA

Interaction Schemas

Defines how other agents should communicate with it via the A2A protocol.

▸ 2 Expand

SECT

Registries: Public to Private

Agent Registries turn Agents into discoverable services, offering two primary paths for developers.

EVID

Public Registries (Marketplaces)

Public registries act like global talent agencies, allowing developers to list specialist agents for licensing.

EVID

Private Registries

Within an enterprise, private registries provide a secure, governed environment for automating internal workflows.

▸ 2 Expand

SECT

Implementing A2A Protocols

The practical implementation of A2A architecture enables exposing AI agents and orchestrating remote agents.

▸ 3 Expand

FRMW

Exposing A2A Agent Steps

Packaging an agent for the A2A ecosystem involves three core steps.

1Define Agent Card→2Implement Agent Executor→3Establish A2A Endpoint

STEP

Define Agent Card

Create a formal agent specification through the Agent Card.

STEP

Implement Agent Executor

Implement the Agent Executor (Translation Layer) to translate A2A requests/responses to underlying frameworks.

STEP

Establish A2A Endpoint

Expose the Executor as an A2A-compliant endpoint.

▸ 3 Expand

SECT

Connecting Remote A2A Agents

An application acts as an Orchestrator, delegating tasks to specialized, remote A2A agents.

JUST

Orchestrator's Cognitive Load

The Orchestrator's primary cognitive load is understanding user intent and managing the overarching workflow.

SUP

Remote Agent Communication

Remote A2A agents are autonomous, domain-bound contractors communicating over the A2A protocol.

▸ 2 Expand

EVID

Connecting Remote Agents Patterns

Connecting to remote agents follows two architectural patterns: direct point-to-point integration or agent registry.

EXMP

Direct Instantiation via Hardcoded Endpoint

Python code shows direct instantiation for specific vendor integrations or private agents using a hardcoded endpoint.

EXMP

Indirect Instantiation via Agent Registry

Python code shows indirect instantiation using an Agent Registry to discover agents and handle authentication.

▸ 3 Expand

SECT

The Extensibility Layer: A2A for UI and Commerce

A2A protocol establishes a unified communication layer, enabling developers to build rich, transactional applications.

SUP

A2A Extensions for Functionality

A2A Extensions provide a standardized pattern for agents to advertise, negotiate, and execute optional, higher-order functionalities.

▸ 3 Expand

EVID

Foundational Frameworks as Extensions

Three foundational frameworks operate as native extensions atop the core A2A foundation.

DATA

Agent-to-User Interface (A2UI)

A2UI is designed to generate dynamic, stateful user experiences.

DATA

Universal Commerce Protocol (UCP)

UCP is engineered to facilitate secure, autonomous agentic commerce.

DATA

Agent Payments Protocol (AP2)

AP2 lays the groundwork for trusted, verifiable agentic payments.

▸ 5 Expand

SECT

Monetizing A2A Agents

Ensuring long-term commercial sustainability is crucial for agent work.

INSG

Agent-as-a-Service (AaaS) Model

The A2A protocol naturally enables an Agent-as-a-Service (AaaS) model, similar to SaaS.

EXMP

Google Cloud Marketplace Monetization

The Google Cloud Marketplace can be utilized as a monetization engine for listing A2A agents.

JUST

Marketplace Handles Billing

Marketplace infrastructure automatically handles complex billing, supporting hybrid pricing like 'Flat fee with usage' models.

SUP

Machine-to-Machine Microtransactions

The A2A protocol supports permissionless, machine-to-machine microtransactions for developers avoiding user accounts.

▸ 1 Expand

FRMW

x402 (L402) Standard Implementation

A2A Extensions framework [A2AEXT] allows an agent's server to implement the x402 standard for pay-per-call endpoints.

JUST

Automated Billing Process

The server intercepts unpaid requests, returns HTTP 402 Payment Required with an invoice, and the calling agent executes payment.

▸ 9 Expand

SECT

Agent-to-UI (A2UI) Interoperability

A2UI interoperability enables agents to generate complete, interactive user interfaces as outputs, resolving the communication gap.

▸ 1 Expand

SECT

The Communication Gap

Humans share insights visually and interactively, unlike agents that typically return raw JSON.

RISK

Raw JSON Breaks Workflow

Agents returning raw JSON requires manual chart building and state management, breaking the vibe coding flow.

▸ 3 Expand

SECT

Generative UI & A2UI

Generative UI enables LLMs to dynamically create user interfaces at runtime based on user intent and context.

JUST

On-Demand Interface Generation

Instead of hard-coding UI states, the model generates appropriate interactive layouts, cards, filters, and controls on demand.

SUP

Shift to Dynamic, Context-Aware UIs

Generative UI represents a shift from static, pre-built interfaces to dynamic, context-aware UIs.

RISK

Security Risks of Arbitrary UI Code

Running arbitrary UI code generated by an LLM poses security risks such as code injection, XSS attacks, and uncontrolled side effects.

▸ 2 Expand

FRMW

A2UI: A Secure Implementation

A2UI is a framework-agnostic standard for declaring UI intent, a Google open-source way for agents to describe interfaces.

JUST

Portable, Declarative Format

A2UI provides a portable, declarative format for streaming raw data or shipping arbitrary code, like sheet music for UI.

INSG

Separation of Concerns for Safety

A2UI's separation of concerns ensures safety: agents request components from a trusted catalog, and clients render them.

▸ 2 Expand

SECT

The Basic Catalog and Bringing Your Own

A2UI ships a basic catalog of 18 ready-to-use components for prototyping and demos.

CMPR

A2UI Basic Catalog Components

The A2UI basic catalog provides ready-to-use components across different categories.

Category	Components
Layout	Row, Column, List
Display	Text, Image, Icon, Divider
Containers	Card, Modal, Tabs
Media	Video, AudioPlayer
Interactive	Button, TextField, CheckBox, Slider, DateTimeInput, ChoicePicker

DCSN

Production Frontends Use Own Catalog

Production frontends should bring their own catalog, mapping existing components to A2UI types.

EVID

Example A2UI Message (v0.9)

An A2UI message (v0.9 format) uses a flat adjacency list of components referenced by ID for easy generation and updates.

▸ 4 Expand

SECT

Generating A2UI: Two Patterns

There are two ways to produce A2UI, with the choice depending on where the layout decision resides.

DATA

Default Pattern: LLM Generates A2UI

The default pattern lets the LLM directly emit A2UI, with the model owning the layout and adapting to user intent.

▸ 2 Expand

DATA

Specialization Pattern: Tool Returns Fixed A2UI

The specialization pattern involves a tool returning a fixed A2UI structure, without LLM tokens spent on UI generation.

JUST

Deterministic Layout Use Case

This pattern is suitable when the layout is deterministic from inputs, such as consistent sales dashboards or booking forms.

EXMP

Tool-as-Template Python Example

A Python function `get_sales_dashboard` builds an A2UI structure with data bindings, acting as a server-side template.

JUST

Data Flow in A2UI

Data values flow via `updateDataModel` messages, allowing clients to re-render on updates without re-sending the structure.

CMPR

Query to Output Mapping

The choice of A2UI generation pattern depends on the query type and desired output.

Query	Return
"What's the average?"	Data (text)
"Compare these regions"	UI generated by the LLM (§4.4)
"Show me my dashboard"	UI built by a tool (this section)
API-to-API	Data (JSON)

▸ 1 Expand

SECT

Interactive Artifacts & The Canvas

Traditional chat interfaces are linear and static, but the canvas approach creates a persistent, editable workspace.

INSG

Canvas as Living Document

The canvas acts as a living document where both the agent and user can modify sections, with changes reflected in real-time.

▸ 1 Expand

SECT

Canvas + A2UI

Combining canvas persistence with A2UI interactivity creates useful workflows for agent-user interaction.

JUST

UI as Communication Medium

The UI is not just rendered; it serves as a communication medium, with the agent observing user interactions and responding.

▸ 2 Expand

SECT

A2UI Best Practices

Adhere to best practices for generating and handling A2UI to ensure efficient and robust interfaces.

▸ 4 Expand

TIP

Let LLMs Generate A2UI

For LLM-generates-UI pattern, use the official `a2ui-agent-sdk` to avoid tedious hand-coding of A2UI JSON.

JUST

SDK Benefits

The SDK builds system prompts, ships JSON-Schema validators, and parses agent-emitted `<a2ui-json>` blocks.

EXMP

SDK Usage Example

Python code demonstrates loading the catalog, building the validator, and rendering a complete system prompt with schema and examples.

EXMP

Agent Execution and Error Handling

Python code shows running an agent to emit `<a2ui-json>`, parsing, validating, and retrying on schema errors.

DCSN

Production Error Handling

In production, wrap `create_ui()` in try/except and fall back to a text response on schema-validation failure.

▸ 2 Expand

TIP

Hybrid Output for Flexibility

Provide both data and UI in agent responses so consumers can choose their preferred output format.

EXMP

Hybrid Output JSON Schema

A JSON example shows hybrid output with separate 'data' and 'ui' fields, along with a 'ui_available' flag.

JUST

Client Behavior with Hybrid Output

API clients ignore the `ui` field and use `data`, while human-facing clients render the A2UI message.

▸ 5 Expand

SECT

Agents and Commerce (AP2 and UCP)

This section explores Agent Payments Protocol (AP2) and Universal Commerce Protocol (UCP) for autonomous procurement.

SUP

Autonomous Procurement Workflow

Vibe coders learn checkout, catalog, and order capabilities, and mandates with their agents for autonomous procurement.

SUP

Key Characteristics and Benefits of Protocols

Protocols offer typed schemas, security, and open source benefits, reducing integration debt and promoting vendor neutrality.

SUP

Transition to Action-Oriented Agents

Agents must transition from 'read' operations to executing 'actions' with real-world financial implications.

JUST

Ensuring Transaction Standards

Prioritizing commerce protocols and robust operational harnesses ensures agents adhere strictly to industry standards for transactions.

▸ 4 Expand

SECT

What is AP2 and UCP?

AP2 and UCP split the workload for autonomous commerce, demonstrated with a food delivery example.

EXMP

Food Delivery Scenario

An AI Smart Assistant uses UCP and AP2 to get food for an apartment, splitting the workload.

▸ 2 Expand

FRMW

UCP (Universal Commerce Protocol)

UCP acts like a universal translator, enabling AI to interact with merchant systems to build orders.

e.g.AI uses UCP to ask about store hours, vegetarian options, and build specific orders like 'One chicken burrito, no onions'.

JUST

Standardized Machine Language

Restaurants publish menus, hours, and customization options in a clean, standard machine language via UCP.

INSG

Easier Non-Human Interaction

UCP standardizes interactions between different providers, making it easier for non-humans to interact.

▸ 3 Expand

FRMW

AP2 (Agent Payments Protocol)

AP2 is an open, shared protocol providing a common language for secure, compliant transactions between agents and merchants.

SUP

AI Payment Within Rules

AP2 allows AI to pay for goods only within rules set by the user.

▸ 3 Expand

EVID

Key Features of AP2

AP2 includes guardrails, a handshake mechanism, and prevention of hidden fees for secure payments.

DATA

The Guardrails (The Mandate)

Users approve digital rules, such as spending limits (e.g., 'up to $25 at Taco Bell'), before AI executes payments.

DATA

The Handshake

AI shows a digital, encrypted 'promissory note' signed by the user, which the restaurant's bank verifies.

DATA

No Hidden Fees

AP2 instantly blocks charges that violate the signed-off rules, preventing unexpected fees.

INSG

AP2 as Secure Lockbox

AP2 is a secure lockbox allowing AI to pay for things using user money, preventing incorrect purchases.

CMPR

Comparison for UCP and AP2

UCP and AP2 serve complementary roles in agentic commerce.

Metric	UCP	AP2
Function	Brain that decides what to buy, handles the menu, and puts the food in the cart.	Is the wallet that securely handles how to pay for it without you getting scammed.
Integration	Integrates with any business provider	Integrates with payment
Benefits	Unified integration: Shared language: Extensible architecture Security-first approach	Authorization & Auditability: Authenticity of Intent: Agent Error and Hallucination: Accountability

▸ 1 Expand

SECT

Conclusion

Adopting foundational standards eliminates technical debt, elevating developers to architects of an autonomous workforce.

INSG

Unlocking New Economies of Scale

Standardized communication layers unlock new economies of scale, transforming how enterprise software is built, consumed, and monetized.