AI Workflow Automation Oracle Fusion Integration and Coupa

AI workflow automation in Oracle Fusion Cloud Applications is built around Oracle AI Agent Studio, a native environment for building and deploying AI agents and workflows that execute multi-step processes across ERP, HCM, SCM, and CX. Workflow agents run a fixed sequence of steps, while hierarchical agents reason through open-ended work. Oracle Integration connects both to outside systems through Model Context Protocol (MCP) tools and prepackaged adapters.

TL;DR

• Oracle AI Agent Studio is the native place to build and deploy AI agents inside Oracle Fusion Cloud Applications.
• Workflow agents follow a fixed node sequence, while hierarchical agents handle open-ended problems through LLM reasoning.
• Oracle Integration exposes integrations as MCP tools, and external apps call agent teams through InvokeAsync.
• Around Coupa, the native layer is Coupa Compose (Navi Agent Studio, Smart Intake and Orchestration, and Navi Connect), with Rossum, Hypatos, Workato, Tray.ai, and Make extending it.
• Approvals and exceptions that cross Fusion, Coupa, and other systems need one owner and one audit record, which is where a governed layer such as Kissflow fits.

If Oracle Fusion runs your finance and supply chain and Coupa runs your spend, each platform automates its own half well. Each half also keeps its own audit trail. The invoice exception, the onboarding request that touches HR, IT, and finance, and the supplier change that needs legal review all live between the two, and that is where audit gaps tend to appear. This page covers the native agent layer in Oracle Fusion, how Oracle Integration and MCP connect it outward, the Coupa equivalent, and a decision table for who owns the rule and the record once a process leaves either system.

What are the core components of AI workflow automation in Oracle Fusion?

Oracle AI Agent Studio is the component everything else attaches to. It is a design-time environment inside Oracle Fusion Cloud Applications.

• Oracle AI Agent Studio: Teams build, test, and deploy AI agents and workflow agents here, without leaving Oracle Fusion Cloud Applications.
• Workflow agents: These agents run a predetermined chain of nodes, such as extracting data, calling a business object, running an LLM, or sending an email.
• Hierarchical (supervisor) agents: A supervisor agent directs other agents, and the LLM's reasoning sets the order of execution.
• Agent teams: Agents are grouped into teams that work a business process together, and external applications start them through an API.
• Pre-built agents and partner-built agents: Oracle ships pre-built agents, and partners publish their own in the AI Agent Marketplace, which Oracle launched in October 2025.
• Cost: Oracle includes AI Agent Studio with a Fusion Applications subscription.
• Limitation: Agents built in AI Agent Studio act on Fusion data and on whatever Oracle Integration exposes to them.

How do workflow agents differ from hierarchical agents?

Workflow agents fit structured enterprise processes where the rules, policies, and service-level agreements are already known. Each run follows the same path, so the result is predictable and easy to explain to an auditor.

Hierarchical agents fit open-ended problems where the next step depends on what the agent finds. The supervisor decides the order at run time. That flexibility suits investigation work and is harder to certify step by step.

Workflow agents chain steps, run steps in parallel, switch between branches, and loop, using the Workflow Control nodes in the table below.

Which nodes does a workflow agent use?

Oracle groups workflow agent nodes into five categories.

Oracle also lists a Code node that runs JavaScript for custom logic inside a workflow.

How does Oracle Fusion integration work for AI agents?

Oracle Integration (OIC) is the bridge between Fusion agents and everything outside Fusion. In Oracle Integration, an integration can be registered as a tool and its project exposed as an MCP server, with its own server URL. Oracle Integration also offers prepackaged adapters, along with REST, SOAP, file, and FTP adapters.

• Native Fusion actions: Agents call Fusion business objects directly, which keeps in-Fusion work inside Oracle's security model.
• Oracle Integration MCP tools: An Oracle Integration flow is published as an MCP tool, so an agent reaches an outside system without anyone building a separate REST wrapper.
• External MCP servers: AI Agent Studio also connects agents to MCP servers that other vendors run.
• InvokeAsync: External applications start a Fusion agent team with an "invoke then poll" pattern: the call returns a job ID, and the caller polls a status endpoint until it reports COMPLETE.

MCP here always means Model Context Protocol, the open standard for connecting AI agents to tools.

How do you connect an external system to an Oracle Fusion agent?

The path runs through Oracle Integration first and AI Agent Studio second. These steps follow Oracle's documentation for using an integration as an MCP tool.

1. Build an Oracle Integration flow that calls the external system through a prepackaged adapter or a REST, SOAP, file, or FTP adapter.
2. Register the integration as a tool, and turn on MCP for its project in Oracle Integration.
3. In AI Agent Studio, open the Tools tab and add a tool of type MCP.
4. Enter the project's MCP server URL, with the Streamable HTTP transport and client credentials authentication.
5. Add the tool to an agent or a workflow agent node, then test the agent against sample requests.
6. Publish the agent once test results match the expected outcome.

Step four is the one to plan around. Oracle's A-Team notes that the Oracle Integration MCP connection in AI Agent Studio does not support identity propagation, so the agent calls the integration as the client application, and the integration cannot apply the permissions of the person who asked.

AI workflow automation with Coupa integration: what runs natively and what connects

AI workflow automation for Coupa combines native orchestration in Coupa Compose and Navi with third-party tools for invoice intake, forms, and ERP sync. The structure matches Oracle's: a built-in agent layer sits at the center, and partners fill specific jobs around it.

What does Coupa automate natively?

Coupa Compose is the native answer: a set of three products Coupa announced at Coupa Inspire 2026.

• Smart Intake and Orchestration: The entry point for intake requests, which orchestrates fulfillment through agents and systems and brings in a person where needed. It is powered by Tonkean, which Coupa acquired.
• Navi agents: Navi is Coupa's built-in agent layer, which runs agent tasks across the Coupa platform.
• Navi Agent Studio: A no-code environment for building custom agents in natural language, which Coupa calls digital teammates, with role-based access settings.
• Navi Connect: Coupa's integration layer for third-party agents and systems, built on Model Context Protocol (MCP) with more than 30 out-of-the-box tools. Coupa scheduled MCP availability for its September 2026 release.

Which specialist and iPaaS tools connect to Coupa?

Specialist tools take the jobs that Coupa's native layer hands off: reading invoices, taking in requests, and syncing records to an ERP.

• Rossum: Coupa acquired Rossum, its document-processing partner since 2024, at Coupa Inspire 2026. Rossum reads supplier invoices and pushes clean data into Coupa.
• T-LLM: Rossum's model is a transactional large language model.
• Hypatos: Hypatos runs Coupa-certified AI agents for invoice processing, with a two-way data exchange with Coupa.
• Tonkean: Tonkean, now part of Coupa, automates intake and approvals and brings more than 250 native connectors.
• Workato: Workato's Coupa connector syncs Coupa data with ERP systems across procure-to-pay and expense processes. Each sync is a recipe that someone builds and maintains.
• Tray.ai: Tray.ai offers a Coupa connector for syncing records with other systems.
• Make: Make's verified Coupa app, available on its Enterprise plan, handles requisitions and approvals, and it carries the same build-and-maintain load.

Ask each of these vendors where the record of an exception that spans Coupa and the ERP is kept. If none keeps it, that record falls to whoever designs the handoff.

When the process leaves Fusion or Coupa, who governs approvals and exceptions?

Native agents stop at the edge of their own platform. Consider an invoice that Rossum captured and Coupa matched, but whose receipt sits in Oracle Fusion SCM and whose budget owner approves by email. That invoice now has three partial audit trails and no single owner. For a CIO who has to answer for audit findings, that gap matters more than any agent count.

Where does each rule and audit record live?

Straightforward system-to-system sync fits an iPaaS. The last row covers the case where a person has to decide, the decision spans two or more systems, and the evidence has to survive an audit.

How does Kissflow fit around Oracle Fusion and Coupa?

Kissflow is an AI-powered enterprise application platform where business teams and IT build, automate, and govern their core work, driven by business logic, not code. Around Oracle Fusion and Coupa, it is the governed execution and exception layer: it takes the structured work that falls at the edges of what those core systems were built to do, and it never replaces Oracle, Coupa, SAP, Salesforce, or Workday.

• AI Builder: Generally available as of May 2026. A single prompt or an uploaded requirements document generates a complete working app, refined through conversation.
• Blueprints: The AI produces blueprints: structured metadata that describes business logic. Every artifact is a standard Kissflow component a developer can open and read.
• One audit log: AI changes use the same audit log as manual edits, every action is timestamped and attributed, and a human stays in the lead on every rule. An admin controls what goes live.
• Placement and hosting: Kissflow runs in the business-process layer, segmented from core systems, on Google Cloud Platform (primary) and AWS (secondary), with data residency in the US, EU, APAC, and Oceania.
• Certifications and uptime: SOC 1 Type II, SOC 2 Type II, SOC 3, ISO/IEC 27001, HIPAA, GDPR, and CCPA, with a 99 percent uptime SLA.
• In production: New store opening and retail lifecycle management, oil rig shutdown and turnaround management, and end-to-end procurement and financial consoles.

See how AI Builder builds a cross-system approval app with one audit log.

Who has evaluated Kissflow?

Kissflow was founded in 2012 and serves over 1,200 customers worldwide. It was included in the Forrester AppGen and Low-Code Platforms Landscape, Q2 2026, and included in the Gartner Hype Cycle for Enterprise Process Automation, 2025. It was named a Strong Performer in The Forrester Wave, Q1 2024, and named an Established Vendor and Customers' Choice in Gartner Peer Insights 2024.

What should a governance and audit checklist cover for cross-system AI workflows?

Every cross-system workflow needs a named owner before it needs an agent. Use these checks before a Fusion or Coupa agent hands work to another system.

• Name one owner for each approval rule, and record where that rule lives.
• Confirm that every agent action and every human decision lands in an audit record that an auditor reads without reconstructing three logs.
• Check whether each agent acts as the requesting user or as a broad service account, and record which one.
• Test every rule change before it reaches live approvals.
• Route exceptions to a person with authority to decide, and record the reason the person gives.
• Accept that some limits remain: no checklist fixes a source system that stores its own approvals outside any shared log.

What are the use cases for AI workflow automation by Oracle Fusion module?

Native agents handle the in-module step, and the cross-system extension is where exceptions appear.

ERP and finance

Agents in Fusion ERP take an invoice to a requisition match and route finance approvals. The extension case is an invoice exception that Coupa flags and Fusion has to resolve, with one record of who cleared it.

Procurement and SCM

Agents move work from supplier quote to requisition inside Fusion SCM or Coupa. The extension case is a procurement and vendor management approval that needs legal and finance sign-off outside both platforms.

HCM

Fusion HCM agents handle onboarding steps inside HR. The extension case is onboarding that needs IT accounts and a finance cost center, where exception routing between HR, IT, and finance tends to break.

CX and service requests

Agents triage service requests inside Fusion CX. The extension case is a request that needs an IT action and a manager approval in another system before it closes.

What's new in Oracle Fusion AI automation in 2026?

Oracle introduced Fusion Agentic Applications on 24 March 2026: 22 applications built into Oracle Fusion Cloud Applications, each run by coordinated teams of AI agents and supported by Oracle AI Agent Studio. The same announcement added an Agentic Applications Builder inside AI Agent Studio, for building agentic applications from Oracle, partner, and external agents.

Which workflow should you start with?

Pick one cell from each line below. The combination tells you whether native agents cover the work or whether it crosses systems.

• Module: Choose ERP, HCM, SCM, or CX, and start where the most approvals still travel by email.
• Build type: Choose a pre-built agent when Oracle or Coupa already ships one, and build a custom agent only for a process neither covers.
• Workflow: Choose AP, intake, supplier management, or ERP sync. If the work is pure sync, an iPaaS fits. If a person decides across systems, it needs a governed approval layer.

Frequently asked questions

Why do AI workflows need Human Approval nodes?

A Human Approval node puts a named person into a workflow agent wherever a decision needs authority. In Oracle AI Agent Studio, the node pauses the workflow, sends the reviewer a message by chat or email, and continues down an approve or reject branch. That branch is also where the reviewer's reason should be recorded, so the audit trail shows why, along with who and when.

Where this leaves your Oracle Fusion and Coupa roadmap

Oracle AI Agent Studio and Coupa Compose have settled the question of where agents live. Fight neither of them. Build inside each platform first, connect them through Oracle Integration and MCP, and use an iPaaS where the work is pure data movement. The harder question is ownership. Once an approval or exception crosses Fusion, Coupa, and an email inbox, no native agent holds the rule or the record. The view here is that governance belongs in the design before the first cross-system agent goes live, because retrofitting it after an audit finding tends to cost more. Your next step is to list your top three cross-system exceptions and fill in the ownership table above for each one.

Book a demo to see how Kissflow gives cross-system approvals between Oracle Fusion and Coupa one owner and one audit log.