# Build AI agents around bounded business jobs

Design and implement AI assistants, agents and multi-step workflows with approved data, scoped tools, measurable tests and human review.

Canonical: https://aroai.com/solutions/agent-development/

## A useful agent has a defined job

Agree the user, input, source systems, permitted tools, expected output, error path and decisions that require a person. Build the narrow workflow first and keep a record of how it performs.

## From prototype to supported workflow

1. **Define the task** Select a repetitive workflow with a clear owner and measurable starting point.
2. **Prepare sources and access** Confirm data rights, identity, scope and system permissions before connecting tools.
3. **Build and test** Create the prompt, tool sequence and approval points; test representative good, ambiguous and failure cases.
4. **Train and hand over** Document how to use the agent, how to monitor it and who can stop or change it.

## What may be included

- **AI assistants and copilots**: Help a person retrieve context and prepare a next step.
- **Workflow agents**: Sequence supported tools with clear permission and confirmation boundaries.
- **Multi-agent prototypes**: Evaluate orchestration only when separate roles, hand-offs and failure modes add measurable value.