Subscribe to release notes

Unified Reasoning Engine

A major upgrade to the AI architecture is rolling out over the course of February. The unified Reasoning Engine replaces multiple channel-specific systems with a single, shared intelligence layer across Voice, Messaging, and Email.

What’s new

• Faster FAQ responses: Simple questions and chitchat in Messaging now receive faster responses.
• Improved language detection: More accurate language detection in Messaging and Email channels.
• Multi-message support: Messaging users can send multiple messages in succession and receive a single, coherent reply.
• Coaching on Voice: Coaching is now applied to Voice conversations.
• Playbooks on Voice: Playbooks can now run in Voice conversations. Performance optimizations for Voice are ongoing—talk to your Customer Solutions Consultant about the right timing for your use cases.

Rollout timeline

The unified Reasoning Engine will roll out gradually throughout February 2026, starting with 1% of customer conversations the week of February 2. No action is required—the upgrade will be applied automatically.

Preparing for the rollout

Voice customers: Review your existing Playbooks and Coaching items. If you don’t want certain Playbooks or Coaching available in Voice, add an availability rule (channel ≠ Voice). If you’re planning to deploy Playbooks on Voice, a phased approach is recommended—your CSC can help you evaluate which Playbooks are ready now and which may benefit from upcoming performance improvements.

Why this matters

This foundational upgrade delivers more consistent AI behavior across all channels, reduces response latency for straightforward inquiries, and enables Voice customers to benefit from Coaching and Playbooks for the first time.

Same-thread email replies for Zendesk Ticketing handoffs

With the SMTP Connector configured, human agent replies to email conversations escalated via the Zendesk Ticketing block can now continue in the same email thread as the original end-user conversation. This provides a more consistent experience for end users and reduces confusion caused by broken or split threads.

Requirements:

• The SMTP Connector must be configured for this behavior.
• Without the SMTP Connector, human agents must use a different email address than the AI Agent’s BYOD address, and replies are sent in a separate thread.
• If using the Ticket Recipient field with a different email address than the AI Agent’s BYOD address, SMTP must be configured for both addresses. Both addresses must be on the same domain.

Learn more about configuring Zendesk Ticketing and the SMTP Connector.

Option to disable Multiple participant conversations in Email

Added a toggle to disable multiple participant conversations in the Email channel settings.

By default, the AI Agent replies to all participants in an email thread and accepts messages from anyone included in the To or CC fields. With this update, AI Managers can now opt out of this behaviour:

• Toggle ON (default): The Agent replies to all participants in the same thread, and everyone has visibility into the conversation.
• Toggle OFF: The Agent replies only to the original sender. If another participant replies, the Agent starts a separate conversation with them.

To configure this setting, go to Channels > Email > Customization and locate the Include CC’d participants in responses toggle.

Introducing Linked Playbooks

Playbooks can now call other Playbooks to create modular, reusable workflows. Call any child Playbook from the Playbook editor with the @ option. Linked Playbooks always return to the parent Playbook when complete, enabling shared steps like authentication or eligibility checks without duplication.

Learn more about Linked Playbooks in Playbook management.

Connect Ada to AI assistants with MCP Server

You can now connect Ada to AI assistants like Claude Desktop and ChatGPT using the Model Context Protocol (MCP). This lets you query and analyze your AI agent data conversationally—ask questions like “Can you review my low CSAT convos from last week and let me know where I should focus my improvement efforts?” and get actionable insights without navigating dashboards.

What you can do

• Analyze hundreds of conversations at once to identify root causes and themes
• Get improvement recommendations based on conversation patterns
• Create visualizations like trend charts and Sankey diagrams

See our prompt library for more examples.

Getting started

MCP Server supports OAuth (recommended) or API key authentication, and works with any MCP-compatible assistant.

Learn more in our MCP Server documentation.

Email CCs for Zendesk Handoffs

An optional Email CCs input field is now available on the Zendesk Ticketing block. This allows CC’d recipients from an email conversation to be passed directly into a Zendesk ticket at handoff, without requiring a custom field in Zendesk.

When configured with the email_latest_CC_list variable, all CC’d participants are automatically included on the ticket, ensuring they receive replies from human agents after handoff.

Learn more about configuring email handoffs in Zendesk in the Zendesk Ticketing block documentation and Email handoffs.

Voice conversation display name improvement

Ada now better identifies callers in the Conversations page of your Ada dashboard by using the phone_number metavariable as the user’s display name for Voice conversations.

Previously, Voice conversations often appeared as “Anonymous”. Now, if a user’s name or display name is not explicitly set, Ada will use the phone_number variable before falling back to “Anonymous.”

This ensures your team has immediate visibility into who is calling without needing to open the conversation details.

Web Import: Automatically synced, always up to date knowledge

Keeping your AI agent informed just got easier. Web Import now captures your website content in minutes and automatically syncs it every day, so your Ada articles stay accurate without any manual effort. With a faster, more reliable crawler and up to 86% more content coverage—including dynamic pages—you can trust your knowledge base to reflect the latest version of your site.

The upgraded Web Import is designed to work seamlessly with your existing workflows. Article updates no longer break Coaching or Rules, ensuring stability across your setup while keeping content fresh. No setup, no notifications, and no manual refreshes—just hands-free, high-quality content for your AI agent.

Learn more about Web Import.

API Usage Report CSV export now contains valid JSON

The API Usage Report CSV export now outputs properly formatted JSON in the Request and Response columns.

Previously, JSON data in these columns was malformed due to missing commas between elements, which prevented customers from parsing and ingesting the data into their systems. The CSV export now uses standard RFC 4180 escaping, ensuring the JSON is valid and can be parsed by any standard JSON parser.

Learn more about the API Usage Report.

Actions Usage Report renamed to API Usage Report

The Actions Usage Report has been renamed to API Usage Report and now includes HTTP Request block tracking alongside Actions.

Going forward, HTTP Request blocks appear automatically in the report as HTTP Request: {METHOD} {URL}.

Learn more about the API Usage Report and HTTP Request blocks.