AI agents in customer service have quickly moved from experiment to everyday tool. According to Salesforce's State of Service research, published in May and based on 3,075 customer service professionals, 66% of service organizations already use AI agents. A year earlier the figure was 39%.
When a service team adopts an AI agent, it has to decide how many requests to hand over and when to pass the conversation to a person. It also has to decide how the work of its people changes. We discussed this during the Athics webinar "AI Customer Experience" with Letizia Olivari, Director of CMI – Customer Management Insights, and Luigi Marzorati, IT Director of Vega Carburanti.
In this guide you will find their insights, the story of Bruno, Vega's virtual assistant, this year's numbers on adoption and on the role of human agents, what the AI Act requires, and five steps to start a project.
Table of contents
- From reactive to proactive customer care
- How fast has this change happened?
- Bruno, Vega Carburanti's virtual assistant: 70% fewer human interventions
- Do AI agents make customer care more or less human?
- How widespread are AI agents in customer service
- Key figures for 2026
- AI Act: transparency obligations for AI agents
- How to implement an AI agent in customer service
- Conclusions
- FAQs
From reactive to proactive customer care

The traditional customer care model is moving from reactive to proactive. Answering quickly is still necessary, but the most advanced companies try to anticipate what the customer needs and offer a solution before the request arrives.
This shift is possible thanks to AI agents for customer support: systems that carry out tasks on their own, make decisions and work with company systems, for example by reading a contract in the CRM or opening a ticket. If you want to understand how they differ from a rule-based assistant, see our guide on chatbots and AI agents.
How fast has this change happened?
Within a year. According to Salesforce, the share of customer service organizations using AI agents went from 39% to 66%. The change is not moving at the same pace everywhere, though.
As Letizia Olivari put it: "The change is happening at two different speeds: customers increasingly expect immediacy, personalization, and a seamless experience, while many companies still struggle to integrate information and make the most of available data. Proactivity, however, is a must-have, as customer care is a strategic touchpoint for customer satisfaction and a necessary step in transforming it into a profit center."
"Customer service has become the heart of the customer relationship," said Luigi Marzorati. Besides solving problems, it is where value is built and relationships get stronger. In the sector where Vega Carburanti operates, for example, changing the daily spending limit on a fuel card often triggers several customer questions that need answering before the change is complete. A proactive AI agent can use that moment to suggest new products, such as cards with higher spending limits or features that anticipate the customer's needs.
Bruno, Vega Carburanti's virtual assistant: 70% fewer human interventions
A concrete example comes from Vega Carburanti, which has deployed Bruno, the AI agent that answers the company's customers. Bruno combines machine learning and generative AI and handles the most frequent requests: information about the Vega card, billing and card replacement. It is integrated with the corporate CRM, so beyond administrative questions it can give personalized suggestions and product information. Every support conversation becomes a sales touchpoint as well.
When a request goes beyond what it can handle, the customer can ask to move to a live chat operator or leave their details to be contacted later. The result measured by Vega is a 70% decrease in requests for human intervention: agents get fewer repetitive questions and more conversations where a person is really needed. The project details are in the Vega Carburanti success story.
This is where the real step forward lies: the agent knows the customer it is talking to and can anticipate their needs.
Do AI agents make customer care more or less human?
From a technology point of view, Letizia Olivari believes AI can increase the human element in customer service, because it makes the work of agents easier. People can then focus on the customer relationship and on understanding what customers actually need.
In practice there are two obstacles. The first is the approach of companies that see technology as a way to cut staffing costs. The second is the training and relational skills of customer service agents, which need to be strengthened and rewarded.
This year's data points in the direction Olivari described. A Gartner survey of 321 customer service leaders, reported by No Jitter, found that 85% are moving agent time toward higher-value work such as customer loyalty and growth. Only 31% have made or plan AI-driven layoffs; 63% are reducing headcount gradually, by not replacing people who leave.
Customers want a balance too. Another Gartner survey of more than 3,500 customers, covered by CX Dive, found that nearly 90% say a company using GenAI in support must make a human agent available. Half of respondents, though, say GenAI makes their interactions easier. The lesson for anyone designing an agent is simple: customers are happy to use it as long as they always have a way out to a person.
In Vega Carburanti's experience, Bruno sped up the launch of new customer-facing services and works alongside human agents. The virtual assistant makes the same information available to internal staff, which makes complex technical topics easier to learn. For more on how people's roles change in a contact center with generative AI, read our article on the human factor in Generative AI contact centers.
How widespread are AI agents in customer service
Until recently the problem was adoption: many companies saw the value of AI, but few actually used it. Today the picture is different. Salesforce's State of Service: AI Agents Edition, fielded between March and April, shows that 66% of customer service organizations use AI agents, up from 39% the year before. 70% of adopters see measurable value within 60 days of deployment, and the first metric to improve is customer satisfaction, ahead of rep productivity and average handle time.
The challenge has therefore moved from adoption to project quality. Luigi Marzorati believes that, for small and medium-sized businesses, the hardest part is finding a partner able to guide them through AI adoption. Vega chose Athics, which followed the project from defining the use cases to launching Bruno.
Read Vega Carburanti's Success Story
Vega's experience shows that AI, when implemented well, makes people's work more valuable and gives agents time for the customer relationship. For a broader look at costs, response times and satisfaction, read our guide on AI agents for customer care.

Key figures for 2026
| Metric | Value | Source |
|---|---|---|
| Customer service organizations using AI agents | 66% (39% the year before) | Salesforce, State of Service |
| Companies seeing measurable value within 60 days | 70% | Salesforce, State of Service |
| Leaders moving agents to higher-value work | 85% | Gartner, via No Jitter |
| Leaders with AI-driven layoffs made or planned | 31% | Gartner, via No Jitter |
| Customers who want to always be able to reach a person | nearly 90% | Gartner, via CX Dive |
| Fewer requests for human intervention with Bruno (Vega Carburanti) | -70% | Vega success story |
AI Act: transparency obligations for AI agents
The transparency rules of the European AI Act apply from August 2026. As the European Commission explains, when people use an AI system such as a chatbot they should know they are talking to a machine, so they can make an informed decision.
For a conversational agent that supports customers the rule comes down to a few design choices. The agent introduces itself as a virtual assistant at the start of the conversation, in chat and on the phone. The customer can ask for a human at any time. When the agent hands over, the human agent gets the conversation history, so the customer doesn't have to repeat the request. The same principle applies to voice channels: we cover it in our guide on voicebot and PBX integration.
How to implement an AI agent in customer service
Here are 5 practical steps to launch an AI agent in customer support:
- Define the objectives: start from the needs of your customer service. Do you want to reduce response times, automate the most frequent requests or offer 24/7 support?
- Map touchpoints and FAQs: analyze recurring customer conversations and identify contact channels (website, WhatsApp, app, phone).
- Choose the right platform: look for platforms with no-code options for quick setup that integrate with CRM and ticketing systems. To estimate the investment, see our analysis of how much a chatbot costs.
- Train and test the agent: give the AI concrete examples of requests and answers and simulate real conversations.
- Plan the handover to a human, then monitor: decide when the agent should pass the conversation to a person, because nearly nine customers out of ten want that option. After launch, track resolution rate, escalations to live agents and user feedback.
Conclusions
AI agents for customer support are now mainstream: two organizations out of three use them, and most see results within two months. As Vega Carburanti's experience shows, the projects that work start from clear objectives, a partner who knows the processes and a real integration with the CRM and company systems.
This year's data also shows that AI agents work alongside people. Companies are moving agents to higher-value work, customers want to be able to reach someone at any time, and the AI Act requires disclosure when a machine is answering. If you want to see how this applies to your customer service, book a Crafter.ai demo.
FAQs about AI agents in customer service
What are AI agents in customer service? They are AI-based virtual assistants that talk to customers on their own. They answer frequently asked questions, handle simple requests, read data from the CRM and provide personalized support 24/7.
What advantages do they offer compared to traditional chatbots? Unlike rule-based chatbots, AI agents understand natural language, use the context of the conversation and can take actions, such as updating a case or opening a ticket.
What results did Vega Carburanti get with its AI agent? With Bruno, the virtual assistant integrated with the CRM, Vega Carburanti cut requests for human intervention by 70%. The agent answers questions about fuel cards and billing and hands the conversation to live chat when needed.
Can they completely replace human operators? No, their role is complementary. According to Gartner, 85% of customer service leaders are moving agents to higher-value work, while the AI agent handles repetitive requests.
Does an AI agent have to disclose that it is artificial intelligence? Yes. Under the AI Act transparency rules, which apply from August 2026, people talking to a chatbot or voicebot must know they are interacting with a machine.
How long does it take to implement them? It depends on the complexity of the project. An initial implementation usually takes from a few weeks to a few months, especially if data and touchpoints have already been analyzed.
What ROI can companies expect from AI agents in customer service? According to Salesforce, 70% of companies using AI agents in support see measurable value within 60 days, and the first metric to improve is customer satisfaction.
Related articles:





