AI01: AI Agent Studio, Agent Creation, Security, Workflows, Debugging, PDF Extraction, Token Management, Name Matching (24-Oct-26)

Create a Free Account to Access

(login if you already have an account)

  • Register (Free)
Lost your password? Please enter your username or email address. You will receive a link to create a new password via email.

0:00 – 1: Case Study Assignments and Solutions       ⟶       

7:00 – 2: AI Agent Studio Course Overview       ⟶       

10:38 – 3: Comprehensive Module Training Roadmap       ⟶       

17:46 – 4: Defining AI Agents and Studio       ⟶       

22:44 – 5: AI Agent Studio Interface Tour       ⟶       

28:33 – 6: Building Agents With AI Assist       ⟶       

35:11 – 7: Creating Custom Security Roles       ⟶       

47:44 – 8: Enabling Profile Options and Processes       ⟶       

57:28 – 9: Agent Access and Deployment Methods       ⟶       

1:00:15 – 10: Delivered Versus Custom Agents       ⟶       

1:07:29 – 11: Building a Workflow Agent       ⟶       

1:12:18 – 12: Designing Agent Workflows       ⟶       

1:19:00 – 13: Testing Agent Data Fetching       ⟶       

1:24:58 – 14: Business Object Parameter Filtration       ⟶       

1:33:22 – 15: Automated Email Output via HTML       ⟶       

1:50:22 – 16: Multi Agent Debugging and Supervision       ⟶       

1:56:59 – 17: PDF Extraction and OCR Capabilities       ⟶       

2:00:23 – 18: Managing Token Usage and Large Documents       ⟶       

2:04:36 – 19: Exact Name Matching vs Wildcards       ⟶       

2:08:03 – 20: Membership Details and Payment Information       ⟶       

Getting Started with AI Agent Studio

Kestrel Outdoor Gear Ltd. designs and sells camping, hiking and climbing equipment through its own stores and online. It has offices in Bengaluru and Denver, and it runs Oracle Fusion Cloud for Financials and HCM.

The accounts payable team spends hours every week answering the same questions from managers: which invoices are open for a supplier, which ones are unpaid, and could someone please email the list. Each answer means running a report, filtering it and pasting it into an email.

Kestrel wants to start using AI Agent Studio. Before anything goes live, it wants the right people to have the right access, and a first working agent that the payables team can ask in plain English.

Key Business Requirements

Concept explain how Oracle Fusion meets it. Hands-on configure it in your practice instance.

  1. ConceptThe leadership team wants a plain explanation of what an AI agent is and what AI Agent Studio does. They also want to know whether a separate system or connection is needed for agents to read Kestrel’s Fusion data.

    Hint:00:17:46 to 00:19:36, and 00:19:36 to 00:21:19See solution

  2. ConceptFinance wants to know how agent usage is measured and charged, and how much usage comes with the Fusion subscription each month.

    Hint:00:21:21 to 00:22:23, and 00:23:23 to 00:23:58See solution

  3. ConceptOracle already ships agents with Fusion. Kestrel wants to know whether it can use or change those agents, where to find Oracle’s ready-made templates and agents built by partners, and when it would build its own agent instead.

    Hint:00:24:29 to 00:26:01, and 00:26:32 to 00:28:28, and 00:59:52 to 01:01:58See solution

  4. Hands-onTo see how quickly an agent can be built, the project lead wants a simple welcome workflow created from a single plain English request, without configuring anything by hand. Name it with your initials, for example AB_KESTREL_WELCOME. It should greet the signed in user with a short motivational message about starting with AI Agent Studio, and it should answer when you type hi.

    Hint:00:28:35 to 00:32:22See solution

  5. ConceptKestrel has two kinds of people. Decide what access each one needs:
    • The IT team, who will build and maintain agents in AI Agent Studio
    • Business users in payables and HR, who will only ask questions to published agents
    • Kestrel also wants to limit agent building to the Financials and HCM pillars it has bought

    Hint:00:32:36 to 00:34:58, and 00:50:25 to 00:53:30 for the developer’s user accountSee solution

  6. ConceptKestrel’s environment is new to AI Agent Studio. List the one time setup that must be done before any custom role for AI Agent Studio will work.

    Hint:00:47:00 to 00:50:17, and 00:53:38 to 00:55:08See solution

  7. Hands-onCreate a custom job role for Kestrel’s AI Agent Studio administrators, named with your initials, for example AB_KESTREL_AI_ADMIN. It must let the holder configure agents in all pillars, chat with agents, and run them.

    Hint:00:35:11 to 00:46:53See solution

  8. ConceptThe payables manager asks: if we give an AP specialist an agent that reads supplier and invoice data, can the specialist now see data their normal roles do not allow? And can we control which users see a particular agent?

    Hint:01:02:00 to 01:05:38, and 01:45:25 to 01:47:25See solution

  9. Hands-onBuild an invoice details agent for the payables team. A user types a question in plain English, such as What are the invoices for ABC Consulting, and the agent returns that supplier’s invoices as a table. Use your initials in every name: a business object such as AB_KESTREL_INVOICES_BO under Financials and Payables, and a workflow such as AB_KESTREL_INVOICE_DETAILS.

    Hint:01:07:43 to 01:19:05, and 01:19:24 to 01:20:29, and 01:23:22 to 01:24:20See solution

  10. ConceptThe payables manager now wants only unpaid invoices above 1000 for the US1 business unit. Explain the two ways the agent can apply these filters, and which one Kestrel should use.

    Hint:01:26:32 to 01:29:47, and 01:32:27 to 01:33:30, and 01:36:50 to 01:38:00See solution

  11. Hands-onManagers want the invoice list in their inbox. Extend your invoice details workflow so that it emails the result to your own email address with the subject Invoice details, formatted as an HTML table.

    Hint:01:33:30 to 01:36:18, and 01:47:28 to 01:49:47See solution

  12. Hands-onMake the invoice details workflow available to business users. Add a starter question so users do not have to type the same question every time, publish the workflow, and run it the way an end user would, outside the builder.

    Hint:01:21:31 to 01:22:29, and 01:38:56 to 01:39:28, and 01:43:45 to 01:44:16See solution

Your Task

  • For each Concept requirement, write a short solution: which Oracle Fusion feature meets it and how. Nothing needs to be set up in the instance for these in this session.
  • For each Hands-on requirement, configure it in your practice instance. Start your setup names with your own initials, for example AB_KESTREL_AI_ADMIN, so they do not clash with other learners in the shared instance.

Solution hints are hidden so you can attempt the case first. Ready to check your approach? Show solution hints

Solution by Requirement

  1. ConceptAI Agent Studio is the platform inside Oracle Fusion where you build AI agents. An agent works like a coworker or assistant: you give it instructions, rules or a flow for a task, and the LLM understands the user’s request in that context and performs the task. 00:17:46 to 00:19:36

    No separate system is needed. AI Agent Studio is built into Fusion, not bolted on, so agents read Fusion data directly through APIs and business objects. An external connection is only needed to bring in data from outside Fusion. 00:19:36 to 00:21:19

    Back to question 1

  2. ConceptUsage is measured in AI Units: each agent run consumes AI Units, and pricing is based on that consumption. 00:21:21 to 00:22:23 The subscription includes 30,000 AI Units per Fusion production environment per calendar month, up from 20,000. 00:23:23 to 00:23:58

    Back to question 2

  3. ConceptFusion has two layers of agents: delivered (embedded) agents from Oracle, and custom agents you build. Delivered agents can be used as they are, extended, or replaced by your own. Oracle builds its delivered agents with the same AI Agent Studio you use. 00:24:29 to 00:26:01
    • AI Agents in the marketplace: preconfigured templates provided by Oracle. Use or copy one, then adapt it.
    • Marketplace: templates built and published by Oracle partners.
    00:26:32 to 00:28:28

    Build a custom agent when no delivered agent covers the business need, in the same way you build a custom report when a delivered report does not fit. 00:59:52 to 01:01:58

    Back to question 3

  4. Hands-on

    Create the welcome workflow with Ask Oracle 00:22:23 to 00:22:52

    Navigation Navigator > Tools > AI Agent Studio

    • In the Ask Oracle field on the AI Agent Studio home page, select the Workflows tag, then describe the workflow in plain English, for example: Create a new workflow agent AB_KESTREL_WELCOME to greet the login user with a nice motivation to start with AI Agent Studio. 00:28:35 to 00:29:40 00:29:40 to 00:30:35
    • Answer the confirmations it asks, for example Yes, go ahead and create it. 00:30:35 to 00:31:36
    • Type hi to test it. It should reply with a greeting. 00:31:36 to 00:32:22

    Back to question 4

  5. ConceptFusion uses role based access, so each group gets only what it needs: 00:32:36 to 00:34:58
    • IT team (agent builders): a custom job role that gives full AI Agent Studio access, as built in requirement 7.
    • Business users: only the chat privilege Access Intelligent Agent Chat and the duty Fai Genai Agent Runtime Duty, so they can run agents without opening AI Agent Studio.
    • Pillar limits: add only the pillar duties Kestrel needs, for example the Financials and HCM administrator duties, instead of all pillars.

    The session gives the developer’s user account three roles: the custom AI Agent Studio admin role, Application Implementation Consultant and IT Security Manager. 00:50:25 to 00:53:30

    Back to question 5

  6. ConceptTwo one time steps, done the first time AI Agent Studio is set up in an environment: 00:47:00 to 00:50:17
    • In Setup and Maintenance, task Manage Administrator Profile Values: set Enable Security Console External Application Integration (ORA_ASE_SAS_INTEGRATION_ENABLED) to Yes at Site level, so the Security Console can work with permission groups.
    • In Scheduled Processes, run Import Resource Application Security Data, then Import User and Role Application Security Data, one after the other, never in parallel.
    00:53:38 to 00:55:08

    Both are already done in the shared practice instance. Do not change the profile option.

    Back to question 6

  7. Hands-on

    Create the AI Agent Studio admin role 00:35:11 to 00:35:41

    Navigation Navigator > Tools > Security Console > Roles > Create Role

    • Basic Information: Role Name and Role Code with your initials, for example AB_KESTREL_AI_ADMIN, Role Category Common Job Roles, and select Enable permission group. Without it the role does not work for AI Agent Studio. 00:35:41 to 00:37:50
    • Function Security Policies: add the privilege Access Intelligent Agent Chat, so the holder can chat with agents. 00:37:50 to 00:40:01
    • Role Hierarchy, Roles and Privileges tab: add Manage All Intelligent Agents. The session also adds the pillar specific manage roles listed in the lab document. 00:41:06 to 00:43:49
    • Role Hierarchy, Roles and Permission Groups tab: add the Fai Genai Agent administrator duty for each pillar (CX, FIN, GRC, HCM, PRC, PRJ, PSC, SCM) and Fai Genai Agent Runtime Duty, which lets the holder run agents. 00:43:49 to 00:46:28
    • Skip Data Security Policies and Users, then save the role. 00:46:28 to 00:46:53

    The session leaves out the optional privileges for external REST tools, Teams and Slack channels, and scheduling. Do not edit roles other learners created.

    Back to question 7

  8. ConceptNo. An agent never bypasses Fusion security. Access to the agent and access to the data are separate: if the user’s normal roles give no access to the data, the agent runs but returns nothing, for example a reply that it has no information. 01:02:00 to 01:05:38

    To control who sees an agent, open its Security tab and add the roles that may use it, for example your custom AI role. A custom role appears there only if permission groups are enabled on it. 01:45:25 to 01:47:25

    Back to question 8

  9. Hands-on

    Create the invoices business object 01:08:51 to 01:09:24

    Navigation Navigator > Tools > AI Agent Studio > Business Objects > Add

    • Name it with your initials, for example AB_KESTREL_INVOICES_BO, with family Financials and product Payables. 01:09:24 to 01:10:18
    • Add a function with a GET operation named get invoices. Take the resource path from the Payables invoices REST API documentation (the lab document has the exact path), with only the fields you want to show and a supplier name query parameter. 01:10:18 to 01:11:21
    • For the response schema, open and close brackets ({}) are enough. Save the business object. 01:11:21 to 01:11:48

    Build the invoice details workflow 01:12:24 to 01:12:56

    Navigation Navigator > Tools > AI Agent Studio > Workflows > Add

    • Name it with your initials, for example AB_KESTREL_INVOICE_DETAILS, with the same family and product as the business object, and keep the default LLM. 01:12:56 to 01:14:28
    • Add an LLM node, read input: prompt it to extract the supplier name from the user’s input, and set its output schema to just the supplier name. 01:14:26 to 01:16:37
    • Add a Business Object Function node: select your business object and its get invoices function, and pass the supplier name from the LLM node. 01:16:37 to 01:18:47
    • Add a second LLM node, format output: prompt it to format the output in a nice table. 01:18:47 to 01:19:24
    • Test with What are the invoices for ABC Consulting. Only that supplier’s invoices come back, as a table. 01:19:24 to 01:20:29 01:23:22 to 01:24:20

    If the table is not formatted well, make the format prompt more precise. The session also suggests trying another LLM.

    Back to question 9

  10. ConceptTwo ways:
    • In the LLM prompt: fetch all of the supplier’s invoices and tell the LLM node to keep only what the user asked for. It works, but the LLM processes every record, so it is slower, uses more AI Units and can be inconsistent. 01:26:32 to 01:29:47 01:32:27 to 01:33:30
    • As business object query parameters: add parameters such as business unit to the business object function, extract them from the question, and pass them to the business object node, so only matching records are fetched.
    Kestrel should use query parameters in the business object and give the LLM only a limited task. 01:36:50 to 01:38:00

    Back to question 10

  11. Hands-on

    Email the result 01:33:30 to 01:34:02

    Navigation AI Agent Studio > Workflows > your invoice details workflow

    • Add a Send Email node after the format node: your own email address, subject Invoice details, and the previous node’s output in the body. The session adds simple HTML tags, such as a bold heading, to the body. 01:34:07 to 01:36:18
    • Change the format node’s prompt to format the output as an HTML table for email. No HTML needs to be written by hand. 01:47:28 to 01:49:47
    • Run it and check that your inbox has the formatted table.

    Back to question 11

  12. Hands-on

    Add a starter question, publish and run as an end user 01:21:31 to 01:22:03

    Navigation AI Agent Studio > Workflows > your invoice details workflow

    • Add a starter question, for example What are the invoices for ABC Consulting, so users can select it instead of typing. 01:21:31 to 01:22:29
    • Publish the workflow. 01:38:56 to 01:39:28
    • Open the AI Agents page from Me, search for your workflow and ask the question. The end user sees only the final output, with no debug details. 01:43:45 to 01:44:16

    Oracle’s path to the AI Agents page is Me > Quick Actions > Show More > AI Agent Studio > AI Agents. Session 1 opens it as AI Chat.

    Back to question 12

Hint finished
Keep watching
Close
Chapter Markers
0:00 – 1: Case Study Assignments and Solutions
7:00 – 2: AI Agent Studio Course Overview
10:38 – 3: Comprehensive Module Training Roadmap
17:46 – 4: Defining AI Agents and Studio
22:44 – 5: AI Agent Studio Interface Tour
28:33 – 6: Building Agents With AI Assist
35:11 – 7: Creating Custom Security Roles
47:44 – 8: Enabling Profile Options and Processes
57:28 – 9: Agent Access and Deployment Methods
1:00:15 – 10: Delivered Versus Custom Agents
1:07:29 – 11: Building a Workflow Agent
1:12:18 – 12: Designing Agent Workflows
1:19:00 – 13: Testing Agent Data Fetching
1:24:58 – 14: Business Object Parameter Filtration
1:33:22 – 15: Automated Email Output via HTML
1:50:22 – 16: Multi Agent Debugging and Supervision
1:56:59 – 17: PDF Extraction and OCR Capabilities
2:00:23 – 18: Managing Token Usage and Large Documents
2:04:36 – 19: Exact Name Matching vs Wildcards
2:08:03 – 20: Membership Details and Payment Information

Log in to access the session materialsThe session deck and reference PDFs for this lesson are free with an OEG account.

Register (Free)
Lesson Content
Previous Lesson
Back to Course
Next Quiz