Review scope: Sampled source-video frames and publisher description. Audio, continuous playback, transcript analysis, simulation results and live-agent behavior were not independently assessed.
About this announcement
ElevenLabs presents ElevenAgents Architect as a conversational way to build and improve AI agents. The publisher says it can analyze transcripts, suggest changes and prepare new agents from a description and supporting files. It also says changes arrive as versioned drafts for review and don't go live without approval, an important part of the workflow rather than a footnote.
The sampled frames first show a question about customers asking to speak to a person. A response identifies a possible account-lookup problem and proposes a new procedure, followed by visible Reject and Approve controls. Later, a separate prompt asks for an agent that can set appointments and answer questions, with documents attached for context. Agent settings then appear on screen.
These are examples from the publisher's video. We didn't test transcript analysis, the suggested fix, simulation results or a live deployment. This review covers sampled frames and the description, without assessing audio or continuous playback.
Notes from the source video
The support example starts with a question someone managing an agent might actually ask: why customers request a person. That gives the response a purpose before any broader claim about optimization. If you're explaining an AI assistant, choose a question tied to a decision the viewer recognizes rather than beginning with a list of capabilities.
The response includes a numerical explanation about failed account lookups and a proposed procedure. The useful visual structure is the connection between the stated problem and the suggested change. The number is part of the promotional example, not an independently verified customer result, so it shouldn't become a performance claim in a review.
A close view makes the Reject and Approve controls prominent. That establishes a decision point after the proposal and helps avoid the impression that a generated recommendation is automatically a production change. For a feature that can affect customer conversations, showing where the user reviews the work is part of explaining how the product is meant to be used.
The new-agent example uses a different colored background and a prompt with document attachments. The visible brief asks for appointment setting and answers based on shared documents; subsequent frames show agent-language and model settings. That sequence connects an initial request with configuration, while stopping short of proving the resulting agent handles those tasks correctly.
Between the interface examples, the sampled frames include people, small interface panels and abstract colored forms. Those scenes broaden the presentation, but the concrete explanation comes from the prompts, proposed procedure and review controls. When adapting the approach, keep those readable steps central so the surrounding visuals don't carry claims the interface hasn't explained.
Design takeaways
- Start with a specific question about an existing workflow, then connect the proposed change to that question.
- Show the review decision clearly when generated changes require approval before use.
- Separate improving an existing agent from creating a new one, and give each example its own understandable input.
- Show attached context and resulting configuration without treating that sequence as proof of correct deployed behavior.
- Keep example statistics attributed to the demonstration unless there is separate evidence for a customer outcome.
Related examples and guides
When you're explaining an AI feature, viewers need to understand what the system proposes and what someone still needs to review. These guides cover planning that demonstration and checking that the finished video represents the product accurately.
Plan your own product video
Start with a product video script, then use the demo guide to work out what you need to show.
Want a video like this for your product?
Use your coding agent to create a video draft, edit its source, and request an MP4 export with Impractical.








