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 10 minutes.
Our main story today is about customers sending their own Agents to contact support.
Today’s headlines are from:
Customers are sending Agents to contact support
Customers have begun sending their own Agents to contact companies. A survey found that 6 in 10 already handle AI Agents that reach out autonomously for consumers or partners. Lorikeet, an AI support platform, reports that up to 3% of the inbound support traffic it sees comes from AI agents and other automated systems, based on its own data and definition. These Agents open chats on company websites and place phone calls, so the customer skips the hold time and the information gathering. Google’s Gemini Call for Me, which places calls to businesses for the user, is in limited testing.
Support teams now have a routing decision with no single right answer.
The options below are the ones vendors and practitioners are discussing.
The right mix will depend on each company’s risk profile, volume, and channels.
Accommodate.
Treat the contact as a customer contact and serve it normally. NiCE has described exploring accommodation as its direction for these interactions.
Observe and tag.
Lorikeet runs detection in observation mode first, tags suspected Agent contacts, and plans to report them as a separate line in metrics. TTEC also recommends flagging likely Agent calls and tracking them separately, so CSAT and handle time reflect human customers.
Ask at the start.
A routing question such as “Are you contacting us for yourself or on someone’s behalf?” sets up a path for each answer. A literal “Are you human?” has limits. A customer-side Agent can answer either way, and a person who uses text-to-speech or pastes AI-drafted text may be misclassified.
Tier by risk.
Allow low-risk requests such as policy lookups or reservations, and require direct customer confirmation for larger refunds or account closures. Commentary on the draft Personal Agent Protocol from Meta and Sierra points to this kind of policy.
Build an Agent lane.
The draft protocol lets a business choose whether Agents use its website, its APIs, or its own conversational Agent. TTEC says some firms may build a separate verified channel, and Alorica has suggested registering Agents through a service such as ID.me.
Restrict.
Add authentication steps for callers, or stop accepting Agent calls.
TTEC’s James Bednar says some brands may take this route because the volume could be astronomical. Regular customers face the added steps too.
Detection is the next question. The signals described publicly include a sender that declares itself an AI, a machine-generated message, leftover AI framing in pasted text, a phone number not tied to the account, and audio analysis.
Sierra’s fleming-1 scores speech in real time and has no published accuracy figures. Much of this work currently sits with AI Agent and bot-management vendors. A search of public documentation did not turn up inbound detection of customer-side Agents from the major helpdesks, so it is worth asking each vendor directly.
Several questions remain open for every option.
Teams have not settled whether an Agent can accept a refund or consent to a data request, and only the account holder can close some accounts.
Identity checks assume a human account holder, and an assistant with the right details can pass them. Sierra notes that some automated callers are fraudsters testing account security.
A Five9 finding cited by SiliconANGLE reports that over half of dissatisfied users could not reach a human, and Agent-to-Agent automation could widen that gap. One low-commitment starting point is to tag suspected Agent contacts for a month and see what share they represent before choosing among the options.
What is the right adoption for your company?
I hope this was helpful with unblocking the ideas and direction you’ll take.
Need to discuss your strategy with someone?
I love to connect with people in CX, Support, and Operations.
Let’s find a place to start.
Ada and Dialpad Join Forces
Ada, an AI customer service agent platform, announced that it is joining forces with Dialpad, a business communications platform, to form a new company focused on agentic customer communications. Ada’s LinkedIn post calls the combination a merger, and Ada’s announcement page gives no ownership structure, pricing, or financial terms. According to the companies, Ada has served more than one billion people and deployed more than 500 customer-facing Agents, and Dialpad powers more than 50,000 businesses.
The stated plans put voice and Agents under one roof. The companies describe faster voice through a direct connection to the Dialpad network, shared workspaces where Support Specialists and Agents work on a request together, Agents that learn from both human and Agent conversations, and real-time guidance for Support Specialists along with automated quality reviews and coaching. The announcement also describes insights across conversations, including predicted customer satisfaction, which a team will want to compare against surveyed CSAT before relying on it. The companies frame customers as increasingly bringing Agents of their own, which connects to this week’s main story.
The announcement lists these capabilities as plans and gives no dates. Ada’s FAQ says nothing changes for current Ada deployments and that customers need to do nothing. Teams on either product can ask about the roadmap, contract continuity, and how data is handled as the systems combine, since the FAQ lists data handling as a question and the answer was not available in the page content reviewed. Teams that use another telephony provider with Ada can ask what the commitment to keeping the platform open to business systems means in practice.
Fin Adds eCommerce and Subscription Connectors
Fin, an AI customer service agent platform from Intercom, announced data connector templates, which are pre-built connectors installed from a catalog grouped by app and configured from the integrations a team already has. The catalog covers Stripe, Chargebee, Paddle, and Recurly for subscriptions, and Shopify, Shopify Storefront, WooCommerce, and BigCommerce for eCommerce, plus a Statuspage template for checking unresolved incidents. Depending on the template, Fin can look up orders and subscriptions, change plans, pause or cancel subscriptions, apply retention coupons, and issue refunds.
Order status, refund, and subscription-change requests make up a steady share of volume for SaaS and eCommerce support. When Fin can read the billing system or store directly, a Support Specialist does not need to open a second tool to answer a cancellation or refund question, and Fin can complete the request itself. For state-changing actions, Fin explains the change and waits for customer confirmation, then hands off to the team if it cannot identify a single account or a request fails. Each install creates an independent copy that a team can edit to match its own refund rules and retention offers.
Turn on identity verification before launch, because templates match customers by verified email or a synced customer ID, and a lookup fails when the verified email does not exactly match the email in the billing tool or store. Intercom also recommends marking email as a protected field and reviewing which response fields Fin can read. The Shopify Cancel Order, Update Order, and Create Refund templates have no confirmation step, so a team must add one in the description or the Fin Procedure. Stripe plan changes handle only items with a quantity of 1, and updates Intercom makes to a template do not change installed connectors automatically. WooCommerce needs pretty permalinks and does not support subdirectory installs, and Paddle needs a live API key.
Zendesk Adds Agent Takeover and Subtickets
Zendesk, a customer service and support platform, announced two additions in its November release notes. The first lets authorized admins and Support Specialists add internal notes, tags, and custom field updates to a ticket while an Agent conversation is ongoing, and take over the ticket to hand control to a human. The second lets Support Specialists link existing tickets or create subtickets from tasks in a task list, with the main ticket tracking each subticket’s assignee, status, and progress.
Agent takeover gives a Support Specialist a way to step into a live conversation: tag the ticket, leave an internal note for context, or correct a custom field without waiting for the Agent to finish. Subticket linking suits work that crosses teams, such as a billing adjustment or an engineering fix tied to one customer request. Because the main ticket shows the assignee, status, and progress of each subticket, the Support Specialist who owns the customer relationship can answer status questions without chasing other teams.
The release notes do not state plan availability or limitations for either feature, so confirm entitlement with Zendesk before planning a rollout. Takeover depends on an authorized access level, so review which roles hold that permission and define when a Support Specialist should step in. For subtickets, the related multiple task lists feature allows up to five task lists on a ticket, which makes a naming and ownership convention worth settling before usage spreads.
Pylon Ships Capacity Routing and Salesforce Viewer
Pylon, a B2B customer support platform, reported 140 updates in a single week, with capacity management for routing and an embedded Salesforce ticket viewer among the highlights. Capacity management lets admins set capacity limits, define exceptions, and enforce routing policies to balance workload across the team. The Salesforce integration can now show Pylon tickets and issue history on the Salesforce account page. The changelog also lists general availability for Support Agent, Automations, Assist Agent, Skills, and Connectors.
Capacity limits and exceptions let routing policy reflect the workload each Support Specialist carries, and the Salesforce viewer puts support history in front of account teams that live in Salesforce. Other additions include a Tickets Touched metric that counts unique issues with at least one message in a time window, CSAT score per ticket in the Analytics View Issues drilldown, an audit-log recovery workflow for accidental bulk deletes and updates, and recording consent and disclaimer settings for phone calls.
Capacity routing needs agreed capacity values and an owner for exceptions, so settle those with Support Specialist workload data before enabling it. Recording consent and disclaimer settings are worth a legal review before call recording is turned on. The Salesforce viewer requires configuring the integration, and a team should confirm which Salesforce users will see ticket history. With 140 updates in one week, assign someone to read the full changelog for items that touch existing configuration, and confirm plan and pricing details for the features now generally available.
Gorgias Becomes Official Claude and ChatGPT Connector
Gorgias, an eCommerce customer support helpdesk platform, announced that it is now an official connector in the Claude and ChatGPT directories. Users add it from either directory without setting up an MCP server by hand. Once connected, the assistant can find and update tickets, audit the Agent setup, draft new guidance, and test answers before they go live.
A support leader can ask an assistant to review how the Agent is configured, draft updated guidance, and test answers before they reach customers. Support Specialists can find and update tickets by asking, in the same assistant that holds their other connected tools. The test step gives a team a checkpoint for Agent changes that sits before publication.
The announcement does not describe permission levels, authentication scopes, or which data fields the connector exposes, so ask Gorgias what the connector can read and change before connecting accounts. Decide which roles may connect it, and confirm whether actions run under the permissions of the connected user. Guidance drafted by an assistant should pass through a human review step before it reaches the Agent.
Siena Handles Loop Returns Inside Chat
Siena, an AI customer service agent platform, announced in-chat returns and exchanges built on Loop Returns, available on live chat and email. The Agent finds the customer’s order, lists eligible items using the return reasons already configured in Loop Returns, asks the store condition questions, and reads back the refund, store credit, or exchange terms. It creates the return only after the customer confirms, and Siena describes itself as the first AI agent to do this with Loop Returns.
Customers complete a return without leaving the conversation, so the self-service portal is no longer a required step. For exchanges, the Agent offers same-product variants that are in stock at the same price, and one return can combine an exchange with a refund. Once the store provides a prepaid label and tracking number, the Agent shares them in the chat and resends them on request. It can also check return status and cancel a return. Only the customer answers pass to the merchant team, so a Support Specialist receives a return that is ready to process.
The announcement does not list prerequisites, plans, or limitations. It implies the store uses Loop Returns, since the Agent works from its reasons and condition questions, so results depend on how carefully those are configured. Labels and tracking numbers depend on the store providing them. Exchanges are described as same-product variants at the same price, so other exchange types presumably still need a Support Specialist, and a team should confirm that with Siena.
Plain Expands Workflows and Business Hours
Plain, a B2B customer support platform, announced a weekly drop led by more than 30 new workflow triggers, conditions, and actions. Workflows can now run when a snooze ends, an email bounces, an assignee changes, or a note is added. They can branch on thread age, linked issues, customer message counts, and other conditions, and they can create a task or run another workflow. The drop also adds business-hours overrides and several smaller changes to Sidekick and Customer Agent.
Business-hours overrides let a team add holidays, early closures, or extra opening hours to a schedule. SLAs and workflows follow the overrides, and running SLA deadlines recalculate while keeping their remaining business time. Sidekick now searches snippets before drafting a reply and tells the Support Specialist which snippet it used. Alternate support email addresses can use a different domain from the primary address, which lets one workspace support several brands.
The page does not state plan availability, so check your plan before assuming access. Customer Agent can now share up to three links from public knowledge, and Plain treats JWT-authenticated sources as public for linking, so review those sources before turning on Share links from knowledge content. Alternate email domains require DNS records and verification before Plain activates the address. A team with existing SLA policies should review them before adding overrides, since running deadlines will recalculate.
Sierra Redesigns Agent Studio Around Ghostwriter
Sierra, an AI customer experience agent platform, announced a redesign of Agent Studio around Ghostwriter, its agent for building Agents. Ghostwriter stays with the user across Agent Studio and can see the current journey, simulation, or dashboard. It replaces tools that were previously split across Explorer, Sierra Assistant, and other specialized agents. Each piece of work happens in a session that holds the conversation and the changes Ghostwriter makes, and those changes stay pending until someone reviews and merges them.
A builder can investigate why a metric changed, examine the conversations, update a journey, and test the fix in one session. Sessions can be shared with colleagues or forked, and Ghostwriter can review a teammate’s changes before they merge. Related Agents, along with their shared data and channels, can be grouped into one project and tested and released together. Release governance adds required approvals and controlled rollouts.
The announcement does not state availability, plan requirements, or limitations. Ghostwriter also monitors how Agents perform, suggests improvements in Slack and Teams, and pushes approved changes live, so a team should define who may approve changes from chat. Line-by-line comparisons, Preview Chat, and Simulations support review, and the quality of that review depends on the owner a team assigns to each merge. A team with several Agents should settle approval rules before multi-Agent projects begin.
Dixa Adds Retry for Failed Messages
Dixa, a customer service platform, announced a way to resend messages that fail to reach the customer. A clear error now appears next to a failed email or SMS message, and the Support Specialist assigned to the conversation receives a notification. Retry is available for email, contact form, SMS, and Facebook Messenger conversations, with up to three attempts.
Before this change, a Support Specialist had to copy the text of a failed message and send it again by hand. The error and the Retry button now sit next to the failed message, which gives the Support Specialist an immediate signal that a reply did not arrive and a one-click way to try again.
Only the assigned Support Specialist can resend, so the conversation must be claimed first. Retries are capped at three attempts. Most failed emails trace back to a suppression list, and retrying usually will not work when the customer address is on one, so read the error before retrying because each retry uses one of the three attempts. Dixa Support can remove an address from the list. The announcement does not state a plan tier or rollout timeline.
Guru Knowledge Agents Return Records Behind Numbers
Guru, a knowledge management platform, announced that its Knowledge Agents can now return the individual records behind numeric answers. Previously the Agents answered analytical questions with figures only. A user can ask how many cards need verification, then ask which ones, and receive the list in chat. The Data Retrieval Tool covers Cards, Connected Sources, Answers, Card Comments, and Card Comment Replies, which is a new dataset.
The feature is aimed at admins, owners, and authors who monitor the health of a knowledge base. A verification backlog count becomes a list of specific cards in the same conversation, which a support team can hand to authors as a work queue. Support Specialists benefit indirectly when the cards behind their answers stay verified and current.
The announcement does not give a release date, plan availability, limitations, or permission details beyond the intended roles. Confirm how existing permissions apply to the records an Agent returns. The usefulness of the lists depends on how consistently a team uses verification and ownership fields, so a knowledge base without those in place will return little that is actionable.
Attio Expands Activities and Workflows
Attio, a CRM platform, announced updates to Activities, Workflows, and its inbox, with limited detail in the changelog entries. Activities let a team log support tickets, social DMs, and product events as structured data on a record, with a custom schema for each activity type. Workflows and Zapier can create, update, or trigger on activities, and reporting on activities is in alpha. The Send message and Send actions blocks in Workflows can now reach a person or a channel in Slack, and a new inbox adds detail and a next step to notifications, with an Action needed view for items waiting on the user.
For support operations that share account context with sales and customer success, Activities give a way to put support tickets on the same record the account team uses. Slack messaging in Workflows can send deal or account context to the right person automatically. In the Action needed view, users can filter, group, and clear items in one click.
The changelog does not state plans or rollout timing. Because reporting on activities is in alpha, treat it as a preview. Each activity type carries its own schema, so define those before logging volume builds up. The same changelog adds admin controls for call recording and email protection domains, including self-service call recording off, so admins should review those settings.
Customer.io Adds Sign-Up Forms and Object Trigger
Customer.io, a customer engagement and messaging platform, announced sign-up forms and a new Object trigger for journeys. Sign-up forms can be built with little to no code and shown in-app, embedded with a script tag on any web page, or hosted on a page Customer.io provides, and each submission adds or updates a person in the workspace. The Object trigger starts a journey when a person, their relationship, and the related object meet all conditions, in any order.
Classic triggers check conditions only when an object or relationship changes, so people who qualify later miss the automation. With the Object trigger, people enter once all conditions are true, conditions can be grouped with OR and nested, and a team chooses between One time entry and Every re-match. By default, people leave the automation when they no longer meet the conditions. Sign-up forms give a team a way to capture contacts for product or status communications without engineering work.
Sign-up forms are rolling out to all accounts, so availability may vary during the rollout. The release note lists the Object trigger for the Essentials, Premium, and Enterprise plans. Automations on classic triggers keep running as before, so migration is optional, and a team should choose the re-entry setting deliberately because the default exit behavior removes people who stop matching.
Mintlify Launches Base Internal Knowledge Workspace
Mintlify, a documentation platform, announced Base, a workspace for internal knowledge where people and Agents work from the same specs, runbooks, and onboarding pages. Base runs on the same platform as Mintlify Docs and includes personal pages for drafts, a Shared with me view, and a Git-backed Teamspace organized in folders. It is in beta, and any team can sign up.
Runbooks and onboarding pages are daily references for Support Specialists. In Base, Agents in Claude, ChatGPT, and Cursor read from the same source through a built-in MCP server, and style guides, review checklists, and page templates can be set up once as skills and applied to every page. Reviewers leave comments and suggestions, and tagging Mintlify Agent in a comment prompts it to edit that section. Pages are stored as Markdown with version history.
Base is in beta, and Mintlify says parts of it will change as early teams use it. Page verification, which marks pages as verified with an expiration date, is coming soon, so stale content remains a risk until it ships. The separate Admin MCP grants write access for editing pages, restructuring navigation, opening pull requests, and managing settings, so limit who can use it. The post does not describe pricing or plan limits.
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.
