ElevenLabs released the alpha of ElevenAgents Architect, which creates and improves agents through voice and text. It analyzes conversation records and tests, adjusts prompts, knowledge bases, workflows, and safeguards, and evaluates proposed changes through simulations. Dashboard changes are saved as versioned drafts for users to review and deploy. It also works through external tools such as Claude Code and ChatGPT, but changes along those paths may be applied directly depending on the tool’s approval settings. Users should not assume every modification automatically waits for review.