Will AI replace call center agents? Klarna said its AI did the work of 700 staff, then quietly rehired humans. See which calls still need one.
Call center agents are still needed, Klarna's own reversal, after publicly claiming its AI replaced 700 employees, is the clearest evidence yet that pure automation ran into real limits. Identity verification and basic guidance can be automated but the work of reading urgency from tone, rebuilding confused conversations, and calming emotion remains squarely human.
Call center agents do far more than answer questions over the phone. Within a limited amount of call time, they have to grasp the caller's emotions, urgency, and situation, provide the right guidance and confirmations, and move the caller toward the next action. Because they work with signals such as tone of voice, pauses, and signs of panic that text doesn't reveal, the job involves a great deal of phone-specific judgment.
The value of this role lies not in reading a script aloud, but in staying steady when the conversation goes off script and organizing the necessary information while calming the caller. Even as voice AI advances, calls involving strong emotions or complicated explanations are still likely to remain with humans.
Calculated by ai-job-risk.net from four public datasets, on a 0–100 scale relative to the 188 occupations we cover. Higher means more of the job's tasks are ones generative AI can do or is already used for. It is not a probability of losing the job.
Task analysis by researchers (Eloundou et al.; ILO). Shown as a percentile among all occupations they rate.
Observed use of Microsoft Copilot and Anthropic Claude on this occupation's tasks. Shown as a percentile among all occupations they rate.
Capital/labor substitution - share decreases as advances in speech and language processing and in artificial intelligence (AI) allow computers to handle customer service interactions with greater frequency.
US figures for the BLS occupation "Customer service representatives". BLS projections assume technology keeps developing at its historical pace; they are not a forecast of AI's effect alone.
Sources
Exposure index calculated by ai-job-risk.net from the datasets below (data as of October 15, 2026). US labour figures are from the BLS, unchanged.
The publishers did not prepare, review or endorse this site. Full citations, licences and method
AI is not fully replacing call center agents, and the clearest evidence is Klarna's own public reversal. In 2024, Klarna announced its AI agent was doing the work of 700 customer service employees; by 2026 the company says that figure has grown to the equivalent of 853 full-time agents and $60 million saved, but in between, Klarna quietly started rehiring human agents through a gig-style hybrid model after customers complained the AI gave generic answers and struggled with nuanced questions. Its CEO now describes human support as becoming a VIP experience rather than something being phased out.
Klarna's experience matches the broader 2026 data. A December 2025 Gartner survey found that only 1 in 5 customer service leaders had actually cut agent headcount because of AI, even though Gartner separately projects conversational AI will reduce contact-center labor costs by $80 billion in 2026, because only about 1 in 10 agent interactions is expected to be fully automated. Voice AI has steadily expanded automated identity verification, first-level guidance, and answers to frequently asked questions, so parts of call center work are highly likely to become even more automated.
But phone-based interactions often involve trouble the caller can't express clearly, or urgency that becomes evident only through tone, pacing, and hesitation. In situations involving anger, confusion, older callers, or mutual misunderstanding, exactly the cases Klarna's customers complained about, a human still needs to rebuild the conversation. Call center agents do more than answer the phone. Their job is to organize the situation through voice-only communication, reassure the caller, and move the issue forward.
The parts most vulnerable to voice automation are the calls that can be handled through fixed questions and simple branching, the roughly 60 to 70 percent of inbound calls that already follow a structured pattern, according to 2026 market estimates, and where the call-center AI market, now estimated near $4.89 billion, is concentrating its investment.
Routine opening tasks such as checking a contract number or providing business hours are easy for voice AI to handle, and this is exactly the layer most AI-assisted contact centers automate first. Roles that rely only on this part of the interaction are likely to shrink.
Questions with fixed answer patterns, such as payment dates, address changes, or basic procedures, are easy to automate and make up the bulk of the roughly 10 percent of interactions Gartner expects to be fully AI-handled in 2026. This is effective for reducing wait times, but even a slight deviation in the question can quickly increase the caller's frustration, precisely the failure mode Klarna's customers reported.
AI can already transcribe and summarize conversations quickly, substantially reducing the burden of record-keeping. Still, the emotional intensity and nuance the next person needs to know often require human supplementation.
First-level routing to the right specialist desk based on the caller's topic is relatively easy to standardize. Simple branching benefits greatly from automation, but calls where the issue changes midstream or remains vague can easily get lost unless a person catches them.
What remains for call center agents is the work of reading both the situation and the emotion through voice and rebuilding the conversation, the exact gap that surfaced when Klarna's AI-only approach ran into complicated, emotionally loaded calls and the company brought humans back.
Even when the topic is the same, priority can shift depending on panic, hesitation, or the length of a silence. The work of changing the response based on voice-only cues remains, and because emotion carries more strongly in speech than in text, how it's received directly affects support quality.
When a caller is confused and the topic jumps around, someone still has to decide what to confirm first, how to calm the person down, and where to restart the explanation. This is not work that can be reduced to reading a script; the ability to restore the conversation and move it forward remains human.
When emotions run high, deciding what to acknowledge and how before diving into fact-finding is crucial. If the order of explanation or response is wrong, the situation can deteriorate quickly, this is the type of call Gartner's survey respondents say AI still handles poorly. People who can build the foundation for trust recovery in a short time are hard to replace.
When billing, contracts, outages, and delivery issues overlap, someone still has to judge how and where to hand the case over. The job is both transferring the call and organizing it so the caller doesn't have to repeat everything, and that level of care has a major impact on satisfaction.
Future call center agents will need the ability to listen without missing details, structure what they hear, and create reassurance through voice alone. As Forrester predicts roughly 30 percent of enterprises will build parallel AI-support functions by the end of 2026, agents increasingly work alongside those systems rather than being replaced by them.
You need the ability to let the caller speak without cutting them off while still mentally organizing what needs to be confirmed. Because calls move in real time, speed in structuring the issue matters even more than in written support, and strong listeners are more likely to retain value even as AI handles the routine third of interactions.
Speech speed, pauses, backchanneling, and phrasing can all change how reassured a caller feels. People who can reduce misunderstanding through voice alone are strong, this is a deliberate skill, different from written communication, and it's exactly what customers said Klarna's chatbot lacked.
You need to judge how far you can handle a case yourself and at what point to involve a supervisor or another department. Holding a case too long can harm quality, but handing it off too early does the same; people who know when to switch appropriately earn trust on the floor.
Even with real-time summaries and response suggestions, the human agent still needs to keep control of the conversation. It's not enough to read the proposed wording; you need to adapt it to the tone of the call. As new roles like AI-operations specialist and conversation designer emerge alongside frontline work, agents who use these tools well while staying in charge of the interaction become stronger.
Experience as a call center agent builds strengths in voice-based situation assessment, emotional handling, and effective handoffs, strengths that transfer well as companies like Klarna prove out hybrid human-AI support models rather than fully automated ones.
Experience structuring situations and handling emotion over the phone can be applied to broader support roles across text and multi-channel environments, expanding from voice support into wider problem-solving work.
Exposure index 99/100 · US jobs 2025–35 -5.3%
Experience in frontline conversation control and priority judgment translates well into improving response quality across intake channels, moving from phone-centered work into broader frontline support.
Exposure index 99/100 · US jobs 2025–35 -5.3%
The ability to gather conditions over the phone while easing anxiety can also be applied to travel consultation and change management, turning strong phone communication into more proposal-oriented guidance.
Exposure index 89/100 · US jobs 2025–35 +0.2%
Experience reading emotional temperature and advancing a conversation can also translate into consultative sales, moving from receiving calls to leading proactive commercial conversations.
Exposure index 93/100 · US jobs 2025–35 -0.9%
The ability to structure conversations without damaging trust is also valuable in ongoing post-sale guidance, moving from reactive call handling into building long-term customer relationships.
Exposure index 83/100 · US jobs 2025–35 +3.9%
The ability to quickly understand someone's situation and choose the right next explanation also applies to candidate handling, using voice-communication strengths for evaluation and coordination in hiring.
Exposure index 85/100 · US jobs 2025–35 +6.4%
Call center agents are still needed, Klarna's own reversal, after publicly claiming its AI replaced 700 employees, is the clearest evidence yet that pure automation ran into real limits. Identity verification and basic guidance can be automated but the work of reading urgency from tone, rebuilding confused conversations, and calming emotion remains squarely human. Long-term prospects will hinge less on script compliance and more on whether someone can preserve quality in the difficult calls AI still can't handle well.
These roles appear in the same industry as Call Center Agent. They are not the exact same job, but they make it easier to compare AI exposure and career proximity.
Call Center Agent scores 99 out of 100 on our generative-AI exposure index (rank 1 of 188). That measures how much of the work AI can do or is used for, not whether the job will disappear. The BLS projects US employment of Call Center Agent to change by -5.3% from 2025 to 2035.
The parts most vulnerable to voice automation are the calls that can be handled through fixed questions and simple branching, the roughly 60 to 70 percent of inbound calls that already follow a structured pattern, according to 2026 market estimates, and where the call-center AI market, now estimated near $4.89 billion, is concentrating its investment. Examples: Identity verification and basic guidance; Voice responses to simple, structured inquiries; Summarizing call content and creating logs; Automating initial routing.
What remains for call center agents is the work of reading both the situation and the emotion through voice and rebuilding the conversation, the exact gap that surfaced when Klarna's AI-only approach ran into complicated, emotionally loaded calls and the company brought humans back. Examples: Reading urgency from tone of voice; Rebuilding the flow of the conversation; Initial de-escalation of complaints; Bridging cases across multiple departments.
Future call center agents will need the ability to listen without missing details, structure what they hear, and create reassurance through voice alone. As Forrester predicts roughly 30 percent of enterprises will build parallel AI-support functions by the end of 2026, agents increasingly work alongside those systems rather than being replaced by them. Examples: Active listening and issue structuring; Control of voice-based communication; Escalation judgment; Working alongside AI voice support, not just using it.
