AI agents are easier to plan when teams name the role before choosing the tool. A research agent, support agent, sales agent, ops agent, and finance agent need different data, approvals, and success metrics.
Agent Role Map
| Agent role | Primary job | Human approval needed for |
|---|---|---|
| Research agent | Gather and summarize source material | Published claims and recommendations |
| Support agent | Draft replies and route tickets | Customer-facing answers |
| Sales agent | Research accounts and draft follow-ups | Pricing, commitments, contract language |
| Ops agent | Route requests and update tasks | Workflow changes and escalations |
| Finance agent | Extract invoice data and flag exceptions | Payments, tax, accounting decisions |
Data Needs
| Role | Data needed | Data to restrict |
|---|---|---|
| Research | Public sources, internal briefs | Confidential strategy unless approved |
| Support | Knowledge base, ticket history | Sensitive customer fields |
| Sales | CRM notes, meeting summaries | Pricing exceptions |
| Ops | Task boards, SOPs | Admin credentials |
| Finance | Invoices, vendors | Bank details and payment approval |
Design Rule
Start every agent with a job description, not a model name. The role determines permissions, review gates, and metrics.
How to Use This
Use this map when planning an agent portfolio or deciding which workflow should get an agent first.
Bottom Line
AI agents should be hired into clearly defined roles. Undefined agents become undefined risks.