Customer Experience News is a weekly newsletter about the most important news and discussions for Customer Experience and Customer Support Leaders.
This is all the weekly news you need in around 5 minutes.
Today’s headlines are from:
Zendesk Separates AI Answer Sources From Replies
Zendesk, a customer service and support platform, announced its September 2026 product release, which separates the sources behind AI generated answers into their own dedicated view rather than listing them inline within replies. The release also expands external knowledge connectors to include Salesforce and HubSpot, and moves the MCP client from beta to general availability, letting admins connect Zendesk to public and private MCP servers inside action flows.
Support Specialists reviewing an Agent generated reply can now open a separate source view instead of scanning a list embedded in the message itself, making it faster to spot which article or record produced an inaccurate answer and fix it at the source. Pairing that visibility with new Salesforce and HubSpot knowledge connectors gives teams more surface area to correct, since a growing share of what an Agent cites will come directly from CRM records rather than help center articles alone.
Teams relying heavily on Salesforce or HubSpot as reference material should audit the accuracy and structure of that data before connecting it as a knowledge source, since Agents will surface it with the same confidence as vetted help center content. The MCP client reaching general availability is worth a look for teams already experimenting with MCP servers elsewhere, though admins should treat it as a new integration surface that needs the same access review as any other connected system.
Fin Adds Incident Detection and Activity Log Exports
Intercom, a customer service platform, announced two updates to Fin, its AI agent. Fin can now export teammate and activity log data to a company’s own cloud storage on a schedule, and it can automatically detect unusual spikes in conversation volume and alert the right team through Slack or email.
When I built out our tagging and signal systems inside Intercom at Frame.io, tracking what happened outside the platform, who touched what and when, always meant stitching together exports by hand. Scheduled exports of teammate and activity log data close part of that gap, giving support leaders an audit trail they can hand to security or compliance without a manual pull. Incident Detection addresses a different problem: as Fin resolves more of the queue, fewer people are watching every conversation, so an automated spike alert catches an emerging issue before it shows up in a CSAT dip days later.
Teams should confirm where exported data lands and who has access to that storage location before turning scheduled exports on, since teammate activity logs can include sensitive detail about how individual agents are performing. Incident Detection has no thresholds to configure, which is convenient to start but means teams cannot yet tune sensitivity if the tool proves too noisy or too quiet for their volume patterns.
Hiver Introduces Skills for Repeatable AI Work
Hiver, a customer service platform built for Gmail and Hiver Omni, announced Skills, a way to teach Hiver AI how a team handles a specific request type using natural language instructions and logic. The release also included eleven smaller updates, including Slack collaboration inside conversations, account level configuration for tags and automations, domain restricted conversation sharing, and new APIs for Shared Inboxes and notes.
Skills let a team codify a process, refunds or order changes, for example, once and have Agents apply it consistently on every matching conversation instead of relying on a macro or a training document that drifts out of date. Built in simulation mode and step level approval give Support Specialists a way to test a Skill before it touches live customers, and to keep a human in the loop on the steps that carry the most risk.
Skills that take action in third party systems, HubSpot was named as an example, will need those integrations already connected and permissioned correctly, or the Skill will fail partway through a process rather than completing it. Teams new to Hiver Automations should expect some setup time translating an existing internal process document into the steps a Skill can actually execute.
Pylon Ships 115 Updates in One Week
Pylon, a B2B customer support platform, announced 115 updates in a single release, the largest of which is a beta Health Score Builder that lets teams configure their own account health signals with a transparent breakdown of each component. The release also gives Assist Agent the ability to set To, CC, and BCC recipients on issue drafts, adds task creation and file sharing inside the customer portal, and expands the API to support linking parent and child issue groups.
A configurable, transparent health score matters more than another black box metric, since teams can see exactly which signals are pulling an account’s score down and act on the specific one rather than guessing. Letting Assist Agent manage draft recipients removes a small but constant manual step for Support Specialists working account escalations that involve multiple stakeholders.
Health Score Builder is in beta, so teams should expect the available signals and scoring logic to change before general availability, and should avoid building downstream reporting on top of it until it stabilizes. With 115 updates shipped at once, teams should budget time to read the full changelog rather than relying on a summary, since smaller fixes and API changes can affect existing workflows.
Front Connects GitBook and Expands Smart QA
Front, a customer operations platform, announced its August product update, headlined by a new connector that lets Front pull from GitBook as a source for AI generated answers. The release also expands Smart QA and Smart CSAT to evaluate every teammate who touched a conversation rather than only the one who closed it, and includes mobile app improvements and proactive messaging through Autopilot Resolve.
Connecting GitBook directly means Agents can answer from documentation that already exists instead of duplicating it into a separate knowledge base, which matters for teams whose product documentation lives in GitBook and changes often. Extending Smart QA and Smart CSAT to every teammate who touched a conversation gives managers a more complete quality picture, since the person who closes a ticket is not always the person who did the most difficult part of the work.
Teams connecting GitBook should review what is actually published there before enabling it as an AI source, since internal or in progress pages could surface in a customer facing answer if permissions are not scoped correctly. Expanding QA evaluation to every contributing teammate will surface more conversations for review, so quality teams should expect their review queue to grow and may need to adjust sampling rates accordingly.
Customer.io Adds Transactional WhatsApp Messaging
Customer.io, a customer engagement platform, announced support for transactional WhatsApp messages, letting teams trigger one to one messages such as order confirmations, appointment reminders, and verification codes directly from their application. The feature extends transactional API support, previously limited to email and SMS, to WhatsApp.
For support and operations teams already using Customer.io for lifecycle messaging, transactional WhatsApp closes a gap for regions where WhatsApp is the primary channel customers expect for time sensitive updates like delivery confirmations or one time passwords. It also reduces the need for a separate WhatsApp specific tool just to handle transactional sends.
Transactional WhatsApp still depends on Meta’s template approval process, so teams should plan for template review time before they can send, the same constraint that applies to WhatsApp broadcast messaging. Teams should also confirm their WhatsApp Business Account is already approved and connected, since this feature builds on that existing setup rather than replacing it.
Linear Moves App Approvals to Enterprise Plan
Linear, an issue tracking platform, now requires the Enterprise plan to enable third party application approvals, the workspace control that lets admins review and approve requests to install outside applications. The control was previously available to admins on any paid plan.
Third party application approvals give admins a checkpoint before a workspace member connects a new app, and that checkpoint matters more as teams connect AI tools and agents directly to their Linear data. Moving the control to Enterprise narrows which teams can enforce that checkpoint by default.
Workspaces on a paid plan below Enterprise that currently rely on this control should confirm whether they retain access or will need to upgrade, and should review which third party applications and AI connections are already installed in the meantime. Teams evaluating an upgrade for this reason alone should weigh it against other Enterprise features they may or may not need.
Notion Agents Can Now Suggest Edits
Notion, a workspace and knowledge management platform, announced that agents can now propose changes instead of making them directly. A person can ask an agent to suggest edits, then move through each one from top to bottom to approve or reject it.
For knowledge base and documentation work, a suggest first mode gives a Knowledge Agent Owner or content lead a way to use Agents for line level cleanup, grammar passes are the example Notion gives, without giving up final review on what actually publishes. It is a smaller change than a new feature, but it addresses a common hesitation about letting an Agent edit a live document unsupervised.
Teams should decide which kinds of edits are appropriate for suggest mode versus direct editing, since the value of the feature depends on someone actually reviewing suggestions rather than approving them in bulk without reading. This works best paired with a clear owner for each document who is responsible for working through suggestions promptly.
Guru Locks Skills to Exact Documents
Guru, a knowledge management platform, announced Skill References, which let admins attach a specific document to a Knowledge Agent skill so the skill always uses exactly that document rather than relying on search to find it. Guru also shipped per channel response settings for its Slack agent, so admins can set different behavior in different channels.
Pinning a skill to a specific document by its ID rather than its name means a brand guide, policy document, or pricing sheet stays connected to the skill even if someone renames or moves the file, which removes a quiet failure mode where a skill silently starts pulling the wrong source. Per channel Slack settings let a team run one Agent two different ways, responding to everything in a dedicated support channel while staying to mentions only in a busier general channel.
If a referenced document is archived or deleted, the skill keeps running but flags the reference, so admins should build a habit of checking flagged skills periodically rather than assuming everything is still correctly grounded. Teams rolling out per channel Slack settings should audit which channels the Agent is currently active in before changing defaults, since existing behavior stays the same until an admin customizes it.
GitBook Agent Can Now Import Existing Docs
GitBook, a documentation platform, announced that GitBook Agent can import existing documentation directly, pointing at a live documentation website or an uploaded ZIP file of Markdown, HTML, or DOCX content and bringing the pages into a change request for review. Imports are shared across a team rather than tied to the person who started them.
Migrating documentation into a new platform has historically been a single person’s project, copying and reformatting content page by page. Letting GitBook Agent handle the import, with a panel showing progress and the option to stop a run in progress, turns that into work a team can start and hand off rather than a solo task that blocks everything else.
Teams should review imported content before merging, since an automated import can carry over formatting or structure that does not match the destination site’s conventions. Merging is blocked while an import is still writing, so teams should plan around that dependency rather than assuming they can merge partial content mid import.
Now this is:
Strategic Support for CX Leaders.
You’ve got ambitious Support targets and new metrics but you’re not sure what to prioritize first.
The list is long, the queue is getting longer, and you don’t have time to step back and think about CX strategically.
What if you could pressure-test your thinking with someone who’s spent 20 years building Customer Support operations?
No pitch, just a conversation.
