Agent
Agent Overview
Agents are AI employees that own evaluatable work. Configure each agent with the instructions, playbooks, tools, and memory it needs to complete tasks consistently.
How agents fit together
An agent combines identity, access, deployment paths, and feedback loops. Start by defining what kind of agent it is, then connect the knowledge and systems it needs, choose where it runs, and improve it with real results.
| Area | Includes | Use it to |
|---|---|---|
| Define its identity | Playbooks, instructions, skills | Set the agent's role, expertise, rules, tone, and operating style. |
| Connect knowledge and systems | Tools, browser, memory | Give the agent access to integrations, external information, and reusable context. |
| Choose where it runs | Surfaces, orchestrations, triggers | Decide where people and systems can reach the agent, and what starts its work. |
| Improve over time | Tasks, benchmarks, evaluations | Review real work, test expected behavior, and measure quality as the agent changes. |
Start here
Playbooks
Give agents durable SOPs, examples, review criteria, and process rules.
Instructions
Set the global behavior, tone, boundaries, and default process for the agent.
Skills
Install reusable expertise, examples, and files for specialized work.
Tools
Connect integrations, MCP servers, and custom functions the agent can call.
Surfaces
Choose where people can reach the agent, from chat to Slack, phone, and meetings.
Memory
Give the agent structured data it can reuse across tasks.