Agent Studio
Agent Studio is where you configure everything your agent can do and how it behaves. Select Agent Studio from the top-left app switcher.
Work moves through three stages: Build the capabilities, Tune how the agent looks and sounds, then Connect it to your product and to external systems.
flowchart LR B[Build<br/>actions, task agents] T[Tune<br/>style, voice, HITL] C[Connect<br/>embed, integrations] B --> T T --> C C -- "learnings feed back" --> B
Author the capabilities the agent can run and ship them safely through the action lifecycle: write, test in the Playground, then publish.
The action lifecycle The end-to-end path: write, test, publish, and observe an action.
Authoring actions Define the capabilities the agent can run and when to use them.
Defining inputs (schema) Tell each action what data it needs from the conversation.
Task Agents Build sub-agents for complex, multi-step workflows.
Evals Grade action behavior so you can trust changes before they ship.
Publishing & versioning Promote versions and roll back when something regresses.
Playground Chat with your agent and test changes before they go live.
Shape how the agent looks, sounds, and hands off to a person.
Style & behavior Set appearance, tone, starters, and response rules.
Voice Configure speech-to-text and two-way voice chat.
Conversations (HITL) Review real chats, debug actions, and train with feedback.
Connect
Section titled “Connect”Put the agent into your product, and wire it to external tools and other agents.
Setup & embed Activate the agent and drop it into your product.
Integrations Connect third-party products: Intercom, Zendesk, Slack, HubSpot.
MCP servers Connect external data and tools over MCP.
A2A agents Delegate to and receive tasks from other AI agents.
Best practices
Section titled “Best practices” Analyzing conversations Read the quantitative what, the qualitative why, then watch the conversations behind the numbers.