---
title: "Voice Agents are here. Context turns them into real customer value."
description: "Voice agents are now technically feasible because synthetic voices sound natural, speech-to-text is real-time, and agents can keep pace with human conversational timing. However, most pilots prove an agent can hold a conversation without proving it can be trusted with a real customer. The missing element is context: knowing who is calling, what is already in motion, and what the agent is allowed to do. Databricks provides this brain by combining customer context, an ontology of business relationships, and governance rules that already live on its platform. The agent uses fast decision models (System 1) for millisecond choices and large language models (System 2) for careful reasoning, with every decision recorded for traceability and improvement."
---

# Voice Agents are here. Context turns them into real customer value.

[Databricks](https://yomu.fyi/company/databricks) · Sathish Gangichetty, Rohit Bhagwat, Usman Zubair · Oct 28, 2026

**Type:** Explainer

## Summary

Voice agents are now technically feasible because synthetic voices sound natural, speech-to-text is real-time, and agents can keep pace with human conversational timing. However, most pilots prove an agent can hold a conversation without proving it can be trusted with a real customer. The missing element is context: knowing who is calling, what is already in motion, and what the agent is allowed to do. Databricks provides this brain by combining customer context, an ontology of business relationships, and governance rules that already live on its platform. The agent uses fast decision models (System 1) for millisecond choices and large language models (System 2) for careful reasoning, with every decision recorded for traceability and improvement.

## Context

Voice is often the fastest or only way customers can reach a business in moments like a driver moving a dock appointment, a drive-thru ordering, or stranded travelers calling a hotel. Historically, voice was too hard to get right, but synthetic voices now sound natural, speech-to-text is real-time, and agents can keep pace with human conversational timing (about 200 milliseconds). However, most enterprise AI investment has focused inward on employee assistants, missing the larger opportunity in customer-facing conversations. Databricks positions itself as the platform that provides the brain—context, ontology, and governance—that turns a voice agent from a generic conversationalist into one that can be trusted with real customer interactions.

## Approach / What changed

Databricks provides the ‘brain’ for voice agents by combining context (customer, order, and policy data), ontology (definitions and relationships between data), and governance (access rules) that already exist on its platform. The agent uses a two-tier AI system: fast decision models (System 1, such as Jev from TypeSafe AI) for millisecond-level choices and large language models (System 2) for careful reasoning. Databricks’ ai\_decide enables quick decisions with confidence scores, and all actions are recorded for traceability and improvement. The open-source Databricks Intelligent Voice Agents (DIVA) blueprint on GitHub provides a working starting point for teams.

## Takeaways

- Voice agents require more than natural speech; they need real-time access to customer context, order data, and business rules to be trusted with actual customer interactions.
- Databricks provides the brain for voice agents via a two-tier AI system: fast decision models (System 1, e.g., Jev from TypeSafe AI) for millisecond decisions and large language models (System 2) for careful reasoning.
- Every decision a voice agent makes on Databricks is recorded for traceability, quality scoring, and continuous improvement, enabling teams to analyze call patterns in natural language.

## FAQ

### How does the voice agent decide which model to use during a call?

The voice agent uses a two-tier system. Fast decision models (System 1), such as Jev from TypeSafe AI, make millisecond-level decisions by using Databricks' ai\_decide on governed data. Large language models (System 2) handle careful reasoning, such as working out the right fix for a delayed order. The two work side by side on every turn of the conversation.

### What is the recommended first step for implementing a voice agent?

Databricks recommends starting with a 90-day pilot covering one channel, one team, and the highest-volume routine call (e.g., order status or driver check-in). Before the pilot, record today's cost per call, handle time, customer satisfaction, and whether callers return. After the pilot, compare results and then extend the same governed setup to more locations and call types.

**Tags:** [AI Agents](https://yomu.fyi/topic/ai-agents), [Context Engine Optimization](https://yomu.fyi/topic/context-engine-optimization), [Customer Service](https://yomu.fyi/topic/customer-service), [Databricks](https://yomu.fyi/topic/databricks), [Voice AI](https://yomu.fyi/topic/voice-ai)

- Source: [Databricks](https://www.databricks.com/blog/voice-agents-are-here-context-turns-them-real-customer-value)
- Source URL: https://www.databricks.com/blog/voice-agents-are-here-context-turns-them-real-customer-value
- Ingested by Yomu: 2026-10-28T21:00:12.611Z

[Read original post](https://www.databricks.com/blog/voice-agents-are-here-context-turns-them-real-customer-value)
