AI Agent reference
This page defines what an AI Agent is, how it relates to the other building blocks, its lifecycle, and — importantly — what it does and does not decide on its own. It is a reference summary; for how to build one, see Build AI Agents.
What an AI Agent is
An AI Agent is an automation that follows one of your insurance business processes — submission intake, claim indexing, claim-to-policy comparison, claim file summarization, and so on. You build an AI Agent by defining a flow: a trigger that brings work in, followed by the nodes that classify, extract, match, route, review, and write back each case.
An AI Agent is the unit you build, publish, and operate. Everything else on this page — templates, flows, nodes, environments, runs, work items, review — describes how an AI Agent is assembled and how it behaves once live.
The AI Agent in the hierarchy
The AI Agent is both the automation you build and run (the subject of this page) and a level of the platform hierarchy: the working area inside a workspace that contains the agent's flows and related configuration. A variable scoped to an AI Agent applies to the flows in that agent. Note: the Bevaya API still refers to this hierarchy level as a project. See Core entity definitions and Environments and variables reference.
How an AI Agent is assembled
| Building block | Role in the AI Agent | Reference |
|---|---|---|
| Template | A working baseline for a common process; the starting point for a flow | Use cases |
| Flow | The node-based definition of what the agent does | Build AI Agents |
| Trigger | Starts the flow and brings in the initial input | Trigger reference |
| Nodes | The configurable steps that classify, extract, match, route, review, write back | Node catalog reference |
| Environment | The runtime context the flow runs in; resolves variables | Environments and variables reference |
| Published version | The recorded version a trigger or the API runs in production | Drafts and publishing |
| Workflow run | One execution of the agent against one case | Workflow run statuses |
| Work item | An exception or approval routed to a person | Work item statuses |
| Human review | Where a reviewer validates or corrects before the flow continues | Human review |
Lifecycle
An AI Agent moves from build to production along this path:
- Start from a template or from scratch — see Create an AI Agent.
- Configure a trigger and compose nodes — define how work comes in and how each case is processed.
- Select an environment and test the draft — run the draft against sample input and inspect each step.
- Publish — record a version that can run in production. Publishing does not start processing.
- Activate the trigger — an active trigger starts workflow runs as work arrives; runs can also be started through the Bevaya API.
- Operate — each run processes a case, may route exceptions to human review as work items, and completes or fails. Track runs in Run history and items in Item status reporting.
The flow stays an editable draft until you publish; you can revise and publish a new version at any time.
Accountability: what an AI Agent does and does not decide
An AI Agent does: read and prepare documents, classify them, extract fields, assemble context, match against your system of record, route work along the path you configured, recommend an action, and — where configured — write results back to your system of record.
An AI Agent does not make the consequential business determination on its own. Where judgment is required, the flow routes a work item to a person through a Flag for Human Review node, and a reviewer validates or corrects before the flow continues. For example, in Claim-to-Policy Comparison the coverage comparison is decision support, not a coverage determination — an adjuster approves before anything is finalized. This human-in-the-loop boundary is intrinsic to how AI Agents are built; the agent prepares and recommends, and accountable people approve.
AI intelligence is an Admin capability. The InsurGPT classification, extraction, and insights that power an AI Agent are AI-gated: building and running flows requires the Admin or Partner role, and AI intelligence is available to Admin. Reviewers work items but do not build agents or use AI intelligence. See Users and access.
Where to go next
- Build AI Agents — create, configure, test, and publish an AI Agent.
- Use cases — the templates AI Agents start from.
- Core entity definitions — how the AI Agent sits in the hierarchy.
- Human review — where people stay in the loop.
- Concepts glossary — concise definitions for every term.