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.
Our main story today is about Salesforce’s deal to acquire Fin, the customer service AI company formerly known as Intercom.
Salesforce Signs Deal to Acquire Fin, Formerly Intercom
Salesforce signed a definitive agreement to acquire Fin, the customer service AI company formerly known as Intercom, for approximately $3.6 billion. The deal is expected to close in the fourth quarter of Salesforce’s fiscal 2027, subject to regulatory clearance. Fin’s AI Agent resolves customer queries end to end across live chat, email, WhatsApp, SMS, phone, and Slack, and the acquisition brings a customer base of more than 30,000 companies into the Salesforce fold.
Closing is still months out, and both companies have shipped substantial updates in just the past six months. Rather than guess at what a combined product will look like, it’s worth looking at what each platform’s Agents can already do for support teams today, and what that suggests about where this combination could go.
Agentforce, Salesforce’s agentic AI platform, reached $1.2 billion in ARR last quarter, up 205% year over year, and the feature set behind that growth is substantial.
The Service Agent handles cases, answers questions, and manages orders without relying on pre-scripted chatbot flows. Agent Builder unifies drafting, testing, and deployment in one workspace, with low-code and pro-code options side by side.
Agent Script pairs deterministic workflows with LLM reasoning, so required steps execute in sequence while the model handles conversational nuance.
The Atlas Reasoning Engine breaks down a request, determines what data and actions are needed, and carries them out.
More recent additions include Intelligent Context, which extracts structure from unstructured sources like PDFs and call transcripts, and External Object grounding, which lets Agents pull live data from systems like an ERP or legacy database through Prompt Builder without replicating that data into Salesforce first. Guardrails and the Einstein Trust Layer are on by default, and Agentforce Voice extends all of this to phone support.
Fin’s AI Agent is built to operate across helpdesks rather than inside one. It runs natively in Intercom, and also deploys alongside Salesforce Service Cloud, Zendesk, and other platforms via API, without requiring a helpdesk migration.
It’s powered by Apex, Fin’s model built specifically for support conversations. Beyond answering questions, Fin’s Agent takes action: processing refunds, updating account details, changing subscriptions, and working through multi-step troubleshooting across whatever channel the customer used, including SMS and WhatsApp, which Agentforce doesn’t natively support.
Like any action-taking Agent, its effectiveness depends on how cleanly it can connect to the systems it needs to update.
For Support Leaders, the opportunity once this closes is access to both approaches without an either-or choice.
Teams already running Service Cloud and Agentforce gain a path to Fin’s broader channel coverage and its purpose-built conversation model, layered onto workflows already built on Flows, Data Cloud, and Agentforce’s governance tools.
Teams running Fin on Zendesk or another helpdesk gain a clearer line into Salesforce’s CRM data and cross-cloud automation if they need it later.
Salesforce has also flagged Fin’s packaged, fast-to-value approach as especially relevant for SMB and commercial organizations that need to launch quickly, alongside Agentforce’s more tailored, enterprise-scale deployments.
The practical move between now and close is to evaluate Agentforce and Fin on what each does today, and start mapping which of those capabilities would actually close a gap in your current support stack.
Fin (Intercom) Ships a Wave of Email Controls
Fin, an AI customer service platform, released six updates this week focused on email handling and ticket intake. Fin now sends a follow-up email when a customer goes quiet, files suspected spam into a dedicated review queue, and lets teams preview Fin’s email replies before they go live. The same release added channel-specific guidance rules for email, controls for how Fin responds on multi-participant threads, and ticket forms that customers can complete over WhatsApp, SMS, Facebook, Instagram, and email.
Operational Impact: The follow-up email closes the loop on conversations that go quiet, confirming whether the issue was actually resolved instead of assuming silence means success. The spam folder gives teams visibility into what Fin is filtering and a way to correct false positives, rather than having those conversations close silently. Preview lets a team test a change to Fin’s email guidance against a real reply before a customer sees it, which matters for anyone nervous about pushing untested instructions live. Channel-specific guidance means email can carry its own tone and escalation rules instead of inheriting whatever was written for chat, and multi-participant controls determine whether Fin replies to everyone on a CC’d thread or stays focused on the original sender. Ticket forms reaching WhatsApp, SMS, Facebook, Instagram, and email let a customer complete a structured request without switching to the Messenger.
Implementation Considerations: Channel-specific guidance and multi-participant controls only help teams that already have email-specific scenarios worth separating out. Teams with simple, unified guidance across channels may not see much change until they identify where email actually needs different handling. The spam folder’s accuracy depends on Fin learning from a team’s own corrections, so the first few weeks will likely require active review of what gets flagged. Preview is only as useful as the test scenarios a team runs through it, picking a single generic test case will miss the edge cases that matter most. Ticket forms across new channels require the form itself to already be built and configured, this extends where an existing form can be sent rather than creating one automatically.
Decagon Joins the Five9 CX Marketplace
Decagon, an AI customer service agent platform, announced it is now integrated and accredited on the Five9 CX Marketplace. The partnership lets joint customers deploy Decagon’s voice Agents on top of existing Five9 infrastructure, with warm handoffs to Five9 human agents when an interaction needs a Support Specialist.
Operational Impact: Teams already running Five9 for voice get a path to add an Agent without a separate telephony build-out or migration. Warm handoffs are meant to carry context, what the customer said and what the Agent already tried, to the Support Specialist picking up the call, which avoids making the customer repeat themselves. Decagon and Five9 point to refunds, account updates, and policy-driven workflows as initial use cases, the kinds of multi-step interactions that previously required a person from the start of the call.
Implementation Considerations: This is a marketplace integration announcement rather than a description of a finished deployment. The actual experience depends on how Decagon’s Agent is configured against a team’s own policies and systems, and on whether the existing Five9 call routing and IVR setup can hand off cleanly to and from the Agent. Teams evaluating this should ask specifically what a warm handoff carries over, call history, detected intent, customer data, before assuming the transition will feel seamless. Marketplace accreditation confirms technical compatibility, not that the integration is plug-and-play for every Five9 configuration.
Gorgias Launches Skills for Its AI Agent
Gorgias, an ecommerce customer support platform, announced Skills, a new way to configure its AI Agent with a dedicated playbook for a specific type of request, such as a return or a warranty claim, rather than letting the Agent decide how to handle it from the whole knowledge base. Skills is rolling out progressively to merchants over the coming weeks, with personalized recommendations generated from each merchant’s own ticket history and existing Guidance.
Operational Impact: Today, Gorgias’s AI Agent draws on a merchant’s entire knowledge base and decides what applies to each ticket, which works reasonably well in general but leaves limited control over high-stakes, policy-sensitive categories like returns and warranty claims. Skills ties a specific intent to a fixed set of instructions, so the same type of request gets handled the same way every time rather than varying based on what the Agent retrieves. Because Gorgias pre-populates recommended Skills from a merchant’s ticket history, merchants get a starting point tailored to what they actually deal with instead of building playbooks from a blank page.
Implementation Considerations: Rollout is progressive, so availability will vary by account for the next several weeks, and merchants who don’t see Skills yet will need to wait. Skills sit alongside the existing Guidance system rather than replacing it, so merchants need to understand which one takes precedence for a given ticket to avoid the Agent receiving conflicting instructions. The personalized recommendations come from historical ticket data, so merchants with limited AI Agent history or inconsistent ticket tagging will get thinner starting recommendations and more manual setup work to configure each Skill correctly.
Linear Agent Can Now Write Code in Sessions
Linear, a project and issue tracking platform used by many engineering and support teams, announced that Linear Agent can now write code using Claude Code or Codex. A coding session starts from an assigned issue, a chat message, or a Slack thread, pulls in that issue’s history and any linked customer requests, and returns a diff for human review. Linear says it already uses this workflow internally to resolve roughly 30 percent of incoming bug reports, mostly on the first pass.
Operational Impact: For support teams whose escalations sit in a shared backlog with engineering, this changes what can happen to a bug report after it’s filed. Linear Agent can investigate a new issue, gather evidence from tools like Sentry or Datadog through MCP, and propose a fix before an engineer looks at it, which shortens the gap between an escalation landing and a fix being available for review. Customer-facing teams that track time-to-resolution on engineering-dependent tickets now have a new variable in that number, assuming the agent’s first-pass attempts are accurate enough to act on without rework.
Implementation Considerations: Coding sessions require a GitHub connection with code access, run on Basic, Business, and Enterprise plans, and consume AI credits, all of which support leaders should account for when estimating Linear costs. The 30 percent first-pass figure is Linear’s own internal number, not a guaranteed outcome elsewhere, and it depends heavily on a team’s documentation quality, test coverage, and how clearly issues are written. Every proposed fix still returns as a diff for review, so someone needs to be available to evaluate and merge the agent’s work. This reduces investigation time more reliably than it reduces total engineering involvement.
Pluno Launches a Troubleshooting Agent for Tickets
Pluno, an AI customer support platform built for complex tickets, announced its Troubleshooting Agent, which investigates the kinds of tickets that currently require an engineer. The Agent checks Sentry, application logs, session replays, deploy history, and the database, correlates what it finds, and proposes a likely cause along with a suggested fix.
Operational Impact: This targets the category of ticket that typically bounces between a Support Specialist and engineering before anyone can act on it. If the Agent’s investigation holds up, a Support Specialist gets a documented hypothesis and a proposed fix attached to the ticket instead of filing a bug report and waiting for someone to pick it up. Pluno frames this as reducing how often engineers get pulled directly into support work, and as giving frontline staff enough context and evidence to run an approved script or configuration fix themselves without a developer in the loop.
Implementation Considerations: The Agent’s value depends entirely on what it has access to. It needs working connections to Sentry, logs, session replay, deploy history, and the production database, and teams without those tools in place, or without clean access controls around them, won’t get the same investigation depth. Letting frontline staff run database scripts or configuration changes based on the Agent’s suggestions is also a governance decision support leaders need to make deliberately, with clear guardrails on what a Support Specialist is allowed to execute versus what still has to route to engineering.
Enterpret Adds Four New Feedback Integrations
Enterpret, a customer feedback analytics platform, announced four new integrations: Pylon, Decagon, Fathom.ai, and HubSpot. The integrations bring support tickets and conversations, AI Agent transcripts, sales call recordings, and CRM activity into Enterpret so teams can analyze them alongside the feedback they already collect.
Operational Impact: Teams running an AI Agent through Decagon now have a way to analyze what that Agent is actually being asked, beyond whatever reporting the Agent platform itself provides, and to compare those conversations against tickets, calls, and surveys in one place. The Pylon integration brings tickets, conversations, call recordings, and survey responses together with synced contact and account data, which Enterpret describes as its most complete support coverage to date. The Fathom integration adds speaker-attributed transcripts from customer calls without manual export, and the HubSpot integration lets teams map CRM activity, deals, and custom objects into the same feedback pipeline.
Implementation Considerations: Each integration is a separate setup with its own documentation, and the value compounds only if a team connects more than one source rather than just one. Teams using Pylon, Decagon, Fathom, or HubSpot should treat this as a chance to consolidate feedback analysis, but the quality of Enterpret’s output still depends on how consistently each upstream source tags and categorizes its data. Adding more conversation volume into Enterpret without a plan for who reviews the output and what they do with it mainly adds more data to look at, not necessarily more insight.
Inkeep’s Content Writer Keeps Docs Current
Inkeep, a platform for building AI agents on a company’s own knowledge base, announced Content Writer, a feature that turns resolved support tickets, GitHub pull requests, and Slack messages into draft knowledge base updates. Each draft queues for human review and approval before it goes live, with the goal of keeping the underlying documentation, and any Agent built on it, current.
Operational Impact: Knowledge bases used to ground Agents tend to drift out of date as products change, and an Agent that answers confidently from a stale article is harder to catch than one that says it doesn’t know. Content Writer turns the resolution work a Support Specialist already does, closing a ticket and explaining what fixed it, into a starting draft for the documentation update that should follow, rather than relying on someone remembering to write that update separately. For knowledge managers, this shifts the job from authoring updates from scratch toward reviewing and approving drafts that are already grounded in a real resolved case.
Implementation Considerations: The quality of generated drafts depends on the quality of the source material. Tickets resolved with vague internal notes, PRs with no description, or Slack threads without context will produce drafts that need significant editing, still faster than starting from nothing, but not a hands-off process. Every draft requires human review before publishing, so this adds a review queue rather than removing the documentation step. Teams need someone with the time and authority to clear that queue regularly, or drafts will accumulate unreviewed and the knowledge base will keep drifting in the meantime.
Plain Adds CSAT-Based Workflow Triggers
Plain, a B2B customer support platform, shipped a set of workflow updates including new CSAT triggers that fire when a survey is sent or completed, with a sentiment condition to branch on whether the response was positive, neutral, or negative. The same release added manual workflow triggers with live execution progress in the thread dock, whole-word matching for message-contains conditions, a higher email signature character limit, and a new section title block for help center articles.
Operational Impact: The CSAT triggers let teams respond directly to survey results instead of reviewing scores separately and acting later. A negative response can route immediately to a manager or open a follow-up task, while a positive response can trigger something different, like a review request. The whole-word matching change addresses a real annoyance: keyword-based workflows that previously misfired on partial matches, Plain’s own example is “plan” triggering inside “planning”, will now only fire on the intended word, which should reduce the false positives that erode trust in automation over time.
Implementation Considerations: Acting on CSAT sentiment in real time only helps if someone is positioned to respond to what the workflow triggers. An immediate negative-CSAT escalation is only useful if a manager or specialist actually picks it up quickly, otherwise it just adds another notification. Teams should review their existing message-contains workflows after this update, since the shift to whole-word matching could change which conversations those workflows now catch or miss compared to before. The help center section title block and the signature length increase are smaller quality-of-life changes that affect how content looks rather than how the team operates.
Kustomer Rolls Out a Bundle of June Updates
Kustomer rolled out a set of June updates covering its AI Concierge, internal collaboration, queue management, and voice channel. Concierge now adapts its responses to brand-specific knowledge sources, with nightly re-indexing and the ability to roll back a configuration change. Agents can also start internal Slack threads directly from a conversation, admins gained controls to restrict user transfers on the Default Queue, and new voice metrics track rejected and timed-out calls alongside attribute-based call routing.
Operational Impact: The Concierge changes address a common complaint with AI Agents that answer from generic training instead of a brand’s own material. Nightly re-indexing keeps Concierge’s source material current without a manual refresh, and the rollback option gives teams a way to undo a configuration change that produces worse answers without waiting on a support ticket. Starting a Slack thread from inside a conversation keeps the back-and-forth a Support Specialist has with another team tied to the ticket it’s about, instead of living in a separate Slack thread that’s hard to trace back to later. The Default Queue transfer restriction and the new voice metrics give supervisors more direct control over where work lands and more visibility into calls that never connect.
Implementation Considerations: The value of brand-specific Concierge responses depends on what’s actually in those knowledge sources. Teams with thin or outdated source material will see Concierge re-index that same thin material every night, not improve on its own. The Default Queue transfer restriction is an admin-level control, so someone needs to decide what the restriction should actually be before turning it on, rather than leaving it at a default that doesn’t match how the team routes work today. The new voice metrics for rejected and timed-out calls are most useful for teams that already track call abandonment as a metric; teams that don’t will need to decide what threshold matters before the numbers are actionable.
Missive Improves AI Rules and Contact Imports
Missive, a shared inbox and team email platform, shipped an update covering AI automation, contact management, and calendar scheduling. The release adds a choice between Fast and Powerful AI models for rule-based actions, one-click attachment copying, default meeting lengths for calendar events, CommonMark markdown rendering in comments, and a guided CSV contact import flow with templates for Google and Outlook formats.
Operational Impact: The Fast versus Powerful model choice for AI rule actions lets teams trade speed for quality depending on what a rule does. A rule that drafts a quick acknowledgment can run on Fast, while one drafting a full reply can run Powerful in the background. The fixes to AI “Create draft” actions, which now use the configured sender alias and include the quoted previous message, address two specific issues that previously made AI-drafted replies look wrong or incomplete to the recipient. The guided CSV import flow with format-specific templates removes a common source of failed imports, where a file looked correct but didn’t match the expected column structure.
Implementation Considerations: Teams using AI rule actions in Missive should review which rules are set to Fast versus Powerful now that the choice is explicit, since the previous default behavior may not match what each rule actually needs. The shared address management changes, with Google Groups, distribution lists, and similar addresses moving to their own settings section, affect how incoming routing is configured, so teams with existing shared address setups should confirm the migration preserved their routing rules. None of these changes require new integrations or data migration beyond what Missive already manages.
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.
