Overview
Domain 4 moves from retrieval to agents. It starts with Select AI on Autonomous AI Database: an AI profile created with DBMS_CLOUD_AI.CREATE_PROFILE names the LLM provider, credential and the tables the model may see. SELECT AI then converts natural language into SQL with actions such as runsql, showsql, explainsql, narrate and chat. Select AI Agent extends this into agentic workflows through the DBMS_CLOUD_AI_AGENT package.
You define tools (SQL, RAG, web search, notification or custom PL/SQL), agents that pair a role with a profile, tasks that give instructions and allowed tools, and teams that run agents and tasks together. A team is activated with SET_TEAM and invoked with SELECT AI AGENT or RUN_TEAM. The agent reasons over a request, chooses tools, observes results and iterates until it can answer, all with database security in force.
How it's tested
Questions describe an assistant someone wants to build and ask which DBMS_CLOUD_AI or DBMS_CLOUD_AI_AGENT call, Select AI action or object definition achieves it. Expect to match objects to responsibilities, choose the action that shows generated SQL without running it, and spot why an agent cannot answer.
-
Pick the Select AI action
A user needs SQL shown, explained, executed or narrated, and you choose showsql, explainsql, runsql or narrate.
-
Map the agent object
A responsibility such as tool permissions, LLM choice or execution order is described, and you name the tool, agent, task or team object.
-
Find the missing setup
An agent fails to reach data or a tool, and you identify the absent profile setting, object list entry or tool assignment.
An analyst wants to see the SQL Select AI would generate, check it, and only then run it. SELECT AI RUNSQL executes immediately, and NARRATE hides the SQL behind prose. SHOWSQL returns the generated statement without running it, which is exactly what the analyst asked for. EXPLAINSQL adds a natural-language explanation, but the core need is to view before running.
Topics 4
Official topics from the Oracle outline, in outline order.
4.1 Configure Select AI profiles for natural language to SQLDBMS_CLOUD_AI.CREATE_PROFILE, object lists and metadata.
- What it means
- A Select AI profile tells Autonomous AI Database which LLM provider and model to use, which credential to authenticate with, and which schemas or tables the model may reference. DBMS_CLOUD_AI.CREATE_PROFILE creates it and SET_PROFILE makes it active for the session.
- What to understand
- The object list limits the metadata sent to the LLM, which affects both accuracy and data exposure. Table and column comments or annotations, enabled through profile attributes, give the model business meaning that improves generated SQL. Only metadata, not row data, is sent for SQL generation, while narrate and chat may send results or prompts to the model.
- How it shows up
- Questions ask which profile attribute limits accessible tables, or why generated SQL misreads cryptic column names.
- Common confusion
- Thinking Select AI sends all table rows to the LLM to write SQL; it sends schema metadata.
4.2 Use Select AI actions and conversationsrunsql, showsql, explainsql, narrate, chat and conversation history.
- What it means
- SELECT AI accepts an action keyword. RUNSQL, the default, generates and runs SQL. SHOWSQL returns the SQL only, EXPLAINSQL explains it, NARRATE describes results in prose, CHAT passes the prompt straight to the LLM, and SHOWPROMPT reveals the prompt sent. DBMS_CLOUD_AI.GENERATE offers the same features from PL/SQL and applications.
- What to understand
- Choosing the right action controls risk and output: showsql suits review workflows, narrate suits business users, chat suits general questions without database context. Conversations created with DBMS_CLOUD_AI conversation routines keep prior prompts so follow-up questions such as 'and last month?' resolve correctly.
- How it shows up
- A user need is described and you select the action keyword that meets it.
- Common confusion
- Mixing up CHAT with NARRATE; chat does not query your tables, narrate summarises query results.
4.3 Build Select AI Agent tools, agents and tasksDBMS_CLOUD_AI_AGENT.CREATE_TOOL, CREATE_AGENT and CREATE_TASK.
- What it means
- DBMS_CLOUD_AI_AGENT lets you define tools such as SQL (natural language to SQL through a profile), RAG (retrieval through a vector index profile), web search, notification by email or Slack, and custom PL/SQL functions. CREATE_AGENT gives an agent a role and an AI profile, and CREATE_TASK holds instructions plus the tools that task may use.
- What to understand
- Separating tools, agents and tasks keeps designs reusable: one SQL tool can serve several tasks, and one agent can carry out different tasks. Tool descriptions matter because the LLM reads them when deciding which tool to call. Custom PL/SQL tools let agents act on business logic, such as raising a ticket, while staying inside database privileges.
- How it shows up
- Expect a scenario that asks where a tool permission, role description or instruction belongs.
- Common confusion
- Putting tool lists on the agent rather than the task; tasks name the tools they may use.
4.4 Orchestrate agent teams and the agent loopCREATE_TEAM, SET_TEAM, SELECT AI AGENT and ReAct reasoning.
- What it means
- A team created with DBMS_CLOUD_AI_AGENT.CREATE_TEAM pairs agents with tasks and an execution process. SET_TEAM activates it for the session, after which SELECT AI AGENT followed by a request, or RUN_TEAM from PL/SQL, starts the workflow. Each agent follows a reason, act, observe loop until it can answer or needs human input.
- What to understand
- The loop explains agent behaviour: the LLM plans, calls a tool, reads the result and decides what next. Teams can chain tasks so one agent's output feeds another. History and tool logs in database views help trace decisions. Human-in-the-loop tools let an agent ask a user for clarification instead of guessing.
- How it shows up
- Questions ask what statement invokes an active team or why an agent repeated a tool call.
- Common confusion
- Confusing SET_PROFILE with SET_TEAM; profiles serve Select AI, teams serve Select AI Agent.
High-yield concepts
The terms, rules and formulas this domain turns into questions most often.
-
AI profile
DBMS_CLOUD_AI configuration of provider, model, credential and objects.
- Remember
- Activated with SET_PROFILE.
- Tested as
- Every Select AI question.
- Trap
- Forgetting to set it for the session.
-
Object list
Profile attribute naming tables the LLM may reference.
- Remember
- Limits metadata exposure and improves accuracy.
- Tested as
- Data-scope questions.
- Trap
- Leaving it broad by default.
-
Comments and annotations
Business descriptions of tables and columns used in prompts.
- Remember
- Improve SQL generation on cryptic schemas.
- Tested as
- Accuracy tuning questions.
- Trap
- Renaming columns instead.
-
SHOWSQL
Returns generated SQL without running it.
- Remember
- Review-before-run workflows.
- Tested as
- Action choice questions.
- Trap
- Choosing RUNSQL.
-
NARRATE vs. CHAT
Narrate summarises query results; chat talks to the LLM directly.
- Remember
- Only narrate queries your data.
- Tested as
- Action choice questions.
- Trap
- Using chat for data questions.
-
DBMS_CLOUD_AI.GENERATE
Function exposing Select AI to PL/SQL and apps.
- Remember
- Same actions without the SELECT AI keyword.
- Tested as
- Application integration.
- Trap
- Assuming Select AI is SQL*Plus only.
-
DBMS_CLOUD_AI_AGENT
Package for Select AI Agent tools, agents, tasks and teams.
- Remember
- Separate from DBMS_CLOUD_AI.
- Tested as
- Agent build questions.
- Trap
- Creating agents with DBMS_CLOUD_AI.
-
Built-in tool types
SQL, RAG, web search, notification and custom PL/SQL.
- Remember
- Tool descriptions guide the LLM's choice.
- Tested as
- Tool selection scenarios.
- Trap
- Writing PL/SQL for what a built-in tool covers.
-
Task
Instructions plus allowed tools for a unit of work.
- Remember
- Where tool permissions are assigned.
- Tested as
- Design questions.
- Trap
- Assigning tools to agents.
-
Team and SET_TEAM
Groups agents and tasks; SET_TEAM activates it.
- Remember
- Invoke with SELECT AI AGENT or RUN_TEAM.
- Tested as
- Execution questions.
- Trap
- Using SET_PROFILE for teams.
-
ReAct loop
Reason, act with a tool, observe, repeat.
- Remember
- Explains multi-step agent behaviour.
- Tested as
- Behaviour and troubleshooting.
- Trap
- Expecting a single LLM call.
Side by side
Concepts candidates mix up, compared on the point that decides the answer.
Select AI actions
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| RUNSQL | Generate and execute SQL | Default action, returns rows | Using it when review is required |
| SHOWSQL | Generate SQL only | Nothing is executed | Expecting results |
| EXPLAINSQL | Generate SQL plus explanation | Teaching or audit use | Confusing with SHOWPROMPT |
| NARRATE | Run query, describe results in prose | Business-friendly output | Expecting raw rows |
| CHAT | Send prompt directly to the LLM | No database query | Asking it about table data |
Select AI Agent objects
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| Tool | A capability such as SQL, RAG or notification | What can be done | Defining tools inside agents |
| Agent | Role plus AI profile | Who does the work | Expecting it to hold tool lists |
| Task | Instructions and allowed tools | What to do and with which tools | Skipping tool assignment |
| Team | Agents paired with tasks and a process | How work runs end to end | Invoking without SET_TEAM |
Exam traps
Each trap is an answer that looks right. Learn the clue that should change your mind.
-
Rows go to the LLM for SQL
- What candidates pick
- Assume Select AI sends table data to write queries.
- Why it looks right
- The model must understand the data.
- Why it's wrong
- For SQL generation Select AI sends schema metadata within the object list; row data reaches the model only with actions like narrate.
Clue to watch for The question is about data exposure during SQL generation.
-
Chat answers data questions
- What candidates pick
- Use SELECT AI CHAT to ask about last quarter's sales.
- Why it looks right
- Chat sounds conversational and helpful.
- Why it's wrong
- CHAT bypasses the database and asks the LLM directly; narrate or runsql query your tables.
Clue to watch for The answer depends on company data.
-
Agents own the tool list
- What candidates pick
- Attach tools when creating the agent.
- Why it looks right
- Agents use tools, so it seems natural.
- Why it's wrong
- In Select AI Agent the task lists allowed tools; the agent supplies a role and profile.
Clue to watch for The question asks where tool access is granted.
-
SET_PROFILE starts a team
- What candidates pick
- Call SET_PROFILE before SELECT AI AGENT.
- Why it looks right
- Profiles are how Select AI starts.
- Why it's wrong
- Teams are activated with DBMS_CLOUD_AI_AGENT.SET_TEAM; profiles alone do not run agent workflows.
Clue to watch for SELECT AI AGENT appears in the scenario.
-
Rename columns to fix SQL
- What candidates pick
- Rename physical columns so the LLM understands them.
- Why it looks right
- Clearer names help models.
- Why it's wrong
- Comments and annotations add meaning without breaking applications that depend on existing names.
Clue to watch for Cryptic column names cause wrong SQL.
-
Custom code for everything
- What candidates pick
- Write a PL/SQL function to send an email alert.
- Why it looks right
- Custom code feels flexible.
- Why it's wrong
- A built-in notification tool covers email and Slack; custom PL/SQL tools are for business logic without a built-in option.
Clue to watch for The action is a standard notification.
Answer framework
Use these steps to design or troubleshoot a Select AI or Select AI Agent solution.
- Decide whether a single natural-language query is enough, which points to Select AI, or multi-step work, which points to Select AI Agent.
- Configure the AI profile: provider, model, credential and a tight object list, plus comments or annotations for meaning.
- Choose the action or tool for each need: showsql for review, narrate for prose, RAG tool for documents, notification tool for alerts.
- Assign tools to tasks, roles and profiles to agents, and pair agents with tasks in a team with an execution process.
- Activate the team with SET_TEAM, invoke with SELECT AI AGENT or RUN_TEAM, and trace decisions in agent history views.
Practice question
Want more like this? The 1Z0-1159-26 practice test has questions for every domain.
Checklist
0 of 12
Tick each one you can do without notes. Your ticks stay in this browser.
Study strategy
Learn Select AI first, then Select AI Agent, because the agent package builds on profiles. On an Autonomous AI Database, create a credential for a provider, create a profile with a short object list, and run the same question through every action keyword. Note what each returns and whether it touched your tables.
Add comments to a few cryptic columns and compare the generated SQL before and after. Try a conversation with two follow-up questions so you see history in action.
Then build a small Select AI Agent team: one SQL tool on a sales schema, one RAG tool over a policy document index, and a notification tool. Create an agent, a task that lists those tools, and a team; activate it with SET_TEAM and invoke it with SELECT AI AGENT. Read the history views to follow the agent's tool choices. Draw the tool, agent, task and team objects on one page with an arrow for each relationship, because many questions ask where a setting belongs. Finish with timed questions on action choice and object mapping. If a practice item stumps you, recreate its scenario in the database rather than rereading notes.
FAQ
How much of 1Z0-1159-26 is Domain 4, Design Select AI Agent Solutions?
Oracle does not publish weights. This Select AI Agent section is central to the exam's agentic theme, and third-party sources suggest a mid-sized share of questions.
What is the difference between Select AI and Select AI Agent?
Select AI turns one natural-language prompt into SQL or prose. Select AI Agent runs multi-step workflows where agents choose among tools such as SQL, RAG and notification.
Which package creates agents?
DBMS_CLOUD_AI_AGENT creates tools, agents, tasks and teams. DBMS_CLOUD_AI handles profiles and Select AI features.
Does Select AI send my data to the LLM?
For SQL generation it sends schema metadata from the object list. Actions such as narrate send query results so the model can describe them.
How do I run an agent team?
Activate it with DBMS_CLOUD_AI_AGENT.SET_TEAM, then use SELECT AI AGENT with your request, or call RUN_TEAM from PL/SQL.
What built-in tools can an agent use?
Oracle documents SQL, RAG, web search and notification tools, and you can register custom PL/SQL functions as tools.
How can I improve wrong SQL from Select AI?
Narrow the object list and add table and column comments or annotations so the model understands business meaning.
Other Artificial Intelligence & Machine Learning certifications
Where the same subject shows up in other exams. Cards marked "Same subject" open the matching domain.
Test your Domain 4 knowledge
You have covered Select AI profiles and actions and the tools, agents, tasks and teams behind Select AI Agent. Check whether you can map a requirement to the right object.
Keep going
- Training Training Built on your Weak areas
- Practice test Questions by domain, with explanations
- Mock exam Full length, timed like the real exam
- Readiness test Find your weakest domain in minutes
- Cheat sheet Key terms and formulas on one page
- Ask the 1Z0-1159-26 AI tutor Get any concept explained your way