The 1Z0-1157-26 costs $0 to sit and $0 to retake. The real cost is the time you spend studying the wrong things, and with six distinct tool ecosystems crammed into 40 questions and no published per-domain weights, any domain you skip is a gift to the fail column. These free practice questions are the fastest way to find out where you actually stand before you open the MyLearn portal.
Free 1Z0-1157-26 Practice Questions
Three free practice sources exist for this exam right now, today, without a credit card. CertCompanion offers 30 free 1Z0-1157-26 questions drawn from its 400-question bank, with detailed explanations per question and full six-domain coverage. Free third-party practice exams host 100+ questions with no registration required, covering all six modules. Oracle's own practice exam lives inside the MyLearn learning path and is repeatable at no cost, community experience consistently marks it as the single closest signal to what the real exam asks.
Why all three matter: Oracle publishes no percentage weight for any domain. You cannot tactically skip Domain 3 to focus on Domain 4, and the 65% pass mark, 26 correct out of 40, leaves little room to absorb a full domain gap.
10 free 1Z0-1157-26 practice questions
These are real questions from our 400-question 1Z0-1157-26 bank - the same pool the full practice exam draws from, not simplified samples. Answer before you expand the explanation.
Question 1
A network-remediation agent has a valid goal and may use approved diagnostics. Its first diagnostic returns a partial result that does not reveal whether a restart is safe. Which next step demonstrates autonomy without granting unrestricted authority? (Select one!)
- A. Ask the model to ignore its permitted-tool list during the incident
- B. End the workflow because one approved tool already completed
- C. Restart every affected device because the objective is restoration
- D. Reason about an allowed next action and continue only if needed
Show answer and explanation
Correct answer: D. Reason about an allowed next action and continue only if needed
Autonomy means reasoning about the next step and deciding whether further work is needed within task boundaries and permissions. Restarting all devices exceeds evidence-based action, ignoring permitted tools abandons authorization controls, and stopping after an incomplete result treats tool completion as task completion.
Question 2
For a production LangChain service, operators need to distinguish unreliable model choices from failures in otherwise correctly requested business functions. Which preparation most directly supports that goal? (Select one!)
- A. Force every invocation to use exactly one tool and one additional model call
- B. Replace all tool schemas with one generic free-form text instruction for the model
- C. Record lifecycle evidence separately for tool selection, parsing, execution, results, and continuation
- D. Remove function results from logs so model prompts remain shorter during investigation
Show answer and explanation
Correct answer: C. Record lifecycle evidence separately for tool selection, parsing, execution, results, and continuation
Production diagnosis benefits from observing distinct lifecycle stages, so teams can tell model-selection behavior apart from parsing, execution, result handling, or continuation behavior. Removing results hides evidence, fixed call counts violate conditional agent execution, and generic free-form instructions weaken the structured tool interface needed by the workflow.
Question 3
Vela Health has an internal dose-reference repository that its clinical-support agent should consult when composing answers. The repository is not temporary conversational context and does not itself run the agent. What should Vela supply? (Select one!)
- A. A knowledge base for the agent
- B. A sandboxed tool environment
- C. A hosted endpoint capability
- D. A platform memory capability
Show answer and explanation
Correct answer: A. A knowledge base for the agent
A knowledge base is application-owned content that the agent should use in its work. Platform memory is an operating capability rather than the supplied reference corpus, sandboxed tools concern controlled execution, and hosted endpoints make the agent available at runtime.
Question 4
A project manager asks whether a requirement for OCI Usage MCP access proves that the agent can calculate pricing, retrieve a particular metric, or bypass normal permissions. What should the architect conclude? (Select one!)
- A. The course establishes the OCI Usage MCP server as an example, but not its exact tools, metrics, pricing logic, or permissions.
- B. The OCI Usage MCP server guarantees every OCI cost metric through a fixed public calculation interface.
- C. The OCI Usage MCP server replaces OCI identity controls whenever an agent is connected.
- D. The OCI Usage MCP server is the database gateway that automatically runs validated SQL Reports.
Show answer and explanation
Correct answer: A. The course establishes the OCI Usage MCP server as an example, but not its exact tools, metrics, pricing logic, or permissions.
The supplied course material establishes inclusion of an OCI Usage MCP server but not its particular tools, returned metrics, pricing formulas, or authorization behavior. It cannot be treated as the managed database service, and standard MCP access never implies that identity or permission controls are bypassed.
Question 5
A developer is implementing a basic agent. The model has requested a function call, but the application has not yet run the requested capability. What should occur next in the execution flow? (Select one!)
- A. End the conversation because a tool call was parsed
- B. Assume the requested function completed and write the final answer
- C. Replace the tool schema with a knowledge-base document
- D. Execute the requested function, then provide its result for inspection
Show answer and explanation
Correct answer: D. Execute the requested function, then provide its result for inspection
Tool-call parsing interprets the requested call, while function execution is the separate next stage that actually runs the capability. The resulting output must then be inspected to guide continuation. Assuming completion, replacing a schema with information, or ending after parsing skips required parts of the loop.
Question 6
A product owner asks for the enterprise-agent component that lets teams define the agent's behavior, available actions, enterprise knowledge, and desired result while OCI handles the operating environment. Which purpose best fits OCI Enterprise AI Platform and OCI Enterprise AI Agents in this scenario? (Select one!)
- A. They replace the organization's business rules with a single universal objective for every enterprise agent
- B. They guarantee a specific authentication protocol for every hosted endpoint
- C. They support building enterprise agents and operating them with OCI-managed capabilities
- D. They convert all knowledge bases into OCI-owned operational memory
Show answer and explanation
Correct answer: C. They support building enterprise agents and operating them with OCI-managed capabilities
The enterprise offering supports building agents with application-supplied instructions, tools, knowledge bases, and outcomes, then operating them with OCI capabilities. It does not replace business rules, reclassify knowledge bases as memory, or establish endpoint authentication guarantees in the supplied foundational material.
Question 7
After an approved maintenance capability is added to an MCP server, an authorized client needs to use it. Which sequence correctly places discovery and authorization before invocation? (Select one!)
- A. Invoke the external function, then connect and negotiate agent ownership
- B. Connect to the server, discover authorized capabilities, then request one
- C. Embed the capability description, then retrieve it through a vector index
- D. Inspect results, select a model, discover access, then invoke
Show answer and explanation
Correct answer: B. Connect to the server, discover authorized capabilities, then request one
At the supported level, the client connects to an MCP server, discovers available capabilities subject to authorization, and then requests information or invokes a capability. Result inspection follows the request, while direct function execution and vector retrieval do not replace the MCP connection and authorized discovery sequence.
Question 8
A support assistant receives a goal to resolve a subscription issue. It checks account status through an API, reviews the returned result, and decides whether another action is needed. Which description best captures the LLM-based agent mental model? (Select one!)
- A. A tool server that executes every available API automatically
- B. A knowledge base that stores conversation history permanently
- C. An LLM that produces one answer from the initial prompt
- D. An LLM combined with tools and an iterative decision loop
Show answer and explanation
Correct answer: D. An LLM combined with tools and an iterative decision loop
An LLM-based agent combines an LLM, callable tools or APIs, and an iterative loop that inspects action results before deciding what to do next. A single prompt response lacks the loop and tool use, a tool server is not the reasoning agent, and a knowledge base supplies information rather than defining the agent model.
Question 9
A developer is building a training workflow that needs a course-demonstrated local utility for calculations, while a separate cloud-finance workflow needs a course-demonstrated integration related to OCI consumption. Which assignment is supported by the supplied material? (Select one!)
- A. Use the local math MCP server for the local utility and the OCI Usage MCP server for the OCI-related workflow.
- B. Use the OCI Usage MCP server for the local calculation utility and the local math MCP server for OCI consumption data.
- C. Use the OCI Usage MCP server for agent delegation and the local math MCP server for semantic document retrieval.
- D. Use the local math MCP server for managed database reports and the OCI Usage MCP server for local SQL troubleshooting.
Show answer and explanation
Correct answer: A. Use the local math MCP server for the local utility and the OCI Usage MCP server for the OCI-related workflow.
The local math MCP server is the documented local utility example, while the OCI Usage MCP server is the documented OCI-related example. The material does not establish database-report, SQL-troubleshooting, agent-delegation, or semantic-retrieval roles for either example, and it does not support exchanging their stated purposes.
Question 10
A research workflow must connect prompt setup, a document-retrieval step, an agent step with tools, and response formatting into one reusable flow. How does LCEL best contribute? (Select one!)
- A. It guarantees the model selects a tool
- B. It executes every external function directly
- C. It replaces tool schemas with raw prompts
- D. It composes workflow components into an agent flow
Show answer and explanation
Correct answer: D. It composes workflow components into an agent flow
LCEL supports composition: it connects workflow components such as prompt preparation, retrieval, agent steps, and formatting into a flow. External functions still run through their execution path, tool schemas remain needed for tool use, and model tool selection is not guaranteed by composition alone.
Why 40 Questions Span Six Completely Different Tool Stacks
Agentic AI is inherently a cross-framework exam: a single 60-minute, 40-question sitting covers LangChain internals, the Model Context Protocol, the OpenAI Agents SDK primitives, OCI Enterprise AI Agents platform concepts, Oracle AI Database-specific patterns, and foundational agent theory. That's not six chapters of one framework. That's six separate mental models, each with its own vocabulary, each testable on its own terms.
The official exam topics page (1Z0-1157-26) lists 23 numbered sub-objectives across six domains. No percentage weight appears next to any domain on that page. With six domains sharing 40 questions, an unprepared domain can cost several of the 14 questions you are allowed to miss.
The exam spans six distinct tool ecosystems in one sitting. Candidates who deep-dived one framework and assumed the rest would be thin coverage find the breadth catches them off guard.
What Each of the Six Domains Actually Tests
This is the official sub-objective checklist. Use it to verify that any practice material you're using actually covers the same scope as the real exam.
Domain 1, Introduction to AI Agents
Official objectives 1.1-1.4 require you to: differentiate AI agents from chatbots and rule-based workflows by autonomy, reasoning capability, and tool use; describe the LLM + Tools + Orchestration loop triad as the core agent architecture; describe Chain-of-Thought (CoT) and ReAct (Reasoning and Acting) as agent reasoning patterns; and explain safety techniques including layered guardrail design and loop stop conditions.
The LLM + Tools + Loop triad is the foundational mental model the entire exam assumes. Every other domain builds on it. Questions test precise distinctions, not "agents use LLMs" but "here is a scenario; is this a chatbot, a rule-based workflow, or an agent, and why."
Domain 2, LangChain for AI Agents
Official objectives 2.1-2.3 require you to: describe LangChain's core abstractions (chat models, prompts, tools, agents) and their roles; apply tools, prompts, and chains to construct an agent; and explain the reasoning and tool-execution flow, specifically how agent.invoke() triggers tool-schema building, function call parsing, tool execution, and loop-continuation decisions.
LangChain Expression Language (LCEL) is tested conceptually, not through live debugging. You need to understand the pipeline structure; you don't need to execute code.
Domain 3, Model Context Protocol (MCP) Fundamentals
Official objectives 3.1-3.5 require you to: explain MCP's role in standardising agent-tool integration across frameworks; define the five component roles (host, client, server, tool, resource/prompt); describe the JSON-RPC 2.0 message format; distinguish the two transport options (stdio for local processes, Streamable HTTP for remote services); and apply MCP components to an agentic workflow, including the OCI Usage MCP Server as a concrete example.
The host/client/server hierarchy is non-intuitive. These are not analogues to generic REST client/server roles. They require precise definition, and the distinction between them appears in both scenario and definitional questions.
Domain 4, OpenAI Responses API and Agents SDK
Official objectives 4.1-4.5 require you to: explain how the Responses API supports single-step and simpler agentic calls; describe the five Agents SDK primitives (Agent, Runner, Tool, Handoffs, Guardrails) and their functions; apply function calling to extend agent capabilities; explain multi-agent handoff patterns between specialised agents; and describe guardrail modes for input/output validation and tracing for observability.
The Responses API versus Agents SDK decision, when to use a single-step call versus a multi-agent loop, represents the central scenario pattern for this domain. Guardrails here are SDK-level constructs; they're different from LangChain tool schemas.
Domain 5, OCI Enterprise AI Agents
Official objectives 5.1-5.5 require you to: describe OCI Enterprise AI platform services and their role in the agent lifecycle; explain how OCI Enterprise AI Agents enables development, orchestration, and execution; describe building blocks (Responses API, tools, memory, vector stores); apply OCI Enterprise AI Agents to build a basic agent; and describe deployment options including hosted endpoints, scaling, sessions, and sandboxed tool execution.
The key testable concept here is the division of labour: the developer owns the agent's instructions, tools, and knowledge bases; OCI owns hosting, scaling, memory, and session management. That boundary is itself a question type.
Domain 6, Agentic AI for Oracle AI Database
Official objectives 6.1-6.7 require you to: explain Oracle AI Database's support for agentic workloads; describe the VECTOR data type and how embeddings enable similarity search; explain the Vector Search workflow (chunking, embedding generation, similarity search, retrieval); apply Vector Search to ground agent responses with enterprise data (RAG); explain the Oracle AI Database Private Agent Factory for no-code agent creation; explain Select AI for natural-language interaction with database data (NL-to-SQL); and explain the Oracle Autonomous AI Database MCP Server for protocol-based tool access.
Four patterns. Each needs a separate mental model. None has a direct analogue outside the Oracle ecosystem.
The Two Domains That Catch Engineers With No Oracle Background
Domain 5 and Domain 6 together account for roughly one-third of the exam. For candidates coming from general AI engineering, LangChain, the OpenAI API, standard MLOps, these two domains cover territory that simply does not exist in those stacks.
Domain 5 introduces OCI as a managed agent execution environment. The concepts aren't intellectually difficult, but they're Oracle-specific: how sessions persist state across multi-turn interactions, how sandboxed tool execution isolates agent-side code, how hosted endpoints scale on OCI infrastructure. Some candidates with strong general AI backgrounds report being tripped up here by the Oracle-platform specifics. The only cure is Oracle's own learning materials. If you're new to the Oracle certification track overall, the full range of Oracle's other certification exams gives useful context for where this credential sits in the broader OCI learning path.
Domain 6 is the harder of the two. The four Oracle AI Database agent patterns aren't variations on a theme, they serve fundamentally different use cases. Oracle AI Vector Search enables semantic retrieval for RAG grounding. The Private Agent Factory is no-code agent creation from database assets. Select AI translates natural language into SQL against structured data. The ADB MCP Server exposes the database as an MCP-compliant tool provider. A scenario question will describe an enterprise requirement and ask which pattern fits. Getting that right requires a separate mental model for each one, not a general "Oracle does AI" summary.
The failure reports are consistent on this point: candidates who felt confident after the first four domains and gave Domains 5 and 6 a light review before booking lost more questions than expected on Oracle-specific content.
The Cross-Framework Traps the Exam Is Designed Around
Is the exam hard? That depends entirely on your background. What's plainly true is that it's designed around cross-framework confusion, and three traps appear repeatedly in community practice materials.
Trap 1: LangChain vs OpenAI Agents SDK tool definitions
| Concept | LangChain | OpenAI Agents SDK |
|---|---|---|
| Define a tool | @tool decorator + schema | Tool primitive + function |
| Run the agent | agent.invoke() | Runner.run() |
| Multi-agent | Chains / sequential | Handoffs between specialised Agents |
These are not interchangeable. Questions will describe a workflow and ask which framework construct applies. Mixing up LangChain tool-call syntax with Agents SDK Handoffs is the most common wrong-answer pattern in practice materials.
Trap 2: MCP roles vs generic REST
A host in MCP is not a server in the HTTP sense. A client in MCP is not a consuming app in the REST sense. The MCP component hierarchy (host contains clients; clients connect to servers; servers expose tools, resources, and prompts) is specific to the protocol and doesn't map cleanly onto REST or gRPC mental models. Multiple experience threads note that candidates who skimmed the MCP module assuming "I know REST, I know this" dropped points on role-definition questions.
Trap 3: Responses API single-step vs Agents SDK multi-step
The Responses API handles single agentic calls and simpler workflows. The Agents SDK is the right choice when you need multi-step loops, persistent state, or handoffs between specialised agents. Scenario questions will describe a use case and ask which abstraction is appropriate. Getting this wrong consistently suggests you've treated them as synonyms. They aren't.
Build a comparison table. For each agent task, define a tool, run a multi-step loop, enforce guardrails, write down how LangChain and the OpenAI Agents SDK each approach it. That table is worth more study time than re-reading any single module.
How to Book and Sit the Exam Through Oracle MyLearn
The 1Z0-1157-26 launched in June 2026 as part of Oracle's 2026 OCI certification refresh. It costs $0 and is taken entirely through the Oracle MyLearn portal. There is no test-centre option; Pearson VUE and PSI are not involved.
"Unproctored" in Oracle's terminology means no human proctor watches your session in real time. It does not mean no monitoring. A browser extension for screen monitoring is required before the exam begins.
No appointment is needed. The exam starts on demand through the learning path page. Close other applications, disconnect VPN, and start when ready. At 90 seconds per question across 40 questions in 60 minutes, the pacing is brisk but not brutal; flag uncertain questions and return to them rather than stalling on a single item.
Your score report appears immediately on completion. The credential is valid for 2 years from the date of passing.
One sentence on career fit: the roles that consistently list 1Z0-1157-26 as a relevant credential include AI Engineer (Agentic AI), Cloud AI Developer on the Oracle stack, Solutions Architect for Oracle AI platform, and Oracle Fusion AI Consultant, all within Oracle-centric enterprise environments.
How Many Study Hours You Actually Need by Background
Don't skip the embedded skill checks inside the Oracle learning path modules. They're free, they're closely aligned with real exam question styles, and they're the best diagnostic signal between modules. Everyone who passed this quickly had one thing in common: they did the skill checks seriously, not as checkbox exercises.
| Background | Estimated hours | What drives the difference |
|---|---|---|
| AI/ML engineer with LLM API + Python experience | 6-10 hours | First four domains are familiar; focus time goes to Domains 5-6 (OCI platform, Oracle DB patterns) |
| Cloud developer with basic LLM familiarity | 10-15 hours | Full six-module learning path plus practice exam; all domains need coverage |
| New to agentic AI entirely | 15-25 hours | Complete the OCI AI Foundations course first (~6 hrs), then all six Agentic AI modules plus supplemental review of LangChain, MCP, and Oracle DB concepts |
The 80%+ threshold matters here. Community experience across Oracle's AI-track exams is that candidates scoring consistently above 80% on timed practice exams before booking pass the real exam comfortably. The official 65% pass mark (26/40 correct) can feel deceptively low until you're in the exam and realise that six domains of scenario questions don't leave many safe questions.
Frequently Asked Questions
Is the exam really free, and will it stay free?
As of November 2026, the 1Z0-1157-26 has been free since its launch in June 2026 as part of Oracle's 2026 certification refresh. The exam is taken through Oracle MyLearn at no cost. Oracle has not announced an end date for the free pricing, but it's worth confirming current pricing on the official Oracle education page before registering; Oracle's pricing policies can change.
What score do I need to pass?
The official pass mark is 65%, which means 26 correct answers out of 40. All 40 questions are scored; there is no unscored buffer published.
How long is the certificate valid?
The credential is valid for 2 years from the date of passing.
Is there a waiting period if I fail and want to retake?
Oracle has not published a retake waiting period for this exam. The retake cost is $0, the same as the first attempt.
Does 'unproctored' mean there's no webcam or monitoring?
No. Oracle uses "unproctored" to mean no live human proctor watches your session. A browser monitoring extension is still required and must be installed before the exam starts. Confirm your browser and extension are configured correctly before exam day using Oracle's system readiness check.
How long does it take for the certificate to appear after passing?
Your score report appears immediately on completion. Candidate reports document CertView load issues during high-volume certification periods; if the certificate hasn't appeared after some time, contacting Oracle support via CertView is the right next step.
What comes next after passing?
The direct professional-level follow-ons are the OCI Enterprise AI Professional (1Z0-1158-26) and the Agentic AI for Oracle AI Database Professional (1Z0-1159-26). Both go deeper into the Oracle platform concepts that Domains 5 and 6 introduce at the foundations level. For candidates newer to OCI infrastructure, the OCI Architect Associate credential complements the agentic AI track with infrastructure design knowledge relevant to production agent deployments.
Can I use the official Oracle practice exam to prepare?
Yes, and you should. The official practice exam inside the MyLearn learning path is free and repeatable. Community experience across Oracle's AI-track exams consistently rates it as the closest available match to the real exam's style and tone. It's not a substitute for broader practice coverage, but it's the highest-signal single readiness check available.
The 1Z0-1157-26 is a free exam with a 65% pass mark, it's not nothing, but on paper it looks approachable. What makes it genuinely demanding is the six-framework breadth in one 60-minute sitting and the Oracle-specific content in Domains 5 and 6 that general AI engineers won't have encountered before. Cover all six domains, give Domains 5 and 6 extra time if your background is outside the Oracle ecosystem, and score above 80% on timed practice sessions before you book. The CertCompanion question bank at /exams/oracle-agentic-ai-foundations-associate-1z0-1157-26 has 400 questions across all six domains with detailed explanations, start with the 30 free questions to see where your gaps actually are.
Ready to find your weak domains before exam day? Try 30 free 1Z0-1157-26 practice questions at CertCompanion.
Checked against official Oracle 1Z0-1157-26 exam documentation and current candidate reports. Last verified 2026-11-13.
Try 3 1Z0-1157-26 practice questions
Pick an answer to see if you are right and read the explanation.
1. At Meridian Labs, the same assistant must preserve useful working context while people interact with it in an ongoing engagement. Which OCI capability is most directly relevant? (Select one!)
Explanation
Memory is an OCI-provided operating capability relevant to retaining working context for an enterprise agent. Knowledge bases are organization-supplied sources, intended outcomes define goals, and application tools define available actions, so they do not represent the platform memory capability.
2. You need to ensure that an expense assistant can call the organization’s approved receipt-validation function during its work. What should the application owner supply? (Select one!)
Explanation
The application owner supplies the tools that define actions an agent may invoke, such as a receipt-validation function. Sessions, scaling, and hosted endpoints are OCI operating capabilities; they support the agent runtime but do not define its available business action.
3. An MCP_User can consume named financial SQL Reports, but a new investigation requires authorized ad hoc SQL through run-sql. Which role is designed to add that capability? (Select one!)
Explanation
MCP_Operator may additionally use run-sql for authorized ad hoc analysis, whereas MCP_User can be restricted to named SQL Reports. MCP_Administrator manages servers, toolsets, tools, and reports; publishing a report is not the documented role distinction for granting ad hoc run-sql access.