Oracle Fusion AI Agent Studio Office Hours: What Enterprises Need to Know Before Building AI Agents

AI agents are becoming a serious enterprise architecture discussion. The focus is no longer only on “what can AI answer?” The real question is: what can AI safely do inside business applications, with the right controls, roles, workflows and auditability? 

Oracle Fusion AI Agent Studio is positioned around that enterprise need. In Oracle’s Fusion AI Agent Studio Office Hours material, the platform is presented as a way to design workflows and agents, deploy and extend prebuilt templates, and create custom agents and multi-agent teams. Oracle also describes it with the message “Built in. Not Bolted on,” which reflects the direction of bringing AI into Fusion application workflows rather than treating it as a disconnected layer.  

For business and technology leaders, the takeaway is clear: Oracle AI Agent Studio is becoming a platform for building governed AI agents that can work with enterprise context, tools, business objects, workflows, guardrails and monitoring.  

Safe Harbor Note 

Oracle’s material includes a safe harbor statement. The roadmap content is intended for information purposes only and is not a commitment to deliver any specific material, code or functionality. Oracle also notes that development, release timing and pricing may change at its discretion.  

This is important for enterprise planning. Organizations should treat roadmap capabilities as directional and build phased adoption plans around what is available, what is planned and what needs validation in their own Fusion environment. 

What Is Oracle Fusion AI Agent Studio? 

Oracle Fusion AI Agent Studio is a platform for building, testing and deploying AI agents for Fusion Applications. The office hours deck highlights capabilities such as agent patterns, human-in-the-loop, security and trust, OCI advantage, native business objects, testing and validation, prebuilt templates, interoperability through MCP and A2A, credentials store, tools, topics, prompts, knowledge store, monitoring, observability and tracing.  

In practical terms, AI Agent Studio helps enterprises move from simple AI conversations to structured AI execution. 

It supports: 

• Prebuilt agent templates  
• Custom agents  
• Multi-agent teams  
• Workflow agents  
• Business object tools  
• REST/API-based interactions  
• Guard railing through topics  
• Human approval patterns  
• Monitoring and traceability  
• Vector-based retrieval  
• Multimodal and RAG capabilities  
• Integration with collaboration tools such as Microsoft Teams and Slack  

This makes it useful for business teams, IT teams, enterprise architects, security teams and implementation partners. 

Copy Template vs Use Template 

Oracle clarifies the difference between Copy Template and Use Template. Copy Template automatically adds a suffix to all artifacts in the agent team and takes the user directly to the agent team canvas. From there, the user can edit agent team settings, agents, tools and topics. 

Use Template provides a step-by-step process for configuring each artifact in the agent team. For project teams, this choice matters. Copy Template can be useful when teams want a faster starting point. Use Template may be better when teams want more guided configuration and review before deployment. 

Pre-Delivered Workflow Agents 

Oracle lists several pre-delivered workflow agents available for reuse. 

Examples include: 

• CX Service Request Resolution Workflow  
• HCM Involuntary Interpreter Agent  
• HCM Learning Creation Assistant  
• HCM Employee Replacement Workflow Assistant  
• HCM Time Assistant  
• SCM Quote to Purchase Requisition Assistant  
• SCM B2B Message Converter Workflow  
• SCM Purchase Order to Sales Order Converter  
• SCM Component Replacement Assistant  

These workflow agents are important because they reduce the need to start from a blank canvas. Instead, organizations can evaluate reusable templates, extend them where required and create a more controlled agent implementation path.   

Can Oracle AI Agents Work Outside Fusion Applications? 

Oracle’s roadmap FAQ states that there is currently no native support for integrations outside Fusion Applications. However, agents can be invoked outside Fusion Applications using REST API starting in 25D. Oracle also notes that federation between Fusion Applications and non-Fusion IDCS, along with the related UI, is planned for 26A. This is a key point for enterprise architecture. 

Organizations that want agents to work with systems such as OTM, WMS Cloud, SAP or other external platforms need to plan carefully. REST-based invocation may support certain use cases, but it will require integration design, security review, API planning and testing. 

MCP and A2A Support 

Oracle’s material highlights two important interoperability capabilities: MCP and A2A. For MCP, Oracle states that capability is coming in release 26A. Users will be able to create connections to MCP servers available as SSE or Stream able HTTP, authenticate in the MCP tool definition, view available tools and select which tools should be included in the configuration.  

For A2A, Oracle states that basic support will be available in 26A for Oracle agents to be called in third-party chat. More enhancements are planned to simplify the process and support the ability to call third-party agents into Oracle agents. For businesses building enterprise AI ecosystems, this is a major direction. Agents will not remain isolated. They will need to communicate with tools, systems and other agents across the enterprise.  

Multimodal Support and Document Interpretation 

Oracle’s FAQ states that multimodal support, including image interpretation, will be available in release 26A. This expands the potential use of AI Agent Studio beyond text-based queries. Future workflows can become more document-aware and image-aware, supporting use cases where users upload files, screenshots, forms, logs or other operational material. For functions such as service, HR, procurement, supply chain and compliance, multimodal support can improve how agents analyze business evidence and route work. 

Guardrails, PII and Sensitive Data Handling 

Oracle’s office hours content directly addresses sensitive data. The deck states that PII detection and blocking or obfuscation, along with configurability for guardrails, are on the roadmap. Until those features are available, Oracle recommends using Topics to add guard railing instructions.  

The examples include instructions telling the agent not to process personal information such as name, address, phone number or social security number, and not to provide financial advice if the user asks something that could be interpreted as regulated advice.  

This is one of the most important enterprise adoption points. AI agents should not be deployed only for productivity. They must be deployed with security, privacy, compliance and user protection built into the design. 

Version Control and Agent Security 

Oracle states that AI Agent Studio currently supports draft and published versions of agents, with no additional version control features available at that time. The material also addresses a common deployment issue: when the Security tab does not show roles in the dropdown. Oracle’s guidance is to ensure that the ORA_DR_FAI_GENERATIVE_AI_AGENT_RUNTIME_DUTY role is assigned and permission groups are enabled. This means organizations need to prepare security roles, permission groups and access models before agent rollout. Agent security should be part of the project plan, not a last-minute configuration task. 

Multilingual Response Support 

Oracle states that agents can support multiple languages in 25D. However, language support may vary by release, product, use case or feature. The deck also notes that only English is available for Enterprise Performance Management features with AI.  

Another important detail: the agent’s response language is not automatically synchronized with the user’s sign-in language or preferred language. The response language must be configured by the administrator during agent design, or the user must explicitly specify the preferred response language in the prompt. For global organizations, this means multilingual AI rollout needs governance. Teams should test language behavior, create user guidance and define where multilingual support is appropriate. 

Exposing AI Agents Through Smart Actions 

Oracle provides an example of exposing an AI agent through Smart Actions in CX Service Centre. 

The prerequisites are: 

• Sandbox access
• Published agent  

The navigation path is: 

Application Composer → Common Setup → Smart Actions 

The setup uses a UI-based action with Agent Team as the action type. This is important because it shows how AI agents can become part of daily application workflows. Instead of sending users to a separate AI tool, businesses can surface agent actions directly where employees are already working. 

Workflow Agents and Vector DB 

The deck explains workflow agent documentation and Vector DB usage. 

Oracle describes three important components: 

Document Processor

Extracts text or content from Fusion UCM files for downstream use. The example given is processing a log file and extracting error details. 

Vector DB Writer

Persists structured JSON as semantic vectors for future retrieval. Oracle gives the example of storing normalized enriched Person and Skill objects as vectors. 

Vector DB Reader

Retrieves semantic objects or fragments using vector similarity search. The example given is retrieving employees with similar skill vectors and career interests. 

Oracle also notes that users do not need to create Vector DB connections because Vector DB is preconfigured in AI Agent Studio. For enterprise AI teams, this is critical for RAG and semantic retrieval. It enables agents to use knowledge and context more intelligently instead of depending only on static prompts. 

Agent Design Architecture: Single Agent, Multi-Agent or Workflow Agent 

Oracle’s architecture slide gives a practical decision model. 

Use a Single Agent When 

One role or persona can handle the task entirely. 

Use Multi-Agents When 

• The path to an answer is not always clear  
• Agents need to reason across multiple domains  
• Faster experimentation and more natural conversations are required  
• The task is exploratory or analytical rather than transactional  

Use a Workflow Agent When 

• Accuracy and consistency are required  
• Every action must be logged and auditable  
• The process touches systems of record  
• The process should be predictable and repeatable  

This is one of the strongest implementation insights in the deck. It helps teams avoid overengineering simple use cases and under-governing critical workflows.  

Oracle AI Agent Studio as a Build, Test and Deploy Platform 

Oracle positions AI Agent Studio as a platform to build, test and deploy AI agents. The platform view includes several capability areas: 

Integration and Extensibility 

MCP support, A2A agent cards, credential store, Microsoft Teams, Slack, dynamic custom headers, common integrations and connectors. 

Observability and Evaluation 

Monitoring, evaluation, test dataset management, agent tracing, A/B comparisons, performance metrics, token usage, RAG metrics, improved debugging, enhanced guardrails and auditing. 

LLM Options 

OpenAI, Meta, Cohere, Google, xAI and Anthropic. 

AI Agents 

New agent templates seeded runnable agent teams, custom agents, agent builder assistant, AI Agent Studio FAQ agent, Fusion business object actions and environment promotion. 

Multimodal and RAG 

Image and document upload in chat, multimodal RAG and SharePoint integration for RAG on external documents. 

Workflow Nodes 

Agent node, document processor, LLM, vector read/write, for-loop iteration, parallel execution, switch branching, human-in-the-loop, code/tool nodes, email trigger, time trigger and event trigger.  

This makes AI Agent Studio a complete enterprise AI agent lifecycle platform rather than a basic AI prompt interface. 

Prebuilt Templates in 25C and 25D 

Oracle’s material lists a wide range of prebuilt templates across 25C and 25D. The 25C list includes templates across Public Sector, Service, Sales and Marketing, Human Resources, Supply Chain, Finance and Procurement. Examples include Service Request Triage Agent, Account Advisor Agent, Sales Analyzer, Benefit Analyst, Pay Analyst, Procurement Policy Advisor, Quality Inspection Advisor, Material Handling Agent and Claims Policy Advisor.  

The 25D list expands coverage across CPQ, service, incentives, HCM, supply chain and manufacturing, procurement, ERP and industry, and public sector. Examples include Quote Advisor, Case Analyzer Agent, Contract Summarization Agent, Lead Advisor Agent Team, Deal Advisor Agent Team, Payroll Run Analyst, B2B Message Converter, Product 360 Advisor, Supplier Portal Advisor, Purchase Requisition Status Advisor and Access Request Assistant. This template library gives organizations a practical place to start. It also helps implementation teams understand Oracle’s recommended agent patterns across business functions. 

Prompt Design Examples from Oracle 

Oracle includes three practical examples that show how prompts should manage tool use and API interaction. 

Get User Session 

The prompt instructs the agent to retrieve user identification details once at the beginning of the conversation, using the get User Session tool to retrieve the employee’s Person Number and Server Time. It then converts Server Time into a date format for reuse throughout the conversation.  

Get Unique ID 

The example uses a business object called COE_INVENTORY_BALANCE and calls an inventory on-hand balance function using item number. The prompt then extracts a unique ID from the response self-link or fallback key and encodes it for API usage.  

Pagination 

The pagination example uses a business object called COE Suppliers, where an offset parameter controls pagination. The agent checks whether the user wants to view more suppliers, calls the API with the current offset, displays results and asks if the user wants the next set. If the user says no, the agent stops and does not call the REST tool again.  

These examples show that prompt design for enterprise agents must be operational. It must manage state, avoid unnecessary tool calls, handle IDs correctly and respect user intent. 

Choice of LLMs 

Oracle’s final slide shows a choice of LLMs across providers such as OpenAI, Meta, Anthropic, Google, xAI and Cohere. The slide includes examples such as GPT-5, GPT-4.1, GPT-4o, Gemini 2.5 Flash, Gemini 2.5 Pro, Claude Sonnet 4, Claude Haiku 3.5, Grok 3, Llama and Command R. It also indicates both comes with Fusion and bring your own options.  

For enterprise customers, model choice will matter by use case. Some use cases may need reasoning strength, some may need speed, some may need multimodal capability, and others may be governed by security, cost or compliance considerations. 

What Enterprises Should Do Before Deploying AI Agents 

Before deploying AI agents in production, organizations should establish a clear readiness model. 

Key preparation areas include: 

• Identify business use cases with measurable value  
• Decide whether the use case needs a single agent, multi-agent or workflow agent  
• Review security roles and permission groups  
• Define guardrails for sensitive data and restricted topics  
• Validate available templates before building custom agents  
• Plan REST/API integration requirements  
• Prepare business object access and tool logic  
• Define testing datasets and acceptance criteria  
• Set monitoring and observability expectations  
• Train business users and administrators  
• Establish ownership for agent maintenance and change control  

AI agent success depends on governance as much as technical capability.  

How NexInfo Can Help 

NexInfo helps organizations move from AI interest to governed AI execution across Oracle Fusion Applications. 

NexInfo can support: 

• Oracle AI Agent Studio readiness assessment  
• AI use case discovery and prioritization  
• Prebuilt template review  
• Agent architecture design  
• Workflow agent design  
• Multi-agent team planning  
• Business object and REST API integration  
• Smart Actions deployment planning  
• Vector DB and RAG readiness  
• Guardrail and topic design  
• Security role and permission review  
• Testing and validation strategy  
• Monitoring and observability planning  
• User training and adoption  
• Post-deployment optimization  

With deep Oracle consulting and implementation experience, NexInfo helps enterprises build AI agents that are practical, secure, scalable and aligned with business outcomes. 

Conclusion 

Oracle Fusion AI Agent Studio is shaping the way enterprises build AI into business applications. The platform brings together templates, tools, topics, prompts, business objects, workflows, guardrails, Smart Actions, Vector DB, observability, multimodal support and LLM choice. 

The most successful organizations will not simply activate agents. They will govern them, test them, secure them and align them with real business workflows. For enterprises preparing for AI-led transformation, Oracle AI Agent Studio offers a path to move from experimentation to controlled execution. 

NexInfo helps organizations adopt Oracle Fusion AI Agent Studio with the right strategy, architecture, governance and implementation roadmap. 

FAQ
What is Oracle Fusion AI Agent Studio?

Oracle Fusion AI Agent Studio is a platform for designing, testing and deploying AI agents for Oracle Fusion Applications. It supports prebuilt templates, custom agents, multi-agent teams, workflow agents, business objects, tools, topics, prompts, knowledge stores, testing, monitoring and observability.

What is the difference between Copy Template and Use Template?

Copy Template automatically adds a suffix to all artifacts and takes the user directly to the agent team canvas. Use Template guides the user through a step-by-step configuration process for each artifact in the agent team. 

Are prebuilt workflow agents available in Oracle AI Agent Studio?

Yes. Oracle lists pre-delivered workflow agents such as CX Service Request Resolution Workflow, HCM Learning Creation Assistant, HCM Employee Replacement Workflow Assistant, SCM Quote to Purchase Requisition Assistant and SCM Component Replacement Assistant. 

Can Oracle AI agents be invoked outside Fusion Applications?

Oracle states there is currently no native support for integrations outside Fusion Applications, but agents can be invoked using REST APIs outside Fusion Applications starting 25D. Federation between Fusion Applications and non-Fusion IDCS is planned for 26A. 

Does Oracle AI Agent Studio support MCP?

MCP capability is planned for 26A. Oracle states users will be able to connect to MCP servers, authenticate in the MCP tool definition, view available tools and select which tools to include. 

Does Oracle AI Agent Studio support A2A?

Oracle states A2A will be available in basic form in 26A for Oracle agents to be called in third-party chat, with future enhancements planned to simplify the process and support calling third-party agents into Oracle agents. 

Can AI Agent Studio interpret images?

Oracle states that multimodal support, including image interpretation, will be available in release 26A. 

How does Oracle AI Agent Studio handle sensitive data?

Oracle’s roadmap includes PII detection and blocking or obfuscation, along with configurable guardrails. Until those features are available, Oracle recommends using Topics to add guard railing instructions.  

What is a workflow agent?

A workflow agent is used when a process requires accuracy, consistency, logging, auditability and repeatability. Oracle recommends workflow agents when touching systems of record and when the process must be predictable. 

What is Vector DB used for in AI Agent Studio? 

Vector DB supports semantic retrieval. Oracle describes Vector DB Writer as persisting structured JSON as semantic vectors and Vector DB Reader as retrieving semantic objects or fragments using vector similarity search. 

Can AI agents be exposed through Smart Actions?

Yes. Oracle provides a CX Service Centre example where a published agent can be exposed through Application Composer → Common Setup → Smart Actions using a UI-based action and Agent Team action type. 

How can NexInfo help with Oracle AI Agent Studio?

NexInfo can help with AI Agent Studio readiness, use case discovery, agent architecture, template assessment, workflow agent design, business object integration, REST API planning, guardrail design, security setup, testing, monitoring, observability and user adoption.