Your Customer Service Job Is Splitting in Two — Here's Which Side to Be On

Two customer service reps. Same industry. Completely different experience of the same technology.

Ylonda Sherrod has been at AT&T for 17 years. She's won the company's top performance award two years running — top 3% nationally. And lately, she picks up every call with a question running underneath it: "Am I training my replacement?" The AI in her call center generates transcripts of her conversations, but it doesn't understand her Mississippi drawl. The summaries are full of errors. Soon, she won't be able to correct them. "If we don't talk about this," she told managers, "it could jeopardize my family."

Armen Kirakosian does similar work for TTEC, a global outsourcing company. He used to spend every call clicking through menus, scribbling notes, searching for information while a frustrated customer waited. Now, before he says hello, AI has already pulled up the customer's profile and flagged the likely problem. He has more time to actually help. "AI has taken the robot out of us," he said.

Both of these outcomes are real, happening right now. The difference between them isn't seniority, personality, or luck. It's what kind of work fills their queue — and that distinction is more specific than most advice about AI will tell you.

What AI Is Actually Taking (Task by Task)

The confusion around customer service and AI comes from treating it as a single job. It isn't. Most customer service roles are actually two jobs bundled together, and AI is aggressively interested in one of them.

The vulnerable half: order status lookups, password resets, basic routing, call summaries, note-taking, FAQ responses. These are retrievable, repeatable, well-documented tasks — exactly what AI handles well. The productivity gains are measurable. A landmark study tracking 5,179 customer-support agents found that access to a generative AI assistant raised issues resolved per hour by 14% on average, and by 34% for newer, less experienced workers. That efficiency had to come from somewhere. It came from eliminating the search-and-retrieve work that previously occupied significant portions of every call.

The growing half: catching AI errors before they reach the customer, de-escalating callers who've already been frustrated by a bot, handling the complex cases that automated systems escalated precisely because they were too hard. Gartner surveyed 321 service leaders and found that 85% are actively expanding what human agents are responsible for — because as routine contacts disappear into self-service, the human queue fills with everything the machine couldn't resolve.

Renz Miguel Marquez, a call-center trainer in Manila, describes his team's AI assistant clearly: it listens to the call and collates information quickly, which speeds resolution. But it also surfaces solutions that don't fit the customer's actual situation. Someone still has to catch that. That verification task wasn't in the original job description. Now it's core to the role.

A.I. has taken the robot out of us.
— Armen Kirakosian, TTEC Customer Service Agent

This applies beyond voice calls. If you handle email-based support, chat queues, or technical tickets, the split is identical. AI automates the retrieval and response-drafting steps across every channel. What remains human is verification, judgment, and repair.

Run this list against last Tuesday. If most of your day was column one, that's not a character flaw — it's information.

The Displacement Nobody Warned You About

Most conversations about AI and customer service focus on chatbots replacing human agents on the front end. But there's a second, less visible threat: AI entering through evaluation rather than conversation.

Megha S., a 32-year-old customer-service worker in Bengaluru earning $10,000 a year, didn't lose her job to a bot that answered calls instead of her. She lost it when her employer introduced an AI tool to review the quality of sales calls. She was told she was the first person replaced by AI. "I've not told my parents," she said. The displacement didn't come through the customer interaction layer. It came through the scoring layer — and that means any function where AI can evaluate, summarize, or rank human work is exposure territory, regardless of how strong that human performance actually is.

I was told I am the first one who has been replaced by AI.
— Megha S., Customer Service Worker, Bengaluru

This is why Sherrod's inaccurate transcripts are more than an annoyance. If an AI system is generating error-filled summaries of her calls — and she won't be able to correct them — those summaries may eventually shape how her work is assessed. Her expertise and her award history don't protect her from a bad data record.

How worried should you actually be? The honest answer involves a range. Forrester's 2026 forecast puts the long-term number at 49% of current customer-service jobs disappearing by 2030 — specifically the high-volume, lower-complexity roles that feed the most automated pipelines. Max Ball, Forrester's principal analyst covering this space, is direct: "There are humans today doing jobs that don't require the level of intelligence that a human has. That work is going to go away."

But the floor holds. Gartner's data shows only 20% of service leaders had actually cut agent headcount due to AI as of late 2025, and 55% reported stable headcount while serving more customers. More striking: Gartner predicts that 50% of companies that cut customer-service staff because of AI will rehire people for similar functions under different job titles by 2027. Automation overreaches, quality suffers, and the human capacity that was cut turns out to have been structural, not surplus. Klarna laid off hundreds of customer-service workers in 2024 citing AI capability, then began rehiring human representatives in 2025.

The threat is real. It has a predictable shape. And it has a window — not for complacency, but for deliberate repositioning before the easy cases disappear from the queue entirely.

What Repositioning Actually Looks Like

The workers navigating this well aren't waiting for their employer to announce a training program. They're shifting their visible value toward the tasks AI demonstrably struggles with: verification, emotional recovery, knowledge governance, and escalation judgment. These aren't aspirational soft skills. They're specific functions that organizations are actively building right now.

Gartner's survey found that 58% of service leaders plan to upskill agents specifically into knowledge-management roles — curating and maintaining the information on which both AI systems and customer self-service depend. This isn't a soft-skills promotion. It's a specific, technical function that the people closest to real customer confusion are best positioned to perform. If you know which knowledge articles are outdated, which exceptions frustrate customers, which policies the bot consistently misapplies — that knowledge is suddenly organizational infrastructure.

Kirakosian's own evolution shows what this looks like in practice. The same AI tools that eliminated his note-taking moved him toward learning and development — translating frontline experience into training that helps other agents work better alongside the technology. He didn't wait for a posting. He made himself the person who understood both the old workflow and the new one.

A 2026 field experiment on Alibaba's customer-service platform found something important about emotional escalation: when a customer was already frustrated with the bot before reaching a human, representatives who intervened early preserved service quality, while those who intervened late often couldn't recover the interaction. Early intervention isn't a personality trait. It's a skill. Knowing when the AI has made things worse — and what to do about it immediately — compounds in value as AI handles more first contact.

Four concrete moves, each grounded in what research shows organizations are actually building: volunteer to audit and update the knowledge base your AI and self-service tools run on; become the person who can identify when an AI-generated suggestion is wrong before it reaches the customer; own the escalations — the cases that arrive already frustrated, already having failed at self-service; learn to document failure patterns in AI interactions, not just your own calls. These roles are being filled from inside, not hired externally, because the people who understand the failure modes are the people already in the room.

If your work is email-based or chat-based rather than voice, the same logic applies. Review AI draft quality before sends, own the complex threads requiring policy interpretation, document the exceptions the system keeps getting wrong. Channel doesn't change the underlying positioning logic.

The Right Response to an Uncertain Forecast

Ylonda Sherrod is still at AT&T. She didn't receive a guarantee — she pushed for information, organized a union task force on AI, and traveled to Washington to testify about what it's actually like to have an AI system misrepresent your work. She is not protected by seniority. She's protected by being the person in the room who saw the disruption clearly enough to act before anyone asked her to.

The workers who navigate this transition well aren't the ones who weren't disrupted. They're the ones who looked at the disruption honestly, identified which part of their work was exposed, and moved toward the part that wasn't.

Here's this week's exercise. Pull up your calendar or task log from the past five workdays. For every task you completed, ask one question: could AI have done this with access to the right data and a verified knowledge base? If yes, that task belongs in column one. If no — because it required catching an error, reading a frustrated customer, making an exception call, or maintaining information that AI depends on — it belongs in column two. If column two is thin, that's not a crisis. It's a map.

The threat isn't coming for the workers who couldn't do the job. It's coming for the ones who were great at the part of the job that's now automatable — and the only way out is to see that clearly, before the queue makes the decision for you.

Recommended Tools & Resources