ElevenAgents Architect screen · Image source: ElevenLabs official blog
On October 6, 2026 (October 7 in Korea Standard Time), ElevenLabs released ElevenAgents Architect. Added to ElevenAgents, the tool builds customer-facing agents through natural-language conversation, then analyzes live conversations and tests to suggest improvements. At launch it was available as an Alpha. (Official announcement)
Build and Improve Agents Through Conversation
To create a new agent, you describe the customer experience you want and attach the relevant documents. Architect configures not only the knowledge base and prompt but also the workflow, voice settings, procedures, guardrails, tools, and simulation setup. For an agent that is already running, you can describe a problem, such as failing tests or rising escalations, and ask for an improvement.
Watch the official ElevenLabs demo video (about 67 seconds)
Video source: ElevenLabs official blog
Applying It to Customer Service by Industry
The announcement gives healthcare, financial services, government, retail and e-commerce, and telecom as examples. Each organization can configure agents around its own documents and real conversation context, covering tasks such as appointments, card activation, resident services, orders and returns, and billing questions.
Healthcare agent example · Image source: ElevenLabs
Financial services agent example · Image source: ElevenLabs
Government services agent example · Image source: ElevenLabs
Retail and e-commerce agent example · Image source: ElevenLabs
Telecom agent example · Image source: ElevenLabs
Improving Resolution Rates with Conversation Analysis and Simulation
Architect finds recurring problems in ElevenAgents Spotlight analytics and customer conversation logs, then creates suggested changes and related simulations. The announcement says it can analyze thousands of conversations to find opportunities for improvement and test the proposals. Teams can review the results and the proposals dashboard before deciding on next steps.
Proposals dashboard · Image source: ElevenLabs
Changes Go Live Only After Approval
Edits are saved as versioned drafts. Owners can review proposals and roll back to an earlier version if needed, and no change is applied to the live agent until it is approved. They check the changed settings, including tools, models, turn-taking, and guardrails, before deciding whether to apply them.
Version and settings review screen · Image source: ElevenLabs
Agent Building for the Whole Team
Architect can be used by voice or chat inside ElevenAgents, and it was announced that it can also be connected to and accessed from Claude, Claude Code, ChatGPT, Cursor, and Grok Bot. Teams can set their own criteria and have Architect simulate guardrails and edge cases.
Diagram of connections to external tools · Image source: ElevenLabs
If you want to extend agent building into voice and conversational interfaces, see the overview of OpenAI Dots, an always-on agent and the self-hosted OpenDots assistant. Latency and model choices for voice support agents are covered in the overview of Microsoft MAI voice models.
Source: X @ElevenLabs