Keep a Human in the Loop

Use human in the loop to engage a person for oversight, approval, or exception handling.

 

Fulfill Your Business Goals

Goal Details How human in the loop helps

Establish accountability over agentic AI

When agentic AI must complete a high-stakes action involving risk and loss potential, your organization must establish accountability over the agent's actions.

With high-stakes actions, most organizations determine the person who is accountable for the action for the purposes of accountability and traceability. Establish this accountability and build trust in agentic AI using human in the loop.

Ensure that agentic AI operates within its assigned scope of authority

Agentic AI can behave in unpredictable ways, and you need ways to oversee it and ensure that it doesn't overstep the authority that you've granted it.

Human in the loop provides built-in oversight and straightforward ways to provide human oversight of agentic AI.

Review the plan before it's implemented

Agentic AI can write drafts, create plans, and more. However, most organizations want an employee to review the work before delivering it.

With human in the loop, employees can provide a control gate by reviewing the plan or draft that agentic AI created.

While building an AI agent, provide guidance about what to do in the event of an error, such as retrying three times and then contacting an employee. Then, when an AI agent encounters an error, it can partner with an employee if needed. The employee can provide real-time input that allows the AI agent to complete its task right away.

 

Human in the Loop Essentials

Area More information

Definition

Agentic AI can behave in unpredictable ways. When a consistent outcome is important or required, keep a human in the loop to ensure that approvals are managed consistently.

Human in the loop

Usage

When agentic AI needs to engage an employee rather than an API or other tool, keep a human in the loop. For example, an automation solution with agentic AI might need a human in the loop for the following tasks:

  • Approve a high-value task, such as when agentic AI needs to create a purchase order above a certain amount

  • Get assistance from a person, such as when agentic AI can't figure out its next steps and asks a person for help

  • Review a deliverable, such as when an agentic AI has created a marketing campaign, blog post, or customer email and needs an employee to review, update, and approve the content.

People can respond to requests using natural language. For example, after reviewing a purchase requisition: "We can get new desk chairs for the office, but the espresso machine is too much of a splurge."

The primary use of human in the loop is with agentic AI. However, you can also keep a human in the loop for automation that doesn't use agentic AI.

To see more real-world case studies, see Review Use Cases.

AI-powered capabilities

Human in the loop supports your agentic AI operations and allows you to ensure optimal outcomes.

To explore all AI features, see AI Innovation and Oracle Integration in Using Integrations in Oracle Integration 3.

Primary users

Agentic AI center of excellence. See Users of Oracle Integration.

Availability

Human in the loop is available in select Oracle Integration editions. See Oracle Integration Editions in Provisioning and Administering Oracle Integration 3.

Additionally, you must enable Process Automation in your instance to work with human in the loop. See Enable Process Automation with Oracle Integration 3 in Administering Oracle Cloud Infrastructure Process Automation.

 

Good to Know

Area More information

Deployment

Simplify your CI/CD (continuous integration and continuous delivery/deployment) pipeline by working in a project. A project is a single workspace for designing, managing, deploying, and monitoring your automation solution. A project offers fine-grained control over code promotion, including selecting the individual components and their versions to promote to higher environments. See Projects.

Error handling

When human in the loop encounters an error, it sends the reason for the error to the AI agent. You determine how to respond (for example, retry) in the AI agent's guidelines.

Additionally, all tasks that are created for keeping a human in the loop have a default expiration period that you can configure.

Monitoring and troubleshooting

Use the built-in observability features to monitor your automation solution and detect issues. See Observability.

Time-saving options

You don't have to build every automation solution from scratch. Instead, get results faster by installing a predefined automation solution. See Recipes and Accelerators.

 

Explore Your Next Steps

Goal More information

Learn about and plan automation

Compare and Plan Automation

Review Use Cases

Start building

Get Started with Human in the Loop in Using Human in the Loop in Oracle Integration 3