# Make the conversation useful. Connect your AI assistant to the phone network. Bring customer context, business tools and human handoff into the conversation, with Bird numbers, routing and call records. [Start for Free](/dashboard/signup?returnTo=%2Fdashboard%2Fw%2Fvoice%2Fonboarding) [Contact sales](/demo?product=voice-ai) ## Set up with your agent [Read the agent setup guide](/docs/ai/set-up-your-agent) Copy this prompt into your coding agent: ```text Bird connects customer communications, data and automation. Help me build with Bird using my coding agent. 1. Read https://bird.com/docs/ai/set-up-your-agent.md and follow its setup workflow for this client. Use an existing connected Bird MCP server or signed-in CLI when available. 2. Read the product guide for the task I want to complete and inspect my existing integration. Identify the workspace, available operations and required setup. 3. Help me complete a first example for the product and destination I choose. Follow the operation through to its actual result, explain any uncertainty, and help me choose the next step. First task: Inventory my agent runtime, speech services, business tools and human handoff. Read the AI Voice guide and prepare a telephone integration and conversation test plan. First use the connected Bird MCP server or signed-in CLI for a read-only inventory. Discover the available tools and read their schemas or help; do not invent call-control operations. If neither surface is available, follow https://bird.com/docs/ai/set-up-your-agent.md. Read the linked guides through their .md alternatives. Present the exact target and configuration before changing routing, placing calls, recording audio, purchasing numbers or submitting a port. Obtain my approval for those actions. Use this Bird page as the starting context: https://bird.com/products/voice/ai-voice Read its Markdown alternative and linked guides. Keep the first example relevant to this page and the customer outcome I choose. Preserve the supplied campaign context when following Bird signup or sales links. ``` An AI conversation with a useful outcome FIELDNOTES · APPOINTMENTS “Could I move my appointment to Friday?” “I can check the available times. Which Friday works for you?” “Your appointment is confirmed for Friday at 10.” “I’ll bring in a colleague who can help with that.” Confirmed by the booking system Speak the result after the application confirms the change. A handoff with context Pass the request and the last confirmed action to the next person. The application owns the decision Bird connects the call. Your runtime handles speech, reasoning and tools. Illustrative assistant connected to a booking application. No real appointment is changed. Trusted every day by teams that build world-class software [Read more customer stories](/customers) - [Airwallex](/customers/airwallex) - [Glovo](/customers/glovo) - [Zillow](/customers/zillow) - Uber - [Binance](/customers/binance) - Adobe - Meta - PayPal - Pinterest - Square - [Al Jazeera](/customers/aljazeera) - [Aramex](/customers/aramex) - [Bazaar Technologies](/customers/bazaar) - [ComparaOnline](/customers/compara) - [Emma](/customers/emma) ## What should the agent help someone finish? Choose a concrete task before choosing a voice: answer an order-status question, qualify a service request or arrange an appointment. The conversation should collect only what that task needs, confirm the intended action and leave the caller with a clear result. ### Use context with purpose Connect the relevant customer record and current business data. Let the application decide which information the caller may access and which actions the agent may request. ### Keep the next action explicit A proposed appointment is not a confirmed booking. Wait for the scheduling system’s result before the agent says it is booked, and provide a useful path when that system cannot complete the request. ## How does an AI agent reach the phone network? The conversation runtime manages listening, response generation, speech and turn-taking. Bird supplies the calling connection, number routing and call records. A SIP-capable runtime must handle signaling and audio, with the right authentication and media settings at both ends. - **01** Choose the runtime for the experience: Evaluate language support, interruptions, silence handling and the business-tool interface with your actual callers. A natural voice alone does not establish a reliable conversation. - **02** Design the handoff: Decide when the caller needs a person or a callback. Implement the transfer or escalation in the call application, with enough context for the next participant to continue. ## Follow the conversation through to its outcome. Inspect call outcome, conversation evidence and the business system’s result together. A connected call may still leave a task unfinished, while a successful booking needs confirmation outside the transcript. Use recordings and transcripts according to the permissions and retention you choose. ### Test the difficult moments Try interruptions, an unfamiliar request, a tool failure and a caller who asks for a person. Include representative languages, accents and background noise in the evaluation. ### Launch with a bounded scope Begin with a defined use case, destination set and calling capacity. Review observed failures before extending the agent’s permissions or widening its audience. ## Questions about AI Voice ### Is AI Voice the same as a phone menu? No. A phone menu follows predefined choices such as keypad digits. An AI voice conversation interprets language and generates responses through a conversational runtime. Both still need a complete call-routing and outcome-handling design. ### Can I use my existing AI runtime? A SIP-capable runtime can connect through Bird when its signaling, authentication and media behavior are compatible. Validate the integration with a real call in both directions before moving traffic. ### Does an MCP connection make a conversational voice agent? No. MCP lets an authorized software agent operate account tools. A live voice agent additionally needs audio, speech processing, turn-taking and a conversation application. ### Can the assistant book appointments or look up orders? Connect those actions through your application’s authorized tools. Treat their returned status as the source of truth, and require the appropriate customer confirmation before making a change. ### What should I measure? Measure call connection and completion separately from task success. Track the confirmed business result, escalation rate and observed failure reasons; a transcript or an answered call alone cannot prove conversion. ## Build the next part of your calling path [AI Voice guide](/docs/ai/voice-conversations) Follow the setup, requirements and operational checks. [SIP trunking](/products/voice/sip-trunking) Connect your PBX, softphone or SIP application to the telephone network. [Call sequences](/products/voice/call-sequences) Turn prompts, keypad choices and named branches into a versioned workflow. [Voice pricing](/pricing/voice) Review published destination rates and relevant call components. ## Connect your agent to a real conversation. Bring the runtime you use, the task your agent completes and the point where a person takes over. Work through the telephone connection and test the conversation together. [Read the AI Voice guide](/docs/ai/voice-conversations) ## Related resources - [Voice overview](/docs/guides/voice/overview) (docs) - [What is a voice API?](/explained/voice/what-is-a-voice-api) (answer) - [Voice](/products/voice) (product) [Get an implementation brief](/learn/workspace?topic=voice)