50 QUESTIONS 90 MINUTES 68% PASSING SCORE $245 EXAM FEE 6 TOPIC AREAS Prepared by DBExam • www.dbexam.com Oracle • AI Database 26ai • 1Z0-1159-26 C E R T I F I C A T I O N G U I D E Embed It, Retrieve It, Act on It Inside the Agentic AI for Oracle AI Database Professional Certification (1Z0-1159-26) Six exam topic areas with their official weightings and all 14 objectives, laid out as a study checklist - plus the full exam fact sheet, who the credential is really built for, what AI engineers are paid in 2026, a five-step route from first read to booked seat, and ten real sample questions with a complete answer key. www.dbexam.com Oracle 1Z0-1159-26 Agentic AI for Oracle AI Database Certified Professional 1 S E C T I O N 0 1 Exam Overview This certification covers AI built inside Oracle AI Database 26ai: vector embeddings and similarity search in SQL, retrieval-augmented generation with PL/SQL packages, Select AI Agent teams, the Autonomous AI Database MCP Server, and Private Agent Factory. It does not test whether you can describe a vector - it tests whether you can pick the right distance metric, get a query to use its index, chunk and embed documents, and run an agent team to completion. Everything below comes from the official 1Z0-1159-26 exam page and DBExam's 1Z0-1159-26 exam syllabus page. Certification track Oracle Cloud Infrastructure Credential Agentic AI for Oracle AI Database Certified Professional Exam number and title 1Z0-1159-26 - Agentic AI for Oracle AI Database Professional Number of questions 50 Exam duration 90 minutes Question format Multiple choice Passing score 68% Exam fee USD 245 (pricing varies by country and currency) Delivery Oracle University, on the MyLearn platform Product version covered Oracle AI Database 26ai Recommended training Become an Agentic AI for Oracle AI Database Professional learning path Two details are worth pinning down before you book. First, despite the title, this is mostly a vector search exam : distance, indexing and RAG together carry 56% of the paper, and agents, MCP and Private Agent Factory the remaining 36% after the 8% foundations area. Second, the questions are written in SQL and PL/SQL - FETCH clauses, DBMS_VECTOR_CHAIN, DBMS_CLOUD_AI_AGENT - so reading code precisely matters. Purchase and scheduling run through Oracle University. www.dbexam.com Oracle 1Z0-1159-26 Agentic AI for Oracle AI Database Certified Professional 2 S E C T I O N 0 2 Who This Exam Is Built For Read the blueprint and four kinds of practitioner fall out of it. None of them is new to Oracle Database: the questions assume you write SQL and PL/SQL already and now need to bring embeddings, retrieval and agents into it. Database Developers AI Engineers Data Architects DBAs Oracle's recommended route is the Become an Agentic AI for Oracle AI Database Professional learning path on MyLearn. General LLM experience helps with the RAG and agent concepts, but most marks sit on database specifics - vector index types and their parameters, exact versus approximate fetch, and the Select AI and agent packages. The wider Oracle catalogue sits on the DBExam Oracle hub. S E C T I O N 0 3 The Six Topic Areas Oracle publishes percentage weightings for this exam, and they are unusually lopsided. Vector Distance and Indexing alone carries 32% - roughly sixteen questions from three objectives - and RAG another 24%. Select AI Agent design is 16%, the MCP Server 12%, and the foundations and Private Agent Factory areas 8% each. Plan your study time accordingly. Apply Vector Distance and Indexing 32% Build Retrieval-Augmented Generation Solutions 24% Design Select AI Agent Solutions 16% Configure Agentic Interfaces and Tool Access 12% Explain Oracle AI Vector Search Foundations 8% Govern Private Agent Factory Deployments 8% Official weightings as published on the 1Z0-1159-26 exam syllabus, sorted from heaviest to lightest. The full objective list follows, straight from the 1Z0-1159-26 syllabus. Read it as a checklist - if you cannot demonstrate a line in a real Oracle AI Database 26ai instance, that is a study item. 01. Explain Oracle AI Vector Search Foundations (8%) • Explain Oracle AI Vector Search concepts, vector embeddings, and semantic similarity search • Explain vector embedding models, ONNX model import, and in-database vector generation www.dbexam.com Oracle 1Z0-1159-26 Agentic AI for Oracle AI Database Certified Professional 3 02. Apply Vector Distance and Indexing (32%) • Apply vector constructor, distance metrics, and exact similarity search to query requirements • Analyze vector index options, memory considerations, and search accuracy trade-offs • Apply approximate and multi-vector similarity search techniques to retrieval scenarios 03. Build Retrieval-Augmented Generation Solutions (24%) • Explain RAG workflow, vector data workflow, and Oracle AI Database support for RAG • Build a RAG flow using chunking, embeddings, vector search, and LLM prompt generation • Use Oracle database packages and services to implement RAG with PL/SQL and external models 04. Design Select AI Agent Solutions (16%) • Explain Select AI Agent framework concepts, execution flow, and core components • Design agent solutions using tools, tasks, agents, agent teams, and built-in integrations 05. Configure Agentic Interfaces and Tool Access (12%) • Explain ADB MCP Server purpose, benefits, and integration with Select AI Agent tools • Configure ADB MCP Server and use Select AI Agent tools through MCP-enabled clients 06. Govern Private Agent Factory Deployments (8%) • Choose Private Agent Factory capabilities for no-code agent use cases • Analyze user roles, security controls, and governance features for enterprise agent deployments S E C T I O N 0 4 What the Credential Is Worth AI engineering is among the best-paid specialisms in software, though the sources disagree on the baseline. For US AI engineers in 2026, ZipRecruiter reports an average of about $129,300, Glassdoor a median of $142,100 , and ScanRole - working from 1,023 employer job postings - about $166,600 . Generative AI skills push that higher: one recruiter's 2026 guide puts LLM and generative AI engineers at $175,000 to $260,000 in base pay. These are aggregator and recruiter estimates, so treat them as a range rather than a promise. www.dbexam.com Oracle 1Z0-1159-26 Agentic AI for Oracle AI Database Certified Professional 4 Why this exam, specifically • It is the professional tier: proof you can build retrieval and agents, not just describe them. • It keeps AI next to the data - vector search and RAG in SQL, not a separate vector store. • It maps to real platform surface: HNSW and IVF indexes, DBMS_VECTOR_CHAIN, Select AI, agent teams, MCP. • It joins two skill sets - database and generative AI - that rarely sit in one person. There is a practical angle too. Enterprises already trust their Oracle databases with the data an AI application needs. The people who can make a similarity search fast and accurate, ground answers in those rows, and let an agent act on them safely are the ones turning AI pilots into production systems. S E C T I O N 0 5 Getting Certified, Step by Step 1 Read the official exam topics end to end Six topic areas, 14 objectives. Mark every line you cannot demonstrate hands-on. 2 Work the Agentic AI learning path Oracle University's path on MyLearn, plus the AI Vector Search and Select AI documentation. 3 Spend half your time on vectors Distance and indexing are 32% of the paper. Build HNSW and IVF indexes and test exact, approximate and partitioned fetches. 4 Build one RAG flow and one agent team Chunk, embed and search documents in PL/SQL, then run a Select AI Agent team and expose a tool through the MCP Server. 5 Rehearse, then book through Oracle University Fifty questions in 90 minutes on MyLearn. Passing bar is 68%; book once you are comfortably above it. For the rehearsal, DBExam's 1Z0-1159-26 practice exam mirrors the real format so you can find the gaps before the exam does. Recommended training runs through the Agentic AI learning path named in the fact sheet, on Oracle MyLearn. www.dbexam.com Oracle 1Z0-1159-26 Agentic AI for Oracle AI Database Certified Professional 5 S E C T I O N 0 6 Oracle 1Z0-1159-26 Exam Sample Questions These ten questions come from DBExam's 1Z0-1159-26 sample questions page, presented here in shuffled order. Note their shape: most give a working system that does something slightly wrong - a query that ignores its index, an agent run stuck waiting, a tool the LLM calls too often - and ask for the precise fix. All ten are single-answer. SQL is shown in a code block. The answer key is at the end of the section. Q1. An agritech company runs exact similarity searches over crop-image embeddings. The embedding model is documented for Euclidean distance, and the vectors are not normalized. The application only needs the ten nearest images in the right order; it never displays or stores the distance values. The team wants a cheaper computation than EUCLIDEAN without changing which images are returned or their order. Which metric should the team use? a) EUCLIDEAN_SQUARED, which keeps the order and skips the square root step b) COSINE, which keeps the order while it ignores the vector magnitude c) HAMMING, which keeps the order by counting the differing dimensions d) MANHATTAN, which keeps the order and uses cheaper sums of per-dimension differences Q2. A healthcare IT vendor stores clinical guideline PDFs in a BLOB column. A developer writes a PL/SQL procedure that should extract plain text from each PDF, then chunk and embed it. The procedure calls DBMS_VECTOR.UTL_TO_TEXT, and compilation fails because the function cannot be found. The Oracle Text CONTEXT component is installed. What should the developer do? a) Call DBMS_VECTOR.UTL_TO_SUMMARY on each BLOB, then pass its plain-text summary to the chunking step b) Call DBMS_VECTOR_CHAIN.UTL_TO_TEXT on each BLOB, then pass its plain-text CLOB to the chunking step c) Apply VECTOR_EMBEDDING to each BLOB directly, then store one vector per PDF with its guideline d) Pass each BLOB straight to DBMS_VECTOR.UTL_TO_CHUNKS, letting it pull the PDF text out before chunking www.dbexam.com Oracle 1Z0-1159-26 Agentic AI for Oracle AI Database Certified Professional 6 Q3. An online marketplace serves product recommendations from an IVF vector index. One fraud-review workload needs noticeably better recall than the default, while every other workload is satisfied with current results. The DBA will not rebuild the index during the current release. What should the developer change for the fraud-review queries? a) Recreate the index with a larger NEIGHBOR PARTITIONS value in its PARAMETERS b) Add WITH TARGET ACCURACY PARAMETERS (EFSEARCH n) with a larger n to those queries c) Recreate the index with a larger EFCONSTRUCTION value in its PARAMETERS d) Add WITH TARGET ACCURACY PARAMETERS (NEIGHBOR PARTITION PROBES n) with a larger n to those queries Q4. An energy retailer exposes a tool named METER_READINGS through its Autonomous AI Database MCP Server. Analysts using an MCP client report that the LLM calls the tool for questions about billing disputes, which the tool cannot answer. The Select AI Agent user wants to rewrite the tool's instruction so the LLM calls it only for meter-reading questions. How should the tool be changed? a) Disable and re-enable the MCP Server with the adb$feature free-form tag, which reloads each tool with its latest instruction b) Move the new guidance into the tool_inputs argument descriptions and leave the tool's instruction attribute unchanged c) Drop the tool, recreate it with CREATE_TOOL using the new instruction, and have MCP clients reconnect d) Call CREATE_TOOL again with the same tool name and the new instruction; connected clients pick it up on their next call Q5. A healthcare-IT vendor's APEX application starts an agent team with DBMS_CLOUD_AI_AGENT.RUN_TEAM to schedule clinic equipment maintenance. The developer created the agent and task without setting enable_human_tool. A support ticket reports that a run never finished, and the team history view shows its state as WAITING_FOR_HUMAN. What explains this state, and how should the application continue the run? a) enable_human_tool defaults to true, so the agent asked the user a question; create a new team so the run can start over cleanly b) short_term_memory_length reached its default of 30 turns; call CLEAR_TEAM to free that memory and let the run continue c) enable_human_tool defaults to false, so a tool call failed; call SET_TEAM again to restart the paused run in this session d) enable_human_tool defaults to true, so the agent asked the user a question; call RUN_TEAM with the user's answer as the prompt www.dbexam.com Oracle 1Z0-1159-26 Agentic AI for Oracle AI Database Certified Professional 7 Q6. An insurance company has configured Select AI RAG on Autonomous AI Database. The DBA created a vector index over policy documents in Object Storage with DBMS_CLOUD_AI.CREATE_VECTOR_INDEX and set the AI profile's vector_index_name attribute to it. A claims analyst runs SELECT AI showsql what does the flood rider exclude and receives a SQL statement instead of an answer drawn from the policy documents. What should the analyst do to get a document-grounded answer? a) Run the same prompt with the narrate action instead of showsql b) Run the same prompt with the showprompt action instead of showsql c) Run the same prompt with the explainsql action instead of showsql d) Run the same prompt with the chat action instead of showsql Q7. A bank runs Autonomous AI Database Serverless, and its fraud table already has an HNSW vector index on the transaction embedding column for interactive lookups. A new quarterly compliance report must list the 20 nearest transactions to a reference pattern, and the auditors require certainty that the result comes from comparing against every vector rather than from an index. How should the developer write the report query? a) Use FETCH FIRST 20 ROWS ONLY with no extra keyword b) Add a NO_VECTOR_INDEX_SCAN hint to FETCH FIRST 20 ROWS ONLY c) Use FETCH EXACT FIRST 20 ROWS ONLY after the ORDER BY d) Use FETCH FIRST 20 ROWS ONLY WITH TARGET ACCURACY 100 after the ORDER BY Q8. An online education platform created this index and then tuned a lesson-search query: CREATE VECTOR INDEX lesson_hnsw ON lessons(emb) ORGANIZATION INMEMORY NEIGHBOR GRAPH WITH TARGET ACCURACY 95; SELECT lesson_id FROM lessons ORDER BY VECTOR_DISTANCE(emb, :q, EUCLIDEAN) FETCH APPROX FIRST 10 ROWS ONLY; The query returns correct results with no error, but it is slow and its plan compares the query vector with every row. Which change lets the query use the index? a) Lower the query's target accuracy so the optimizer can choose the index b) Rebuild the index with DBMS_VECTOR.REBUILD_INDEX to refresh the graph c) Use COSINE in the query so its metric matches the index's metric d) Raise VECTOR_MEMORY_SIZE so the HNSW graph has room in the vector pool www.dbexam.com Oracle 1Z0-1159-26 Agentic AI for Oracle AI Database Certified Professional 8 Q9. A logistics company is building a RAG pipeline for shipping-policy documents stored as text in a POLICY_DOCS table. A developer writes this query to split each document into chunks, and it fails: SELECT d.doc_id, VECTOR_CHUNKS(d.body BY WORDS MAX 100 OVERLAP 10) FROM policy_docs d; What change makes the query work? a) Wrap the call in VECTOR_EMBEDDING so the SELECT list returns one vector per document b) Move the VECTOR_CHUNKS call into the FROM clause as a lateral virtual table and select CHUNK_TEXT c) Move OVERLAP 10 ahead of MAX 100 inside the call and keep the call in the SELECT list d) Change BY WORDS to BY CHARACTERS and raise MAX to 400 characters, keeping the call in the SELECT list Q10. A patent analytics firm splits each patent into chunks and stores one embedding per chunk with its doc_id. An analyst runs: SELECT doc_id, chunk_id FROM patent_chunks ORDER BY VECTOR_DISTANCE(emb, :q, COSINE) FETCH EXACT FIRST 3 PARTITIONS BY doc_id, 2 ROWS ONLY; Which description matches the result set? a) The three nearest chunks from each of the two nearest patents, in distance order b) Up to two nearest chunks from each of the three nearest patents, in distance order c) The six nearest chunks overall, in distance order, from any patent d) Up to two nearest chunks from each of the three nearest patents, grouped together by patent Answer key Q1 Q2 Q3 Q4 Q5 Q6 Q7 Q8 Q9 Q10 a b d c d a c c b b S E C T I O N 0 7 Where to Go Next Read what other developers are running into on the Oracle community before you book - vector index and Select AI questions surface there first. The DBExam 1Z0-1159-26 exam page keeps the syllabus, sample questions and practice exams in one place. www.dbexam.com Oracle 1Z0-1159-26 Agentic AI for Oracle AI Database Certified Professional 9 Quick reference • Official exam page — mylearn.oracle.com • Exam syllabus + practice test — dbexam.com/oracle • Community threads — forums.oracle.com • Scheduling — oracle.com/education/training/buy Before you book — the guide in one card Exam number 1Z0-1159-26 Passing score 68% Questions 50 Duration 90 minutes Fee USD 245 Delivery Oracle University / MyLearn Topic areas 6 Objectives 14 Heaviest area Vector Distance and Indexing (32%) Next heaviest Retrieval-Augmented Generation (24%) Good luck - and remember what the blueprint is really asking: not "what is a vector?", but "can you find the right rows fast enough for an agent to act on them?"