Will AI replace call center agents? What the record shows (2026)

If you run a contact center, or work in one, the numbers arrive weekly. A fintech says its assistant does the work of hundreds of agents. A carrier says calls are down by half.

The obvious reading is that agents are next. The record TIN has checked reads differently. AI is taking contacts, but the replacement figures were mostly modeled, and the one AI a regulator opened up was people.

Start with the official US outlook. It projects customer service representative jobs to fall from 2,666,000 in 2025 to 2,524,100 in 2035, a 5 percent decline [source].

What “will AI replace call center agents” actually asks

Will AI replace call center agents asks whether software can take over the work a contact center pays a person for: resolving the customer’s problem, and owning the answer when it is wrong.

Answering a simple question is the part software already does. The harder part is the exception, the refund, the complaint, and the liability for what was said. That is what this post tests.

Will AI replace call center jobs?

It is expected to shrink them, not end them. The Bureau of Labor Statistics projects a loss of 141,800 customer service representative jobs over the decade [source]. That is 100 off the subtraction of its own two totals because BLS rounds each published value to the nearest 100 separately ([source]).

The agency names the cause. It expects less demand “as their tasks continue to be automated,” with self-service, social media and apps handling simple tasks [source].

It also expects companies to keep in-house service centers “particularly for complex inquiries such as refunding accounts or confirming insurance coverage” [source]. And it still projects about 289,500 openings a year [source].

These are forecasts, not measurements, and TIN did not check them. They do say something the headlines skip: the official view is attrition at the simple end, not an empty floor.

When will AI replace call center jobs?

For a share of contacts, it already has at some companies. At its October 2026 AI investor day, Deutsche Telekom said AI agents handle about 40 percent of US customer contacts [source].

The same report says US customer service calls have dropped by 55 percent [source]. These are company figures given at an investor event, reported by trade press. TIN has not checked them, and they count contacts, not agents.

That gap is the whole question. A contact handled by AI is not a job removed. Whether it becomes one depends on what the company does with the people, and the record shows companies choosing differently.

What the “work of 700 agents” figure really measured

It measured workload, not layoffs. Klarna said its assistant handled 2.3 million conversations in its first month, two-thirds of its service chats, and did “the equivalent work of 700 full-time agents” [source].

Klarna’s headcount did fall, from 5,527 at the end of 2022 to 3,422 at the end of 2024, per its IPO prospectus [source]. Its CEO attributed much of that to attrition under a hiring freeze, “not solely due to AI” ([source]).

Then the reversal. In May 2025 he said cost had been “a too predominant evaluation factor” and “what you end up having is lower quality” [source]. CNBC reported Klarna would recruit more human agents ([source]).

Can AI take contacts without cutting agents?

Yes, on the Octopus Energy record. Its CEO said the AI tool was answering 34% of customer queries by the end of April 2023, “the work of 250 people in the UK alone” [source]. He called it “unlikely to lead to job cuts” ([source]).

A person reviewed and sent each AI-drafted reply. In 2024 the company announced 4,000 new jobs [source]. Same “work of N people” framing as Klarna, opposite headcount decision.

AI in call centers: how much still goes to a human?

Less than before, but never zero in the files TIN holds. Decagon reports Notion’s support agent cut ticket resolution time by up to 34%, doubled deflection, and has a 3.4% ask-for-human rate [source].

Those are vendor figures, with Notion’s head of customer experience named on the page. They describe a support channel where people still sit behind the agent, handling what it routes to them.

Will AI replace BPO jobs?

Not in the one case a regulator opened up. The SEC found Presto’s drive-thru voice AI, on its original version, “required human agent intervention, including entering the order, in all instances” [source].

Those agents were contracted “at various off-site locations, including in the Philippines and India” [source]. Presto had told investors its automation rates ran from 95% to 99% [source].

The metric counted only restaurant staff as intervention, so the off-site agents never showed up in it. Even the advanced 2023 pilot needed a human on about 70% of orders [source]. Presto consented to the order without admitting or denying the findings.

Who answers for what the bot says?

The company does. A Canadian tribunal ordered Air Canada to pay C$812.02 after its website chatbot gave a customer wrong bereavement-fare advice [source].

The tribunal wrote: “It makes no difference whether the information comes from a static page or a chatbot” [source]. The decision never describes the bot as AI. The point stands for any automated agent: removing the human does not remove the liability.

The proof

TIN’s case file on Klarna prints the 2024 claims, the 2025 reversal and the headcount figures that do not reconcile, each with its own source: Klarna’s AI assistant and the return to human agents.

TIN’s case file on Octopus Energy shows the satisfaction figure moving between statements rather than merging it: AI-drafted support replies with no job cuts.

TIN’s case file on Notion names its limit, a single vendor source anchored to a named executive: Decagon’s support agent at Notion.

TIN’s case file on Presto quotes the SEC order line by line: the drive-thru voice AI that needed humans on every order.

TIN’s case file on Air Canada rests on the signed tribunal decision and two newsrooms: the airline held liable for its chatbot’s advice.

The broader question of AI support and headcount is argued in AI customer service automation and does AI customer service cut jobs. None of these files counts call center agents laid off because of AI, and this post claims no such number.

What each record shows

The bottom line

AI is replacing calls faster than it is replacing call center agents, and the gap between the two is where most of the headlines live. Every big replacement figure TIN has checked was a workload equivalence, and the one AI system a regulator audited was people in another country. The agents most at risk are the ones on simple, scripted contacts; the ones who handle exceptions and own the outcome are what Klarna went back to hiring.

The transferable test for any automation claim: ask who is left out of the denominator.