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 the customer signal that AI support is quietly filtering out.
Today’s headlines are from:
AI Is Resolving Tickets while Losing the Signal
When Robert Cabral’s team at Runway launched AI support, the first thing that broke was the handoff. A customer would explain their situation, get transferred to a Support Specialist, and find themselves starting from scratch. Robert’s team caught it in the first week. They built a system that packages the full conversation and any customer-submitted attachments as an internal note before a Specialist opens the ticket. That fix landed.
What is still open is quieter and harder to close.
I have been thinking about this problem for a long time. Before the term signals was used, Chris DiNicolas and I were working inside Intercom and Airtable while at Frame.io, building a tagging and categorization system to turn raw ticket volume into something the product team could act on. Manual tags, automated rules, a structure we kept rebuilding as the data outgrew it. We gathered so much each month that organization became the constraint.
Chris continued evolving that system after I moved into Adobe Support Operations, and what he eventually built is one of the more complete solutions to this problem I have seen. The through-line from what we were doing then to what Robert is working through now is the same: AI makes the volume manageable but does not make the signal visible.
When every ticket passed through a Support Specialist, signal was ambient.
Someone noticed the same complaint three times in a week. Someone mentioned it in a channel. That informal loop was inefficient but it worked. Robert described what breaks when AI handles first contact: a customer reports that generations are running slow, AI delivers standard browser-check guidance, and the same response goes to the next fifty customers before anyone on the team registers a pattern. The ticket resolves. The signal does not.
Robert is building a detection layer to surface patterns across AI conversations before they become a volume problem. It is not finished. Discord runs in parallel as a secondary signal source because the community surfaces what the ticket queue does not. He is monitoring both while he builds something more systematic. The visibility problem does not announce itself, which is what makes it easy to defer.
AI closed the resolution loop and opened a gap that most teams have not named yet, let alone solved.
📺 Watch or listen to the full conversation with Robert Cabral on Unresolved: The signal AI doesn’t know it’s missing
→ For a look at how this problem was eventually solved at scale at Frame.io, Chris DiNicolas’s piece on building a scaled AI-native VOC system is well worth the read: Building the Scaled AI-Native VOC
Fin (Intercom) Adds SLAs and Live Coaching to Phone
Fin (Intercom), a customer support and messaging platform, announced two updates to Fin (Intercom) Phone. SLAs can now be applied to inbound phone calls, with speed-of-answer targets set inside workflows, SLA status visible per conversation in the inbox, and phone SLA performance tracked alongside chat and ticket SLAs in existing reports. Separately, managers and supervisors can now enter live calls using two modes: Join (barge) to enter the call audibly, and Coach (whisper) to speak privately to the agent without the customer hearing.
For support teams running phone alongside chat and tickets, SLA enforcement on calls has historically required separate telephony tools or manual tracking. Bringing phone SLAs into the same reporting layer as chat and ticket SLAs means performance data consolidates in one place instead of requiring cross-platform reconciliation. The coaching tools change the dynamic for real-time QA. Supervisors can redirect an agent mid-call before a customer frustration compounds, without interrupting the conversation. Whisper mode removes the all-or-nothing choice between staying silent and jumping in audibly.
Phone SLA requires a workflow step, not a global setting. Teams that have not structured phone workflows in Fin (Intercom) will need to build or update them before enforcement applies. Barge and whisper require supervisors to be actively monitoring the call dashboard, which assumes staffing coverage and tooling that not every team has in place. Teams using third-party telephony and treating Fin (Intercom) Phone as a secondary channel should audit whether this feature set justifies consolidation before investing in configuration.
Help Scout Lets Teams Build Views by Conversation Activity
Help Scout, a customer support platform for growing businesses, announced thread-level views. The update allows teams to create inbox views filtered by activity within a conversation, including which customers are waiting on a reply, conversations with low satisfaction ratings, or messages containing specific topics, combinable with existing filter conditions.
Views built on thread-level activity let queue managers surface conversations that need attention without manual review. A view filtered for waiting on reply makes backlog prioritization visible at a glance instead of requiring agents to scroll. CSAT-based filtering creates a direct path for supervisors to follow up on poor ratings without exports or manual flags. Combined with existing filters, teams can build sharper queue logic rather than relying on catch-all inboxes.
The value of these views depends on the consistency of the underlying signals. CSAT ratings require customers to respond, which limits coverage on teams with low response rates. Topic-based keyword filtering will surface false positives if message content is inconsistent or informal. Teams should test view logic against real ticket samples before using it as a queue management tool. The full list of supported filter conditions is in Help Scout’s help center.
Zendesk June Release: AI Agent Expansion, Ticket Summaries, and Auto Link Discovery
Zendesk, a customer service platform, published its June 2026 product update covering three changes relevant to support operations. AI agent capabilities are being expanded to all Zendesk customers, removing plan-tier distinctions and unlocking features including agentic reasoning and API integrations across the customer base. AI ticket summary is now available at no additional cost for Suite and Support Professional plans, providing a pooled monthly allowance of 500 uses across the team, with no setup required. Automatic link discovery for the web crawler now allows Zendesk’s knowledge crawler to index external web content by following site links automatically, without a sitemap or manually specified URL list.
The AI agent expansion removes a common blocker for teams on lower-tier plans that wanted to pilot autonomous deflection but hit feature walls. Teams that have been waiting on a plan upgrade before deploying AI agents should re-evaluate timelines. Ticket summaries reduce the time agents spend reading thread history on complex or reopened conversations. The 500-use monthly cap pooled across a team is low enough that high-volume teams will need to triage which conversations receive summaries, making the tool more practical for escalation queues than high-volume simple ticket flows. Automatic link discovery changes the calculus for teams maintaining external documentation in marketing sites or developer portals, where content previously had to be manually mapped before the crawler would pick it up.
AI agent expansion does not mean all teams are ready to deploy. Teams without clean conversation data, defined escalation rules, or tested knowledge sources will face the same readiness challenges regardless of plan tier. The pooled 500-use monthly summary allowance requires monitoring; a team of 50 agents sharing 500 uses has roughly 10 uses per agent per month before the pool runs out, half the per-agent nominal cap. Web crawler auto-discovery carries no sitemap requirement, but teams should audit what content the crawler indexes on the first run, particularly for sites that serve multiple product lines or audiences.
Brainfish Ships Live Agent Handoff for Zendesk
Brainfish, an AI-powered support automation platform, announced Live Agent Handoff for Zendesk. The integration transfers a customer from the Brainfish AI directly into a Zendesk Agent Workspace conversation in the same window, with the full AI exchange including messages, retrieved sources, and confidence scores loaded into the agent’s view before the agent types a reply. The customer does not switch surfaces or re-explain context. Live Agent Handoff is generally available to Brainfish customers running Zendesk.
The persistent failure point with AI support deployments has been the handoff. A customer who spends several minutes explaining a problem to an AI and then gets transferred to a human who starts from scratch loses whatever trust the first interaction built. Live Agent Handoff carries the entire AI conversation, retrieval chain, and escalation reason into the Zendesk ticket automatically. For support leaders, AI-to-human escalations become structured events with attached metadata rather than plain tickets with notes, making CSAT and handle time analysis per escalation cohort possible. Brainfish states early teams report 30 to 45% reductions in handle time on handed-off conversations; these figures come from Brainfish and have not been independently verified.
Live Agent Handoff requires both Brainfish and Zendesk Agent Workspace. It does not require a separate Zendesk license. Teams not currently running both platforms will need to factor in 1 to 2 weeks for integration and deployment. Handoff trigger configuration, whether intent-based, confidence-threshold-based, or customer-initiated, requires teams to define their threshold for AI escalation before setup. Teams without clear definitions of when the AI should hand off will find the configuration difficult to calibrate. The content clustering feature that surfaces knowledge gaps from handoff transcripts depends on handoff volume; teams with low handoff rates will not generate enough signal for meaningful clustering.
Gorgias Opens MCP Server to All Customers in Beta
Gorgias, an ecommerce-focused customer support platform, announced its MCP (Model Context Protocol) server is now in open beta. The first-party MCP server connects a Gorgias workspace to AI tools including Claude, ChatGPT, Cursor, and any MCP-compatible application, enabling teams to surface customer insights, audit helpdesk configuration, and review AI Agent performance through natural language in AI chat interfaces. Read access is fully live. Write operations such as Macro edits and AI Agent configuration changes are gated during beta and rolling out incrementally.
MCP connectivity shifts how support teams can interact with their own data. Instead of navigating dashboards and export flows, a support leader can ask a natural language question and get an answer drawn directly from live Gorgias data. Gorgias highlights Voice of Customer analysis (top return drivers, CSAT patterns), helpdesk hygiene (stale rules and macros, tagging gaps), and AI Agent review as practical starting points. The distinction between reads and writes matters for planning: teams should not build workflows that depend on live configuration changes through the MCP until write restrictions lift.
Gorgias MCP is included at no extra cost on any plan, but requires a separate active subscription to an AI tool. The LLM is not included. Teams without Claude, ChatGPT, or another MCP-compatible tool already licensed will need to account for that cost separately. Setup is documented as approximately 5 minutes, which is realistic for teams with standard configurations. The quality of AI-generated analysis depends on the underlying Gorgias data. Teams with inconsistent tagging, incomplete ticket dispositions, or sparse knowledge articles should address data quality before drawing conclusions from MCP outputs.
Plain Rebuilds Ari AI with Agentic Search and Retry Logic
Plain, a B2B customer support platform for Slack, Teams, and email channels, announced updates to Ari, its AI agent. Rather than a single classification pipeline that routes to human handoff when confidence is low, Ari now iterates: it searches knowledge sources, evaluates the results, refines the query if needed, drafts an answer, and uses a judge to retry with feedback rather than immediately escalating. Question classification has also been improved, and every tool inside Ari now reads from a single consistent view of the thread, fixing edge cases where Ari only saw partial context.
The prior classify-and-handoff approach means borderline questions that could be answered with a refined search escalate to a human instead. The iteration loop attempts to close that gap before handing off. For support teams managing high-volume Slack channels where agents are frequently context-switched, reducing unnecessary handoffs has direct throughput value. The retry-on-rejection model allows the system to improve a resolution attempt mid-conversation without surfacing the failure to the customer.
Agentic search with a retry loop means Ari takes more steps before responding, which adds latency to initial replies. Teams accustomed to near-instant AI responses in Slack should test response times in their specific channel configurations before communicating changes to customers. The quality of search refinement depends directly on knowledge source coverage. If the knowledge base has gaps, a refined query finds the same absence of content the first query found. Retry logic does not compensate for missing documentation.
Birdie Adds Custom Domain Support for Recording Pages
Birdie, a screen recording tool for customer support teams, announced Custom Domains for its recording pages, now available to all customers following a beta period. Instead of a Birdie-branded URL, customers who click a recording request link now see a URL on the team’s own domain, such as capture.company.com rather than notion.birdie.so, without changes to the recording flow or any download requirement. More than 100 teams deployed custom domains during the beta.
Customer hesitation to click recording links is consistent friction in async support workflows. The domain a customer sees before clicking is the first trust signal, and a third-party subdomain raises questions about whether the link is legitimate. Custom Domains address that hesitation at the URL level. For teams supporting enterprise customers in security-sensitive environments, a link on the company’s own domain is also easier to pass through customer security review, which reduces the manual exception requests support teams are often asked to facilitate.
Custom domain setup requires a DNS configuration, which means coordination with whoever manages the company’s domain records. Support teams that do not control their own DNS will need to open a request before the feature can go live. Teams that have already shared recording request templates, macros, or documentation with customers will need to update those materials once the domain changes. Existing Birdie-branded links continue to work unless deprecated.
Linear Launches Native Code Review with Linear Diffs
Linear, a project management and issue-tracking platform, announced Linear Diffs. The feature allows teams to review pull request code changes directly within Linear from any issue with a linked PR, with reviews syncing back to GitHub in real time. A guided review mode, in beta for Business and Enterprise plans, surfaces core changes first, provides explanations per section, and separates primary changes from secondary ones such as glue code. A Review Inbox consolidates review notifications alongside other Linear work, sorted by proximity to shipping.
For support and operations teams that use Linear to track bugs and feature requests, code review has previously required switching between Linear and GitHub. Linear Diffs removes that context switch for teams that want to stay in one tool. For CX-adjacent teams that file bugs and track fixes, the ability to see diff status on an issue means they can verify what changed in a bug fix without needing direct GitHub access. The guided review mode is particularly targeted at reviewing AI-generated code, where the volume of changes per PR has grown faster than traditional review tools were designed to handle.
Linear Diffs requires workspace admins to enable code access through the GitHub integration upgrade, and individual users must separately opt in to review diffs in Linear. Guided reviews are free during beta but limited to Business and Enterprise plans. Teams on lower plans have access to basic diff review without the guided format. Linear syncs reviews back to GitHub, but teams should verify that existing GitHub branch protection rules and required reviewer workflows still function as expected after enabling the integration.
GitBook Brings Agent Inline to the Editor
GitBook, a documentation platform used by technical and support teams, announced improvements to GitBook Agent inside the editor. The Agent is now accessible via a floating panel in the editor itself, without switching to the sidebar. When text is selected, the Agent is available directly from the format toolbar. On every change request, the Agent is automatically added as a reviewer to check documentation against existing content with no manual setup. Free users now have access to the Agent with a soft weekly limit.
For support and knowledge teams using GitBook to maintain internal documentation or help center content, the prior workflow broke whenever edits were small, since opening a sidebar disrupted the editing flow. The inline floating panel keeps the Agent available during active editing without that interruption. The automatic reviewer addition on change requests is more significant for larger teams: every documentation update receives an AI consistency check against the existing content library without a manual step, creating a structural check against content drift over time.
The automatic reviewer addition changes what happens when a change request is submitted. Teams with established change request workflows should verify that the AI reviewer does not disrupt approval requirements or notification settings they have configured. Free users gaining Agent access with a weekly limit could affect teams that have been using the Agent heavily under a paid plan if plan assignments are not reviewed. Teams should confirm plan settings before rolling out the inline Agent to the broader team.
Attio Redesigns Android App and Speeds Up Email Sync
Attio, an AI-native CRM platform, published its May 28 changelog with two notable changes for support and success teams. The Attio Android app has been fully redesigned and released to the Google Play Store. Email sync performance improved significantly, with average parse time more than four times faster and long hangs affecting some mailboxes resolved.
For support and success teams that use Attio to track customer relationships, email sync latency creates a real workflow problem: if the CRM does not reflect recent customer emails, agents act on stale context. A four-times improvement in parse speed means email history reaches agents faster, which matters most in handoff scenarios or when multiple people are working an account simultaneously. The Android redesign benefits teams where account managers or support leads access Attio outside the office or during customer visits.
The Android update is an app store update, not an automatic enterprise deployment. Teams using mobile device management should verify the new version is compatible with their MDM policies before pushing an update. Email sync improvements are server-side, but teams that experienced the long hang issue Attio describes should verify their mailbox connections after the update to confirm the fix applied. The changelog also includes dropdown menu and navigation refinements that will adjust daily workflows, though these are UX improvements rather than configuration changes.
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.
