The Gist
Which service interactions are shifting to AI, and which stay human? Routine, high-volume tasks are moving to automated AI agents, while human agents are being repositioned toward complex, empathy-driven cases that build loyalty and trust.Is AI customer service actually getting cheaper? Gartner predicts GenAI cost per resolution will exceed offshore human agent costs by 2030, forcing organizations to rethink AI deployment around engagement value rather than pure cost reduction.Are companies planning to cut service jobs because of AI? Nearly 80% of customer service organizations plan to transition agents into new roles blending AI oversight, knowledge curation, and high-touch service delivery rather than eliminate positions outright.
A restaurant kitchen operates on a simple principle: the prep cook handles repetitive, high-volume tasks so the head chef can focus on plating, flavor, and the creative decisions that earn the reviews. Customer service organizations are arriving at a similar division of labor, except the prep cook is now an AI agent processing password resets, order status queries, and billing adjustments at a pace no human team could match.
The shift is well underway, and it is accelerating. A Gartner survey of 321 customer service leaders found that 91% are under executive pressure to implement AI in 2026, with improving customer satisfaction, operational efficiency and self-service success identified as the top priorities. The goal is no longer to bolt AI onto a legacy workflow. Service leaders are redesigning operating models so that automation and human expertise function as an integrated system rather than competing resources.
This article examines the forces driving that redesign. It explores why the traditional cost-reduction framing is losing ground, how AI and human agents are being reconfigured into complementary roles, and what practical steps marketers and CX leaders can take to align their teams with the emerging service model.
FAQ: How AI and Human Agents Are Splitting Customer Service Work
Editor’s note: These questions are drawn directly from the Gartner, NBER, Forrester and Zendesk research.
What does Gartner project for GenAI cost per resolution by 2030?
Gartner predicts it will exceed $3 per resolution, potentially higher than many offshore human agents, reversing the original cost-savings case for AI.
What does the NBER study find about AI’s effect on agent productivity?
Across 5,179 support agents, AI assistance increased productivity by 14% on average, with a 34% gain for novice and low-skilled workers.
What percentage of service organizations plan to redeploy rather than eliminate agents?
Nearly 80% plan to transition agents into blended roles combining AI oversight, knowledge curation, and high-touch service.
Why do many customers still prefer human agents over AI?
64% of customers say they’d rather companies not use AI for customer service, and 53% would consider switching providers if they learned AI was involved.
Why is the AI-to-human handoff considered the biggest operational challenge?
98% of leaders call smooth handoffs essential, but 90% say they struggle to execute them, and repeated context loss frustrates customers and erases efficiency gains.
From Code Automation to CX Automation: What Claude Routines Signal
One of the clearest recent signals of where AI-powered automation is heading came from Anthropic’s April 2026 launch of Claude Code Routines, a research preview feature that lets developers define AI-driven tasks and set them to run on a schedule, respond to API calls or react to GitHub events. Routines execute on Anthropic’s cloud infrastructure, which means they continue working when a developer’s laptop is closed. The feature packages a prompt, one or more code repositories, and a set of connectors into a single configuration that runs autonomously.
At first glance, Claude Routines is a developer productivity tool. Its documented use cases center on backlog maintenance, code review, deploy verification and documentation updates. A routine can triage overnight alerts, correlate stack traces with recent commits, and open a draft pull request before the engineering team starts their day. The feature operates through Claude Code, Anthropic’s command line tool for agentic coding, and is available to Pro, Max, Team, and Enterprise subscribers.
From Developer Tool to Service Infrastructure
The CX relevance emerges when you consider what Claude Code can connect to through the Model Context Protocol (MCP). Routines is currently oriented toward developer workflows, but the connector architecture extends into CX territory. MCP connectors already exist for Salesforce Service Cloud, Zendesk, Slack and other platforms central to customer service operations. A Zendesk MCP integration allows ticket creation, status updates, automated triage and routing based on issue type or customer tier. A Salesforce Service Cloud connector enables case data queries, knowledge article retrieval and customer interaction tracking. These are the same operational building blocks that define customer service workflows.
The pattern matters more than any single feature. When a scheduled routine can pull open support tickets from Zendesk, classify them by intent and urgency and post a prioritized summary to Slack before a support shift begins, the line between code automation and service operations automation dissolves. The infrastructure enabling these workflows, scheduled AI tasks connected to operational systems through standardized protocols, is the same infrastructure that customer service organizations need to automate the routine and preserve human attention for the complex.
What Matters Here: What Does Claude’s Connector Architecture Mean for Service Automation?
Claude Code Routines is a developer tool today, but its MCP connectors already reach Salesforce Service Cloud, Zendesk, and Slack — the same systems that run customer service operations, blurring the line between coding automation and CX automation.
The Pressure to Automate Meets the Demand for Empathy
Customer service teams have operated under a persistent tension for years: executives want faster resolution at lower cost, while customers want to feel heard and understood. AI has entered this dynamic as an accelerant, capable of resolving routine inquiries at a fraction of the cost and time of a human agent. Gartner predicts that by 2029, agentic AI will autonomously resolve 80% of common customer service issues without human intervention, leading to a 30% reduction in operational costs.
Yet customer sentiment reveals the limits of that automation promise. A Gartner survey found that 64% of customers would prefer that companies did not use AI for customer service, and 53% would consider switching to a competitor if they learned AI was being used. For marketers managing brand perception and CX leaders guarding loyalty metrics, these numbers represent a direct constraint on how far automation can extend before it erodes the relationship it was supposed to protect.
The practical implication for marketing and CX teams is that AI deployment strategy must account for customer psychology alongside operational efficiency. Service interactions involving complaints, billing disputes or emotionally charged situations require active listening and adaptive judgment that AI cannot reliably deliver. The workflow design question becomes less about what AI can do and more about where the customer relationship benefits from a human presence.
What Matters Here: Why Won’t Automation Alone Satisfy Customers?
64% of customers say they’d rather companies not use AI for service, and 53% would switch providers over it — a hard ceiling on how far automation can extend before it costs the relationship it was meant to protect.
Related Article: The Evolution of Customer Loyalty in an Always-on World
Why the Cost-Reduction Playbook Is Running Out of Runway
The initial business case for AI in customer service was built on cost arbitrage. Industry benchmarks place human agent interactions at $6 to $8 each while AI-driven resolutions run $0.50 to $0.70. Conversational AI was projected to save $80 billion in labor costs by 2026, and early adopters reported returns of $3.50 for every dollar invested.
That math is evolving in ways that complicate the pure cost-savings narrative. Gartner predicts that by 2030, the cost per resolution for generative AI will exceed $3, potentially higher than many B2C offshore human agents. Rising data center costs, a vendor pivot from subsidized growth to profitability, and increasingly complex use cases are all contributing factors.
Regulation Is Forcing Humans Back Into the Loop
Regulatory pressure adds another dimension. Regulations mandating easy access to human agents will encourage customers to bypass AI by default. Gartner predicts that by 2027, 50% of companies that attributed headcount reduction to AI will rehire staff under different job titles.
The strategic reframe emerging from these data points is significant. Organizations that deployed AI primarily to cut costs may find the math working against them as costs rise, customer trust erodes, and regulations force human agents back into the workflow. The alternative strategy, deploying AI to drive customer engagement rather than pure cost elimination, positions the technology as a complement to human capability. Forrester predicts that 2026 will be defined by foundational work rather than transformation, with service quality dipping as companies wrestle with AI deployment complexity.
What Matters Here: Is the AI Cost-Savings Case Starting to Break Down?
Gartner projects GenAI cost per resolution will top $3 by 2030 — potentially pricier than offshore human agents — while looming right-to-human regulation is already pushing half of AI-driven layoffs toward rehiring by 2027.
How the AI-Human Service Model Is Taking Shape
The operating model emerging across customer service organizations follows a recognizable pattern: AI handles the volume, humans handle the value. AI agents manage high-frequency, low-complexity interactions such as order tracking, password resets, and FAQ responses. Human agents focus on escalations, emotionally sensitive situations, and cases requiring judgment.
The Handoff Is the Real Bottleneck
The handoff between the two layers determines whether the system feels seamless or frustrating to the customer.
Several capability areas define how this model functions in practice. The following table maps the primary service functions against their AI and human roles, along with the integration factor that determines whether the experience feels unified or fragmented to the customer.
Service FunctionAI RoleHuman Agent RoleIntegration FactorRoutine Inquiry ResolutionAutonomous handling of FAQs, order status, billingEscalation only for exceptionsKnowledge base quality and coverageComplaint and Dispute HandlingInitial triage, context gathering, sentiment detectionEmpathetic resolution, negotiation, judgment callsSmooth handoff with full context transferProactive Service OutreachAutomated alerts for outages, delivery updates, renewalsPersonalized follow-up on high-value accountsCRM data accuracy and segmentationAgent Assist and CopilotingReal-time response suggestions, knowledge surfacing, case summarizationDecision-making, tone calibration, relationship buildingTrust in AI suggestions and override capabilityKnowledge ManagementAI-generated draft articles, content gap detectionReview, curation, contextual accuracy validationContent governance and update cadenceQuality AssuranceAutomated interaction scoring, compliance flaggingRoot cause analysis, coaching, process improvementInterpretability of AI scoring criteria Where the Model Gets Tested in Practice
Several elements in this model deserve closer attention. The agent assist and copiloting function is where AI and human roles overlap most directly. A National Bureau of Economic Research study of 5,179 customer support agents found that AI assistance increased productivity by 14% on average, with a 34% improvement for novice and low-skilled workers. The AI gathers context, suggests responses, and summarizes conversation history so the human agent can focus on tone, timing, and the emotional register of the interaction.
Knowledge management is another area where the model is evolving rapidly. Gartner found that 58% of service leaders aim to upskill agents into knowledge management specialists, acknowledging the need for accurate, continually updated content to support both AI systems and customer self-service. AI can draft and detect gaps, but human review ensures contextual accuracy and prevents the kind of confident-but-wrong answers that erode customer trust.
The handoff between AI and human agents remains the most critical operational challenge. Zendesk research indicates that 98% of leaders say smooth AI-to-human transitions are essential, while 90% admit they struggle with the mechanics. When a customer moves from an AI conversation to a human agent and has to repeat their issue, the efficiency gains from automation are offset by frustration.
What Matters Here: Where Does the AI-Human Handoff Break Down Most Often?
98% of service leaders call smooth AI-to-human handoffs essential, but 90% admit they struggle with the mechanics — the gap where efficiency gains get erased by customer frustration.
Key Actions for Leaders Redesigning the AI-Human Service Split
The following table highlights the most important lessons, actions and strategic considerations emerging from Gartner’s shifting AI cost-per-resolution projections and the operational data on AI-human service handoffs.
Key AreaWhat HappenedWhy It MattersRecommended ActionCost assumptionsGartner projects GenAI cost per resolution will top $3 by 2030Undermines the “AI is always cheaper” business caseModel AI ROI on engagement value, not per-resolution cost aloneCustomer trust64% of customers say they’d rather avoid AI in service; 53% would switch providersOveruse of automation risks loyalty and retentionReserve AI for low-stakes interactions; keep humans on disputes and complaintsAgent role redesign~80% of service orgs plan to redeploy, not cut, agentsSignals a shift from headcount reduction to role transformationReframe agent success metrics around resolution quality, not handle timeHandoff mechanics98% of leaders call AI-to-human handoffs essential; 90% say they struggle with itPoor handoffs erase efficiency gains and frustrate customersVet vendors specifically on context-transfer and escalation-detection capabilityKnowledge base readiness61% of leaders report a backlog of articles needing updatesAI self-service is only as accurate as its source contentPrioritize knowledge management investment before scaling AI self-service Preparing Your Team for the Blended Service Model
For CX leaders and marketers evaluating their next steps, the data points toward several practical priorities. First, resist framing AI deployment purely around headcount reduction. Gartner predicts that over 50% of customer service organizations will double their technology spend by 2028 without an equivalent reduction in talent.
A useful starting exercise is to audit your current service interactions by complexity and emotional intensity. Map which interactions can be fully automated, which benefit from AI assist with human oversight, and which should remain human-led. The ambiguous middle ground, where most organizations struggle, is where pilot programs and iterative testing yield the most insight.
Once the customer does get through to the human representative, they have come to expect a different level of empathic skill and emotional intelligence due to all that AI can handle. Some of the most successful uses of AI don’t replace human judgment; rather, they enable the human to focus on what they alone can do. If you are searching for others who share your struggle in transitioning from one to the other, customer service management LinkedIn communities and industry-specific forums such as the Zendesk community would be good places to start.
— Chris Cornella, VP Business Development
US Professional Funding
When evaluating AI customer service platforms, ask vendors specific questions about handoff mechanics. How does the system transfer context from an AI conversation to a human agent? How does it detect when a customer is escalating emotionally and needs human intervention? The answers will reveal more about operational readiness than any feature comparison chart.
Fix the Knowledge Base Before You Scale
Invest in knowledge management before scaling AI self-service. AI systems are only as accurate as the knowledge bases they draw from, and 61% of leaders report a backlog of articles to edit. Running an AI agent against an outdated or inconsistent knowledge base scales inaccuracy, which compounds into support tickets that human agents then have to clean up.
Finally, redefine agent success metrics. If human agents are increasingly handling complex, high-stakes interactions, measuring them on average handle time penalizes the behavior you want to encourage. Metrics that reward resolution quality, customer retention outcomes and knowledge contribution will better align agent performance with the blended model your organization is moving toward.
Learn how you can join our contributor community.