Generate with AI

Status: Current | Last reviewed: 2026-10-06

Describe the agent or workflow you want, and an AI assistant drafts it for you. You review and edit the result like any other draft.

Generation can inspect existing agents, tools and detailed capability guidance before planning. It can also research public sources and, when the Library connection is configured, learn from previous demo videos and transcripts. It should explain relevant examples and defaults, and ask about consequential business choices instead of requiring you to supply technical field names. A Voice demo brief stays focused on one voice agent. Routine call ending and summaries belong in its prompt, while protected member information calls for native Customer Authentication with supported defaults. It asks about unresolved business decisions, not routine authentication options or phone-system ownership.

For scenario-based requests, the assistant has shared guidance for voice, Copilot, supervisor and visual experiences, including corrections, pending approvals and spoken wrap-up. It can plan relevant checks and define exactly what a latency target measures. A proposed test plan is not an executed test: review the saved draft and run the relevant Preview or Performance checks.

The default OpenAI authoring setup is GPT-6 Astra with low reasoning effort, selected from the generation benchmark. This also applies to new Studio designs and editor Assistant chats. It controls the AI building your draft; the models used by the finished agent follow its own configuration.

Don't see the option? Generate with AI must be turned on for your tenant (see Enabling it) and the deployment must have the feature enabled. If either is missing, the option does not appear.

Generating an agent

  1. Open Agents → New agent and pick the agent type (Voice, Chat, Copilot, or Translation) as usual.
  2. On the "How would you like to start?" step, choose Generate with AI. (For Translation agents this step only appears when the feature is enabled.)
  3. Give it a name, an optional description, and describe what you want in the large text box. Detail pays off:
  4. Who does it talk to, and about what?
  5. What should it do? What should it refuse or hand off?
  6. Tone and language(s).
  7. Systems it needs, such as booking, CRM, or order lookup.
  8. Click Start Generating.

The agent appears with a Generating… badge. Opening it shows the generation view instead of the editor: the AI's plan, a live progress log, and your original request.

Generating a workflow

Use the same path for workflows: Workflows → New workflow → Generate with AI. Add a name, description, and flow brief: what starts it, which steps run, who approves what, and which messages go out. Workflows can take more technical briefs than agents.

Answering the AI's questions

If the AI needs a decision from you, such as whether the agent should book appointments or only check availability, it pauses and asks. You'll see:

Answer them and click Send Answers. Generation resumes where it left off. You can leave the page at any time. The bell tells you when you're needed or when the result is ready.

When it finishes

You get a "ready" notification. Opening the agent/workflow now shows the normal editor with everything the AI built: pipeline stages or workflow nodes, prompts, and selected tools. Review it, tweak anything, test it, then publish it when ready. Nothing publishes on its own; the AI writes a draft.

If generation fails, the view shows a plain-language reason (for example, a needed integration isn't configured). The draft keeps anything saved before the failure. You can edit it by hand or delete it and try again. Cancel works the same way: the run stops and the draft stays editable.

Enabling it (admins)

  1. Make sure an integration for the AI engine exists under Settings → Integrations → Services (for Claude Code: an Anthropic API key, or AWS Bedrock / Google Vertex credentials; for OpenAI Agents API: an OpenAI API key with Agents read/write and Responses write permissions).
  2. Open Settings → User Experience → AI Generation.
  3. Toggle Enable Generate with AI, pick Claude Code or OpenAI Agents API under AI engine, select matching credentials, and save.

The option appears for everyone in your tenant who can create agents or workflows.

The same selection powers new editor Assistant chats and Design conversations. Existing conversations keep their original engine. Start a new chat/design to try the new selection; you can keep the original draft for comparison. The engine that authors a draft is independent of the models used by that agent.

OpenAI retains authoring conversations in the United States until deleted and does not support Zero Data Retention. Choose a connection appropriate for the information you include in briefs, references and diagnostic conversations. Closing or expiring an authoring session requests deletion of its OpenAI session; your saved TrueServe draft and conversation history remain.

Tips for good results