Businesses and technology researchers are debating whether artificial intelligence will replace traditional call centre roles or become a complement to human agents. Companies are piloting more advanced "AI agents" that can act autonomously, while analysts and service firms warn of limits and implementation challenges.

Business and technology research firm Gartner forecasts that agentic AI will autonomously resolve a large share of routine customer-service problems in the coming years. Gartner predicts that AI will autonomously resolve 80% of common customer service issues by 2029.

Executives in large technology firms have also signalled a potential reduction in human-run operations. K Krithivasan, chief executive of Tata Consultancy Services, told the Financial Times that AI may soon mean there is "minimal need" for call centres in some regions.

The term "AI agents" refers to systems capable of operating with greater autonomy than traditional rule-based chatbots. These agents can generate human-like responses and make decisions without following a fixed list of pre-programmed replies, potentially extending the range of issues handled without human intervention.

Many organisations are already using chat interfaces that are not truly generative AI. Those systems typically follow rules and guided flows, which can resolve straightforward queries but fail when conversations deviate from expected patterns.

One customer reported a non-AI chatbot telling them a parcel had been delivered and then showing a photograph of the package at the wrong front door, with no option to continue the conversation after the evidence was shown. The parcel firm involved says it is investing £57m to improve the service and offered a statement:

"Our intelligent chat facility uses tracking data to suggest the most helpful responses and ensure the customer's parcel is delivered as soon as possible, if this has not happened as scheduled,"

"Our data confirms the vast majority of people get the answers they need from our chat facility, first time, within seconds. We're always reviewing feedback to ensure our services are as helpful as possible, and we continue to make enhancements on a rolling basis."

By contrast, another delivery company temporarily disabled its less-restricted AI chatbot after it began criticising the company and swearing at users, illustrating risks when models are given too much freedom without safeguards.

Analysts and vendors say adopting generative AI is not a simple cost-cutting exercise. Emily Potosky, an analyst at Gartner, highlights both potential and pitfalls in deploying AI for customer service:

"You can have a much more natural conversation with AI," she says. "But the downside is the chatbot could hallucinate, it could give you out-of-date information, or tell you completely the wrong thing. For parcel delivery I would say rules-based agents are great because there are only so many permutations of questions about someone's package."

She also emphasises the financial and data demands of these systems:

"There's this idea that knowledge management becomes less important because generative AI can solve the fact that their knowledge is not particularly well organised, but actually the opposite is the case,"

"Knowledge management is more important when deploying generative AI."

For many businesses, developing useful AI agents requires extensive, well-organised training material. Companies that have outsourced customer service to lower-cost centres often possess detailed documentation and recorded interactions that can be used to train models, making those operations natural starting points for automation.

Software vendors are already offering AI platforms aimed at handling first-line customer interactions. Joe Inzerillo, chief digital officer at Salesforce, describes how call-centre archives and training materials can feed agentic systems:

"You have a huge amount of documentation, and that's all really great stuff for the AI to have when it is going to take over that first line of defence," he says.

Salesforce says its customer-service platform, AgentForce, is being used by customers across sectors and reports several results from early deployments. Mr Inzerillo describes adjustments made to make the AI more effective and empathetic in customer exchanges:

"While a human might say 'sorry to hear that', the agent just opened a ticket,"

He also recounts an example of an overly rigid safety rule and its consequences:

"This backfired when customers asked legitimate questions about integrating Microsoft Teams with Salesforce," says Mr Inzerillo. "The agent refused to help because Microsoft appeared on our competitor list."

Salesforce reports that 94% of customers offered the choice opt to interact with AI agents, that some deployments show customer satisfaction rates "in excess of what people get with humans," and that the firm has reduced customer-service costs by about $100m. On headlines about job cuts, the company said: "A very large percentage of those people got redeployed in other areas around customer service."

People working in or managing customer-service teams note there will still be situations where human contact is preferred or necessary. Fiona Coleman, who runs a firm that uses AI to give call-centre workers more flexible shift patterns, emphasises that some interactions require human judgment:

"There are times where I don't want to have a digital engagement, and I want to speak to a human," she says.

"Let's see what it looks like in five years' time - whether an AI can do a mortgage application, or talk about a debt problem. Let's see whether the AI has got empathetic enough."

Regulators and lawmakers are already responding. Legislation currently proposed in the United States would require companies to disclose when callers are speaking to AI and to transfer callers to a human on request. Separately, Gartner has suggested the European Union may mandate a consumer "right to talk to a human" as part of future protections.

Organisations are experimenting widely with customer-facing AI, seeking gains in speed and scale while trying to avoid errors and reputational risk. Analysts warn that costs, training-data needs and knowledge-management efforts are substantial, and some experts and managers say there will remain occasions when customers demand human interaction. As deployment grows, legislative and consumer-protection questions are also shaping how and where AI agents are used.