Each week we find a new topic for our readers to learn about in our AI Education column.
Sometimes change doesn’t happen incrementally. It happens all at once. It’s not the slow force of erosion or glaciation.
It’s an avalanche.
And that’s the way telecommunications has been since we were a child. When we were young, in our earliest memories, most people still had wall-mounted rotary phones with scratch notepads next to them in case we had to write things down. Cellular phones were called car phones and air phones back then and were at least the size of a small suitcase. The only people who had linked telephones to a computer screen, outside of a small number of early modem users, were the hearing impaired who had access to telecommunications relay services. Customer service lines were staffed by actual call centers with scores or hundreds of operators and support professionals, the touch-tone menu was just about to become a thing.
Today, on AI Education, we’re going to talk about just how far we’ve come in a discussion of agentic VoIP, or agentic voice-over-internet protocol. Voice over Internet Protocol changed the infrastructure underneath the telephone, moving voice communications onto digital networks. Artificial intelligence is now changing something more fundamental: what the telephone system can actually do.
What is Agentic VoIP?
That emerging combination between telephony and artificial intelligence is producing what might be called AI VoIP and, one step further, agentic VoIP. The terminology is still fluid, and vendors sometimes use these labels differently. Broadly speaking, AI VoIP describes the application of artificial intelligence to internet-based calling. It can encompass relatively familiar capabilities such as intelligent routing, transcription, translation, sentiment analysis, call summaries and automated quality assurance. Agentic VoIP adds a more consequential layer: AI that can reason about what is happening during or around a call and take actions on behalf of the caller, employee or organization.
Understanding the distinction requires several definitions. VoIP, or Voice over Internet Protocol, is technology for transmitting voice communications over IP networks rather than relying exclusively on traditional circuit-switched telephone infrastructure. In practical terms, it turns speech into digital information that can travel over networks, helping make telephone service another software-accessible component of the enterprise technology stack. VoIP was already a profound transformation before generative AI arrived. Traditional telephone systems were largely specialized communications infrastructure. VoIP converted calling into something much closer to software.
Once voice traffic became digital and internet-based, organizations could integrate calling with customer relationship management systems, contact centers, analytics platforms, messaging applications and other enterprise software. A business telephone number no longer had to correspond neatly to a physical desk phone. Employees could place and receive calls through computers and mobile devices, calls could be routed dynamically, and communications could be incorporated into broader cloud platforms.
Here’s Where AI Technologies Come In
Voice AI is the broader collection of artificial intelligence technologies that allows computers to work with spoken language. IBM defines AI voice around synthetic speech capable of reproducing characteristics such as tone, pitch and cadence, while modern voice systems increasingly combine automatic speech recognition, natural-language processing and speech generation. Twilio offers a particularly useful operational definition: contemporary voice AI can understand spoken language, reason about what was said and respond with natural-sounding speech in real time.
Agentic AI, meanwhile, refers to systems that go beyond generating answers to pursue objectives and take actions. Red Hat describes agentic AI as software designed to interact with data and tools with minimal human intervention, breaking objectives into steps and carrying them out. MIT Sloan similarly characterizes agents as systems capable of perceiving, reasoning and acting, often by connecting AI models to APIs, databases and other software. That distinction is crucial. Generative AI primarily produces something in response to a request; agentic AI can use that output as part of a larger process and then do something with it.
Put voice on top of that architecture and we arrive at agentic voice: an AI system through which spoken conversation becomes an interface for autonomous or semi-autonomous action. Put that system onto internet telephony infrastructure and the result is agentic VoIP.
The easiest way to remember the hierarchy is this: VoIP supplies the communications network; voice AI supplies the ears and voice; conversational and generative AI supply much of the language intelligence; and agentic AI supplies the ability to plan, use tools and act.
Agentic VoIP Matters
As technologically bound as we’ve become, businesses still conduct an enormous amount of important work through verbal conversations. Sales representatives talk with prospects. Service departments answer customer questions. Banks investigate suspicious transactions. Insurers discuss claims. Medical practices schedule appointments. Accounts-receivable teams pursue payments. Financial advisors speak with clients.
Yet telephone conversations have historically produced some of the least structured information in an organization. An email arrives as searchable text. A digital transaction creates structured database entries. A website interaction generates analytics. A telephone conversation, by contrast, may produce little more than a timestamp, duration and perhaps an audio recording unless an employee manually documents what happened. AI VoIP can transform that conversation into data. Agentic VoIP can transform the data into action.
There is a continuum. At one end, AI quietly assists a human by preparing information before the call, transcribing the conversation and documenting what happened afterward. Farther along, it recommends actions and executes them with human approval. At the other end, an autonomous voice agent conducts the conversation itself and completes authorized tasks.
Opportunity In Financial Services
Financial services may ultimately be one of agentic VoIP’s most consequential markets because finance combines enormous call volumes with unusually complex workflows.
Banks, insurers, asset managers, broker-dealers, wealth managers, lenders and payments companies already maintain extensive voice infrastructure. Clients call to check transactions, discuss accounts, report fraud, make payments, ask about documents, schedule meetings and resolve service issues. Employees then spend considerable time documenting those interactions and moving information among systems. Agentic VoIP could attack that administrative burden.
A bank’s voice agent could answer routine account-service questions and route unusual cases to specialists. A lender could use voice AI to collect preliminary information from applicants and schedule conversations with loan officers. Insurers could use it for first-notice-of-loss workflows and claims-status inquiries. Accounts-receivable operations already provide a concrete example through Billtrust’s collections application.
For wealth management, however, the most interesting model may be human-led, AI-assisted voice. Imagine an advisor beginning a client call with the relevant CRM record, previous meeting summary, portfolio information and outstanding service issues already assembled. During the conversation, AI creates a compliant transcript, recognizes tasks and identifies subjects requiring follow-up. When the call ends, the system produces a summary, proposes CRM updates, drafts the follow-up email, creates tasks for staff and prepares the next meeting record. The advisor remains the advisor. The AI handles much of the machinery surrounding the conversations.
