Layer 2 — AI Agents · Chapter 3

Major Types of Agents

Book chapterLayer 2

Agents can be classified by the environment in which they work and the kind of responsibility they assume.

Agent typeTypical workMain challenge
Research agentsSearch, compare sources, summarize evidence, and prepare reportsSource quality and factual verification
Software agentsWrite, test, debug, and deploy codeRepository context and safe execution
Operations agentsUpdate records, route work, coordinate vendors, and monitor exceptionsIntegration with legacy systems
Customer-service agentsAnswer questions, resolve cases, issue credits, and escalate disputesPolicy compliance and customer trust
Personal agentsSchedule, shop, organize information, and manage communicationsPrivacy, identity, and permission
Robotic agentsNavigate and manipulate the physical worldSafety and unpredictable environments

Horizontal and vertical agents

A horizontal agent works across many industries, such as a general coding or research agent. A vertical agent is designed around one profession or workflow, such as insurance claims, freight dispatch, or medical billing.

Copilots versus autonomous agents

Copilots remain under direct human control. Autonomous agents are responsible for completing more of the workflow. Many businesses will move gradually from copilots to bounded autonomy rather than making a single leap.