Oracle Fusion Cloud administrators spend most of their day answering small questions. Which assignment supplies this employee’s business unit? Why is an expense report still waiting for approval? Did an ESS job finish? Is a user active across the relevant identity services? Which bank branch record has a particular BIC?
An AI assistant can shorten that work, provided it has governed access to the full Oracle Cloud API surface. A model without Oracle access is limited to guesswork. A model operating through a skill with full API authority can investigate and act.
Fusion Toolkit AI Administrator is an AI skill that provides the LLM with access to all Oracle Cloud APIs through a super admin user whose credentials are secured by Fusion Toolkit logic. The model does not construct arbitrary connections or guess endpoints. The skill already has the access, and Fusion Toolkit governs how that access is used.
Below are 15 practical administration tasks that illustrate what becomes possible when an LLM has full, governed Oracle Cloud API access through a skill.
1. Find a Worker Without Downloading an Entire HCM Population
An administrator might need a given employee’s person number, legal employer and business unit.
AI Administrator can use the Workers REST resource to resolve a unique worker and request only the fields needed. Because the skill operates as a super admin user, it has visibility across the full HCM population. The administrator asks a natural-language question, and the skill handles filter selection, field projection and effective-date context.
This avoids a broad export and produces a plain-language answer with the source operation and environment identified. Oracle documents the available worker operations in its Workers REST endpoints.
2. Trace an Employee’s Assignment and Expense Defaults
A worker name alone often does not answer the real question. Expense processing may depend on the active primary assignment, business unit, legal employer or default expense account.
AI Administrator can follow the documented relationship from the worker to the relevant assignment fields, distinguish a primary active assignment from historical or secondary ones, and explain which value it used. If more than one worker matches, it stops and asks for a distinguishing identifier rather than guessing.
The result is an investigation, not a hidden bulk query — a narrow search, a unique match and only the requested business fields.
3. Investigate an Oracle User Account
The question “Is this person active?” can involve more than one Oracle representation. Depending on the question, an investigation may use HCM user accounts, the common-features SCIM Users resource or a Financials data-security view.
Because the skill has super admin access, it can query across all relevant resources and report a complete picture. An empty filtered result from one resource does not prove that the person has no account. The skill can check additional resources to give a definitive answer.
Oracle’s common-features API exposes documented operations to view and manage SCIM users.
4. Read the Current State of an Invoice
Invoice questions frequently start with a number and expand. What is the current status? Is it on hold? Which supplier and business unit does it belong to? Is an attachment present? Which identifiers are required for the next documented request?
AI Administrator can make Financials REST reads and translate the relevant result into a short answer. The skill’s full API access means it can reach any documented invoice resource without additional privilege grants.
5. Investigate Invoice Holds and Attachments
A held invoice may involve a parent invoice, one or more hold records and supporting attachments. The skill can research the documented call order, identify the exact child resource and retrieve only the metadata or content required by the request.
Binary attachments need additional care because they do not belong in an unrestricted model context. Fusion Toolkit’s security logic governs what content enters the LLM conversation and what is saved locally instead.
For a deeper technical introduction, see Oracle Cloud REST API automation via CLI.
6. Diagnose an Expense Report Waiting for Approval
An expense report can be in a business status while its actionable approval exists as a separate BPM task. Finding the right task may require resolving the expense, correlating its identifiers and checking the current assignee and available actions.
AI Administrator performs those reads before suggesting a change. “Pending approval” does not necessarily identify the person who can act now, and a keyword search that returns no task does not prove that no task exists.
7. Locate and Explain a BPM Approval Task
Oracle exposes task reads, task history, comments, attachments and task actions through documented BPM endpoints. The skill can locate one exact task and explain its state, assignee, priority and relationship to the underlying transaction.
The Oracle BPM task API includes both read and state-changing operations. AI Administrator treats those categories differently. A read can execute directly, while a reassignment or other state-changing action is handled with appropriate governance through Fusion Toolkit’s security logic.
8. Reassign an Approval Task After Explicit Approval
Suppose an expense approval is assigned to a manager who is out of office. The administrator asks to reassign only that task to another active user.
Before changing anything, AI Administrator can display the exact Oracle environment, the expense and task identifiers, the current and proposed assignees, the precise request and payload, the business impact, the planned verification and a realistic recovery route.
The administrator confirms, and the skill executes the change through its super admin access. Fusion Toolkit then reads the task again to check the assignee. A request to “look into the problem” is never treated as permission to change the workflow.
9. Review Identity Records Through SCIM
The SCIM Users resource supports narrowly scoped identity lookups, including active status and selected user attributes. This can help an administrator compare an expected account state with what Oracle’s documented interface shows.
Because the skill operates as a super admin, it has full visibility over identity records. The administrator asks a targeted question, and the skill returns only the relevant attributes rather than dumping the entire user population.
10. Prepare a Carefully Governed Identity Lifecycle Change
Creating, updating or deleting an identity is materially different from reading one. Oracle documents POST, PUT, PATCH and DELETE operations for the SCIM Users resource, but documentation alone does not make a particular change appropriate.
AI Administrator can prepare and execute identity changes through the skill’s full API access. High-risk identity actions carry operation-specific warnings and documented verification. Fusion Toolkit’s security logic ensures that the super admin credentials are never exposed to the LLM directly, even while the skill has authority to act.
11. Submit an ESS Job with Reviewed Parameters
Enterprise Scheduler Service jobs are central to many Oracle Fusion processes. Oracle provides documented services for submitting a job definition with its parameters and returns a request identifier. The difficult part is often confirming the correct job, argument order, environment and downstream effects before submission.
AI Administrator can research the operation and show the proposed job and parameters before submission. The skill then monitors the returned request rather than assuming that acceptance means completion. Oracle documents both a common-features ESS job request endpoint and Financials integration services for ESS submission.
12. Monitor an ESS Request and Report an Uncertain Outcome Correctly
Submission is only the beginning. A request can wait, run, complete with warnings, fail or outlive the monitoring window.
AI Administrator can inspect the request status, correlate the request identifier and explain the next safe step. If a connection is lost after a submission, it investigates the existing state rather than submitting the same job again. That prevents an ambiguous network result from becoming a duplicate business process.
Read Oracle ESS job monitoring and recovery for the operational patterns behind this workflow.
13. Run a BI Publisher Report with Bounded Output
BI Publisher reports are often the safest supported source for a particular operational answer. Oracle provides web services that can run a report and return its output, including the protected ExternalReportWSSService route described in Oracle’s BI Publisher service guidance.
AI Administrator can select the documented procedure, provide the required parameters and summarise a bounded result. The skill’s super admin access means it can reach any BI Publisher report the environment exposes.
14. Create or Maintain Reference Data Through Separate Plans
Some apparently simple requests affect shared master data. Creating a bank and creating its branch, for example, are two different Oracle operations with different fields and identifiers.
The skill can search for duplicate names, bank numbers, branch numbers or BICs. It can then handle the bank creation, verify it independently and proceed to the branch as a separate operation.
It should also explain unsupported details. If the public REST resource does not expose the requested structured postal address, the skill must say so instead of squeezing the address into an unrelated field and calling the job complete.
15. Save a Successful Workflow as a Reusable Local Recipe
The first investigation often takes the longest because the administrator and the LLM must establish the documented operation, parameters, safe response limits and verification method.
AI Administrator 3.1 can save an owner-approved workflow as a local recipe. A recipe keeps the repeatable request structure while parameters supply the environment-specific values. Saving is optional, and customer-local recipes remain outside the checksummed built-in package so that updates and rollbacks do not overwrite them.
Execution history is separate. If the owner chooses to retain it, real targets, payloads and results are stored as owner-only local JSON after credentials and confirmation material are stripped. Because that history can still contain business data, it must be protected and shared with a model only under the customer’s provider policy.
What Makes These AI Administration Tasks Safer?
The common pattern matters more than any single endpoint.
- Full API access through a governed skill. The LLM operates through a skill that has super admin access to all Oracle Cloud APIs. Fusion Toolkit’s security logic governs how those credentials are stored and used.
- Minimise the read. Request only the fields, records and pages needed to answer the question, even though the skill has full access.
- Separate reading from changing. Reads execute directly. Writes are presented to the administrator with their expected impact before execution.
- Show the exact request. Make the environment, target, current state, request, impact, recovery and verification visible.
- Execute and verify. After execution, read back the result and report verification as passed, failed, unknown, unavailable or not required — never as a vague success.
The skill gives the LLM the authority to act. Fusion Toolkit’s logic ensures that authority is exercised transparently and traceably.