Agentic AI is shifting expectations from “help me understand” to “help me execute.” In an enterprise setting, that execution must be grounded in real data, connected to your enterprise apps, and delivered with predictable, reliable action. In practical terms, agentic AI only becomes enterprise-grade when it can:

  • Get the right data from your apps
  • Take the best action safely by invoking secure, trusted, and reliable tools
  • Follow repeatable patterns so outcomes are consistent and governable

In this blog, I want to introduce you to Oracle Integration’s new agentic AI capabilities and give you a quick overview tour of the UI via a demo video.

My friend and colleague, Kishore Katta, or KK, as he is known, has been working with a number of customers on a ‘Smart Invoice Validation and Processing Agent.’ Therefore, I am going to use that use-case as context for this blog. An agent in this scenario has the job of receiving new invoices from suppliers, validating the invoice and extracting details, comparing the terms of the invoice against the buying agreement/contract in place for that supplier, and then creating the invoice in the company ERP application. It also uses HITL for human approvals, if needed. Big shout-out to KK for building the agent and tools used in the demo. If you want to learn more about HITL and why it is a critical part of enterprise agentic AI, check out this blog.

What’s new in OIC 26.01

In OIC 26.01, you’ll see a new AI tab within an OIC project, alongside familiar concepts such as integrations and connections. The AI tab is where you build and manage agentic artifacts—specifically:

  • Tools
  • Agents
  • Agent Patterns
  • Prompt Templates

These are not abstract AI concepts—they map directly to how you operationalize agentic execution in an enterprise:

  • Tools define what can be done
  • Agents define what should be achieved
  • Patterns define how reasoning and action are orchestrated
  • Templates define how execution is initiated repeatedly at runtime

AI Tools: the enterprise “action layer” (and the most important part of your enterprise agentic journey)

What is an AI Tool in OIC?

An AI Tool in OIC is how you expose enterprise capability — implemented as an OIC integration — as an action that an agent can call. In this release, OIC allows you to use REST-based integrations as AI tools, effectively wrapping an existing integration so it can be invoked by an agent.

A key point: tools aren’t just endpoints. OIC lets you add descriptions and detailed guidelines, including guidance for each parameter, to help the agent invoke the tool correctly and consistently.

Why AI Tools matter (more than which agent framework you use)

Most organizations already have a significant investment in integration flows that encode business logic and guardrails: validations, ERP actions, enrichment, exception handling, workflow routing, and more.

AI Tools are the mechanism for unlocking that investment for agentic AI:

  • You can take an existing integration (for example, “get invoice details”) and publish it as a tool with clear semantics and usage guidance for the agent. Another example might be “create invoice in ERP”
  • You get a growing, reusable catalog of governed actions that can be called by AI systems — without rebuilding the same logic across different agent implementations.
  • OIC becomes the catalog of trusted tools available to all your enterprise agents.
  • You can sleep easy at night knowing your growing fleet of agents are not invoking your enterprise systems directly in a inconsistent (and sometimes hallucinogenic) manner.

MCP: Model Context Protocol

A defining capability in this release is that OIC tools are automatically exposed as MCP tools. When you enable the MCP server in the project settings, tools created in that project are automatically published through the built-in MCP server. Tools can be secured with Basic Auth or OAuth 2.

That matters because many customers will end up with multiple agent experiences across their organization. With MCP exposure, OIC can serve as a central, governed place to publish enterprise-grade actions that are reusable across agent platforms.

In short: build the tools once in OIC; let many agent runtimes consume them.

AI Agents: orchestration of objectives + tool use

Before we jump into agents in OIC, I want to make an important point. When talking about the technology that you use on your agentic AI journey, focusing on which agent framework to use would be a mistake. New agent frameworks are prolific. Every SaaS app has agents built-in now (Oracle has Fusion AI Studio), Oracle Cloud has two or three different paths you can go down (excellent overview of them here), and you can even build agents right in the Oracle Database (Private Agent Factory).

The point is, you will end up using many different agents. Great post here on this topic: Which Agent?.

What is an AI Agent in OIC?

An AI Agent is where you define the objectives, goals, and behavioral guidelines in natural language — that is, what you want the agent to do, in what order, and how it should behave under specific conditions. In other words, an agent in OIC is a process that has a brain (LLM) with objectives and goals, access to tools with which it can take actions towards its objectives, and a context in which it can continue thinking, acting, and reasoning, until it has achieved its goal(s).

In the demo video below, the agent is an invoice processing agent. Its guidelines include steps such as:

  • Determine whether the submitted document is a valid invoice
  • Perform a risk assessment
  • Search contract terms after risk assessment
  • If terms align, create the invoice in ERP
  • If terms do not align, create a human-in-the-loop (HITL) workflow approval task
  • If approval is received, create the invoice in ERP

This is important: the agent is not just generating a narrative summary such as a chat interface to LLMs. It is executing a multi-step business process by calling tools and then deciding on the next best action to achieve its goals.

Watch the video:

Agent Patterns: guiding reasoning for predictable outcomes

Enterprise teams need repeatability. OIC supports agent patterns—a way to dictate how an agent should reason and act.

In the current release, OIC includes a pattern based on the widely understood ReAct approach (reasoning + acting), with additional patterns planned in the coming release. Patterns help drive predictability by shaping how the agent sequences thoughts, decisions, and tool calls.

Prompt Templates: operationalizing execution with runtime parameters

OIC also introduces prompt templates, which let you standardize how an agent is invoked along with the ability to define parameters that are passed to the agent at runtime. This is crucial for production use cases where each run differs by inputs (invoice ID, document location, supplier, business unit, etc.), but you want consistent instructions and behavior.