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.
No main story this week so we’re diving right into the updates.
Today’s headlines are from:
Decagon Releases Duet Autopilot, a Self-Improving AI Agent
Decagon, an AI customer service agent platform, announced Duet Autopilot, a feature that automatically analyzes production conversations, identifies where the agent is underperforming, proposes updated agent logic, and validates those changes through simulation before surfacing them for human review. The system runs overnight and requires human approval before any change reaches production. Alongside it, Decagon published DuetBench, a formal benchmarking framework for evaluating agent improvement end-to-end.
Operational Impact: Teams that currently spend time manually reviewing conversation logs to find failure patterns and then rewriting agent instructions get that cycle automated. Autopilot monitors 100% of interactions, traces issues to their root cause in the agent’s operating procedures, proposes targeted fixes, and runs those fixes against a golden test set before a human reviewer sees them. The reviewer gets a packaged proposal with diffs, test results, and full context rather than a blank-page task. Over time, the system also expands its own test coverage as new conversation types appear in production.
Implementation Considerations: The quality of Autopilot’s proposals depends on the quality of the existing Agent Operating Procedures and the volume and variety of production conversations feeding its analysis. Teams with thin conversation volume or loosely written procedures will see less precise output. The governance model places a reviewer in the approval seat for every proposed change, which means someone needs enough context to evaluate what is being proposed. Teams without that coverage risk approving changes they do not fully understand, which creates its own exposure.
Fin Overhauled Its Voice and Phone Workflows
Fin/Intercom, an AI customer service platform, released several phone and workspace updates. Fin Voice 2 launched with improved voice capabilities for AI-handled calls. New phone workflow features include a Collect Data step that captures caller information by keypad or voice input before the call connects to a Support Specialist, and live call quality indicators that flag degraded connections in real time with a breakdown across network, microphone, and speaker. Admins can now also configure role-based sidebar templates, setting default section layouts per role for the Intercom inbox.
Operational Impact: The Collect Data step removes the need for Support Specialists to ask callers for account numbers or case IDs after picking up. That information arrives in the conversation before the call begins, so agents start the interaction with context rather than spending the first minute gathering it. Call quality monitoring surfaces connection problems before a customer complains, giving agents a chance to address degraded audio in the moment. Role-based sidebars let ops teams standardize what agents see when they open a conversation, reducing per-teammate setup and creating consistent information access across a role.
Implementation Considerations: The Collect Data step requires phone workflows to be built out in Fin AI Agent > Workflows. Teams without maintained phone workflows need to construct that structure before the feature adds value. Role-based sidebar templates are configured in Settings > Teammates > Roles > Inbox and require a deliberate audit of what sections each role actually needs. Getting that configuration right upfront determines whether the standardization is useful or simply limits agent flexibility.
Hark’s AI Agent Sets Up Feedback Capture at Key Customer Moments
Hark, a voice-of-customer platform, launched Core Moments, a feature that uses an AI agent named Emmy to automatically configure feedback capture at specific points in the customer journey, including cart drop-off, first purchase, returns, and product usage. The feature connects to Shopify and Klaviyo and pairs qualitative customer feedback with the behavioral metric shift tied to each moment, such as a change in cart drop-off rate alongside the customer’s own explanation for it.
Operational Impact: Most feedback programs run on quarterly surveys disconnected from what a customer was actually doing at the time. Core Moments ties qualitative feedback directly to a behavioral signal, so when a metric like cart drop-off or first-purchase rate moves, teams have customer language attached to the moment rather than having to infer a cause. For CX and ops teams, this shifts the question from “what changed” to “why,” without a separate research effort.
Implementation Considerations: This is built specifically for Shopify brands and requires connecting Shopify and Klaviyo. Hark says setup takes about five minutes and is handled by their team rather than the customer’s developers, but that claim should be tested during onboarding rather than assumed. Teams without a Shopify storefront, or without Klaviyo in their stack, will not be able to use this feature as configured. For SaaS B2B teams evaluating Hark for support feedback generally, this update is most relevant if ecommerce operations are part of the business.
Dixa Sidebar Themes and Knowledge Article Versioning
Dixa, a customer service platform, released two updates. Agents can now choose from eight built-in sidebar themes in light or dark mode, or create a custom theme. The previous default style remains available as the Classic theme. Separately, Knowledge Article Versioning now automatically saves a snapshot of each help center article each time it is published, with side-by-side version comparison using word-level diff highlighting, one-click restore, and independent versioning per translation.
Operational Impact: The theme options reduce low-level friction for agents during long shifts, and the Classic fallback means no one loses a preference they had. The versioning feature addresses a more significant operational gap: knowledge managers who edit large article sets without version control face a slow recovery when a publish introduces errors or removes content that should have stayed. Per-translation versioning is particularly useful for multi-language teams, where a single-language update can otherwise create confusion about what is live in each locale.
Implementation Considerations: Sidebar themes are an agent-level preference — admins cannot enforce a standard theme across the team. Find it in Settings > Preferences > General > Appearance. A browser tab reload or desktop app restart is required. For versioning, snapshots are captured on each deployment, not on every draft save, so work-in-progress that is never published will not have versions to restore. Teams managing translated content should review how their translation workflow interacts with the independent publish toggle before making changes.
Linear Adds Team Documents and Shared Agent Skills
Linear, a project management platform, added a dedicated documents section to every team page, giving teams a home for notes, specs, and shared references that do not belong attached to a specific issue or project. Teams can create new documents directly from their team home page, pull in existing documents from issues and projects, and organize them into sections. The same release introduced shared skills for Linear Agent, allowing teams to create and distribute reusable instruction sets for recurring agent workflows like writing specs, running planning sessions, or drafting updates.
Operational Impact: CX and support ops teams that use Linear to manage internal projects, tooling evaluations, or process documentation now have a team-level home for that content. Without it, team context tends to end up attached to individual issues or stored in a separate tool, where it becomes hard to find or goes stale without a clear owner. Shared skills allow the agent workflows a team has established to be standardized across teammates rather than rebuilt per person.
Implementation Considerations: A new documents section fills quickly after launch and then stagnates without assigned ownership. Teams that do not designate someone responsible for maintaining specific documents will likely see that pattern repeat. The feature is accessible by clicking a team’s name in the sidebar, or via the keyboard shortcut o then t. Shared skills require someone to write, test, and maintain the instruction sets, and the quality of the instructions determines the quality of what the agent returns.
Help Scout Launches Company Profile for Account Health
Help Scout, a customer support platform, released Company Profile, a view that consolidates a customer account’s support metrics and activity into a single place. Accessible from the inbox, it surfaces response times, CSAT, conversation volume, and topic trends for a given company so Support Specialists can review account history before or during a customer interaction without pulling separate reports.
Operational Impact: Support teams that carry account health responsibilities, or that get pulled into renewal and escalation discussions, can access a summarized view of a customer’s support history in the flow of their work. The stated use cases include renewal preparation, account health reviews, and investigating changes in support patterns. Having that context in the inbox removes a step that previously required opening a separate analytics view or asking a colleague.
Implementation Considerations: The topic trends section is only as useful as the team’s conversation tagging and categorization practices. Teams with inconsistent or sparse tagging will see volume data without interpretable context. Response times and CSAT will reflect accurately regardless, but the topic breakdown requires clean upstream data. Teams considering this feature should audit their current tagging discipline before setting expectations for what the profile will show.
Gorgias Redesigns Its AI Automation Analytics
Gorgias, an ecommerce customer support platform, released a redesigned AI and Automation analytics section with two main pages: an Overview covering automation performance broadly, and an AI Agent page with dedicated sub-reports for Support Agent and Shopping Assistant. Each page follows a consistent layout with four metric cards, bar and line charts adjustable by channel or store, and a detailed sortable table. A metric glossary now documents how every formula in the section is calculated, and custom metric views and column preferences are saved per user.
Operational Impact: Support ops teams building internal accountability around AI automation now have a structured reporting surface designed for that purpose. The separation of Support Agent and Shopping Assistant performance into dedicated sub-reports reflects the different ways those roles contribute to a support operation. The metric glossary matters in practice: when stakeholders do not share a definition of “resolution rate” or “automation rate,” performance reviews stall on terminology. Having every formula documented in one place provides a reference that support and leadership teams can point to.
Implementation Considerations: Teams that have been tracking Support Agent and Shopping Assistant performance as combined metrics will need to recalibrate their benchmarks when reviewing the new sub-reports separately. Custom metric preferences are saved per user rather than per team, so organizations that want consistent reporting across multiple ops team members will need to align on a shared configuration manually. Teams using channel and store filters for more granular analysis should verify that their conversation tagging is consistent enough to make those filters meaningful.
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.
