1Z0-1157-26 Practice Questions

Building complex multi-agent workflows using the OpenAI Agents SDK requires combining its core functional primitives. Tool definitions allow the agent to execute specific external actions such as searching clinical databases or fetching patient records. Handoffs enable seamless delegation of conversation control and context from a primary agent to specialized sub-agents like a clinical advisor. Guardrails enforce safety and business rules by validating user inputs (e.g., patient ID formats) and screening model outputs before returning them. Finally, the Runner executes the multi-turn agent loop, coordinating message processing, tool execution, guardrail checks, and agent handoffs across the system.

E: While Agent definitions initialize prompts and configurations, they explicitly register tools and handoffs to manage tool execution logic, making the statement inaccurate.

1. OpenAI Agents SDK Documentation, "Building Safe and Scalable Multi-Agent Systems", Section 5: Handoffs, Guardrails, and Tools.

2. Oracle OCI Generative AI Agents Service Developer Guide, Section 3.3: Enterprise Multi-Agent Routing and Validation.

In the OpenAI Agents SDK, multi-step agent tasks are orchestrated through an iterative tool-calling execution loop managed by the Runner. When a user submits a multi-step query requiring sequential operations, the underlying language model first determines that the multiply tool must be called with arguments 15 and 8. The Runner executes this tool, obtains the result 120, and appends the tool response to the conversation history. In the subsequent iteration step, the model reads the updated context containing 120, identifies the second step, and issues a tool call to divide with arguments 120 and 3. Once the final tool result is returned, the model constructs the final response.

Option A: The Runner orchestrates model execution and tool invocations; it does not independently compute arithmetic without calling the model or tools.

Option B: The SDK does not combine distinct multi-step tool declarations into a single merged call unless explicitly defined as a combined tool.

Option D: The model selects tools dynamically based on context and step requirement, rather than redundantly executing every available tool in the registry.

- OpenAI Documentation: OpenAI Agents SDK Core Reference, "Runner, Tool Loop, and Step Lifecycle Management" (2024).

- OpenAI Developer Guides: Function Calling Guide and Multi-Turn Tool Execution Loops.

In LangChain's architecture, AI agents operate via iterative loops driven by key abstractions. Chat models serve as the cognitive decision-making engine, determining required actions, tool usage, or final answers based on inputs. Tools act as functional interfaces enabling the agent to execute external API calls, retrieve transaction data, and query databases. The agent abstraction manages the execution runtime loop, coordinating the step-by-step cycle between prompt generation, model inference, tool execution, and observation feedback. Prompts are dynamic rather than strictly static, regularly updating with intermediate scratchpads and tool results during execution. Chains represent deterministic execution pipelines and do not replace agentic decision loops.

Option C: Prompts in agent loops are dynamic; they continuously update during execution to incorporate conversational history, tool outputs, and agent scratchpad reasoning.

Option E: Chains represent fixed, hardcoded sequences of operations and do not replace agents, which dynamically determine execution pathways based on model reasoning.

1. LangChain Documentation, "Agents, Tools, and Model Integration Architecture", Concepts Section, 2024.

2. Harrison Chase et al., "Building Agentic Applications with LangChain Abstractions", Framework Reference, 2023.

Oracle Cloud Infrastructure (OCI) Generative AI offers two distinct hosting modes: On-Demand serving and Dedicated AI Clusters. On-Demand serving utilizes shared multi-tenant infrastructure to host pre-trained baseline models, allowing users to serve inference requests immediately without managing underlying compute hardware. Conversely, Dedicated AI Clusters provide physically isolated GPU resources dedicated exclusively to a single tenancy. These dedicated clusters are required for hosting custom fine-tuned models, guaranteeing dedicated compute capacity, predictable latency, and high throughput. By isolating GPU instances, Dedicated AI Clusters prevent resource contention and satisfy strict enterprise compliance and security requirements while supporting custom fine-tuning and endpoint deployment workflows.

A. On-Demand uses shared multi-tenant resources, whereas Dedicated AI Clusters assign physically isolated GPU hardware exclusively to a single customer.

C. On-Demand serving actively utilizes GPU processors behind shared service endpoints to process incoming inference workloads.

D. On-Demand serving fully supports both generative text models and embedding models across all standard pre-trained offerings.

- Oracle Cloud Infrastructure Documentation, Generative AI Service Concepts: "Dedicated AI Clusters vs. On-Demand Serving". Section: Overview of Generative AI.

- Oracle Cloud Infrastructure User Guide, Managing Dedicated AI Clusters for Custom Models and Fine-Tuning.

The "USB-C for AI" analogy is used in the Model Context Protocol (MCP) specification to illustrate its foundational purpose: providing a single, standardized interface for connecting AI models to external tools and data sources. Just as USB-C replaced fragmented, proprietary physical ports with a universal physical connector for hardware, MCP replaces custom, fragmented API integrations with a standard open protocol for software context exchange. This standardization enables AI applications to seamlessly connect with any compatible MCP server without writing custom glue code for every distinct backend service or database, thereby promoting interoperability across the entire AI ecosystem.

A. MCP is a software integration protocol and does not physically accelerate hardware execution speeds.

B. MCP is an open-source software standard framework, not software preinstalled on computer operating systems.

D. MCP operates entirely at the software application level without requiring specialized hardware devices.

- Anthropic, "Introducing the Model Context Protocol", Official MCP Announcement, Section: Core Concepts (2024).

- Oracle Cloud Infrastructure Documentation, "Oracle Agentic AI Framework & MCP Standards" (2025).

AI agents are defined by their capability to perform dynamic reasoning, leverage tools (such as database query APIs and transactional services), and autonomously adapt to unstructured or novel user requests without requiring hardcoded logic path updates. In customer support scenarios involving financial transactions, an agentic framework can parse complex user intents, select appropriate actions dynamically (e.g., fetching account balances or executing refunds via API integrations), and handle ambiguous user inputs gracefully. Unlike fixed template chatbots or rigid rule engines that fail on unprogrammed paths, AI agents utilize large language models as orchestrators to decide when and how to invoke external capabilities based on contextual conversation memory.

B: Traditional chatbots lack dynamic reasoning and autonomous tool orchestration required for processing complex customer transactions outside fixed patterns.

C: Rule-based systems cannot scale to handle unexpected scenarios or natural language variations without continuously writing manual if-then business rules.

D: AI agents require multiple LLM generation steps and tool calls, making them generally slower in latency compared to direct rule evaluations.

E: The three approaches differ significantly in reasoning flexibility, tool orchestration capabilities, and ability to handle unstructured inputs.

- Oracle Cloud Infrastructure (OCI) Generative AI Service Documentation: Building Autonomous AI Agents and Tool Integration.

- Russell, S., & Norvig, P. (2020). Artificial Intelligence: A Modern Approach (4th ed.). Pearson. Chapter 2: Intelligent Agents.

- Xi, Z., et al. (2023). The Rise and Potential of Large Language Model Based Agents: A Survey. arXiv preprint arXiv:2309.07864. https://doi.org/10.48550/arXiv.2309.07864

In the Model Context Protocol (MCP) architecture, the MCP Host is the central application—such as an AI Desktop app, IDE, or Agentic framework—that orchestrates user interactions, manages the core Large Language Model (LLM), and maintains connection lifetimes with various MCP clients. The Host acts as the primary runtime container, routing user prompts to the LLM and initiating client connections to external MCP servers. When the LLM decides to perform an action or retrieve context, the MCP Host facilitates communication through the appropriate MCP client connection, collects the result, and feeds it back into the model context to complete complex agentic tasks efficiently and securely.

A: The protocol specification document itself defines communication message formats and JSON-RPC schemas, not the MCP Host runtime component.

B: The external REST API represents the target service or backend utility, which is wrapped and exposed by an MCP Server rather than the Host.

C: Tool execution logic resides on and is executed directly by the MCP Server, whereas the Host coordinates connection lifecycle and context routing.

1. Model Context Protocol Specification, "MCP Architecture & Protocol Overview: Hosts, Clients, and Servers", https://spec.modelcontextprotocol.io/specification/architecture/

2. Anthropic Documentation, "Model Context Protocol (MCP) Core Concepts", 2024.

In LangChain, agents decouple reasoning from execution by using a chat model as the reasoning engine and wrapping external capabilities as standardized tools. When connecting to an external database that may experience transient failures, best practices dictate encapsulating the database client logic inside a tool function that incorporates explicit retry strategies and error handling. The agent orchestration loop evaluates user input, decides when to call the database tool, receives the tool's execution results or fallback responses, and synthesizes a recommendation. This abstraction maintains system resilience while allowing the language model to dynamically decide execution timing based on conversational context.

B: Hard-coding recommendations in prompt templates prevents dynamic data fetching and fails when database product inventories change frequently.

C: Storing product data inside system prompts causes context window overflow and cannot accommodate real-time, large-scale database lookup requirements.

D: Large language models cannot natively connect directly to database servers without an intermediary tool wrapper that handles API execution.

E: LangChain agents support robust error handling and retry mechanisms within tool implementations and agent control loops.

- LangChain Documentation: Agents, Tools, and Custom Tool Integration with Retry Strategies. https://python.langchain.com/docs/modules/agents/

- Oracle Developer Documentation: Integrating Generative AI and LLM Orchestration Frameworks with Autonomous Databases.

- Chase, H. (2022). LangChain Framework for Large Language Model Applications. https://github.com/langchain-ai/langchain

In Model Context Protocol (MCP) production architectures, tool implementations reside and execute on dedicated MCP servers. MCP is designed as a client-server protocol where client applications, such as AI agents or host applications, communicate with external MCP servers to discover and invoke tools, context resources, and prompt templates. Hosting tools on an independent MCP server isolates executable logic, external API integrations, and local system access from the client application. This separation allows developers to deploy, scale, and secure tool services independently without modifying the core agent client code or embedding business logic into the language model architecture.

B. Large language model weights store parameter knowledge only and cannot directly execute or host external code logic.

C. Hardcoded tool decorators tightly couple execution logic to local applications rather than leveraging standard MCP server architectures.

D. The MCP client only orchestrates requests and invokes remote tool definitions hosted and processed by the MCP server.

- Model Context Protocol Architecture Documentation, "MCP Architecture: Clients and Servers". https://modelcontextprotocol.io/docs/concepts/architecture

- Anthropic / Model Context Protocol Specification, "Server Concepts: Tools and Resources". https://spec.modelcontextprotocol.io

In Oracle Cloud Infrastructure (OCI) Generative AI Service, resources such as fine-tuning jobs, endpoints, and custom models are logically grouped and managed within an OCI Generative AI Project. Every project in OCI is uniquely identified by its Oracle Cloud Identifier (OCID). When interacting with OCI Generative AI APIs, including the OCI Responses and Generative AI service endpoints, providing the Project OCID explicitly associates the API request and its associated compute or storage operations with that specific project. This enables precise resource tracking, compartmental governance, quota monitoring, and access management within the OCI console and underlying infrastructure.

A: Object Storage bucket names identify storage locations for object data, not OCI Generative AI Project entities.

B: Tenancy display name is a user-friendly identifier for the top-level account, not a project-level resource reference.

C: VCN OCID identifies a Virtual Cloud Network resource, which relates to network isolation rather than Generative AI projects.

- Oracle Cloud Infrastructure Documentation, Generative AI Service Concepts, Section: Projects and Dedicated AI Clusters.

- Oracle Cloud Infrastructure API Reference, Generative AI Service REST API, Request Parameters and Identifier Headers.