AI Voice Agent: How to Build One With No Code on ElevenLabs

An AI voice agent is software that picks up the call, answers the question, books the slot and updates the record — with no person in the loop. I spent an afternoon building one on ElevenLabs, with no coding at all, and by the end of it the agent was fielding questions about courses, fees and admissions the way a trained human rep would. My honest read is that most sales calls will be handled by AI voice agents within a few years. That is not a science-fiction prediction. The tooling to build one is already sitting in front of you, and it takes about as long as a decent lunch break.

What follows is the whole build, step by step, plus the part almost nobody explains properly: what this is actually worth once it works.

What an AI voice agent actually does

The most useful way to think about an AI voice agent is not as a chatbot with a voice bolted on. It is closer to a digital employee — a role you define, staff and deploy, rather than a feature you switch on.

The four jobs it takes over

A voice agent handles four kinds of work straight out of the box:

• Books appointments — it takes the call, finds the slot, and confirms it

• Answers customer queries — the same twenty questions your team answers every day

• Updates CRM systems — so the record reflects the conversation without anyone retyping it

• Sends emails — the follow-up that usually gets forgotten at 6pm on a Friday

Individually, none of those is glamorous. Together they are most of what a front-line rep does on a normal day.

Why businesses are adopting it now

The reason this is spreading quickly is not that the agent is smarter than a person. It is that it acts exactly like a human rep while taking the repetitive work off the roster, and it does that around the clock. A human rep answers the fees question fifty times a week and gets slower each time. An agent answers it the ten-thousandth time exactly as it answered it the first. Businesses are adopting AI voice agents rapidly for that single reason: it is a straight productivity gain on work that was never a good use of a human.

Step one: sign up and choose a business agent

The build starts on ElevenLabs. You sign up, and the first real decision is choosing a business agent rather than a general-purpose one. That choice matters more than it looks, because it sets the frame for everything after it — a business agent is scoped to represent an organisation, which is what makes the later steps (knowledge base, tools, deployment channels) coherent rather than a pile of settings.

Step two: pick the use case

Next you select the use case. ElevenLabs gives you concrete options rather than a blank canvas, and the three worth knowing are:

• Customer support — inbound, reactive, high volume

• Outbound sales — the agent initiates

• Career guidance — advisory, question-led

Pick the one that matches the job you are actually staffing. This is the step people rush, and it is the one that quietly determines whether the agent sounds like it belongs to your business or like a demo.

Step three: add a knowledge base and tools

This is where an AI voice agent stops being generic and starts being yours. It comes in two halves, and they do different things.

Knowledge base documents give it context

You upload documents — the material that describes your business. This is what the agent answers from. Without it, the agent is a well-spoken stranger guessing at your pricing. With it, the agent knows your courses, your fees and your process, because you handed it the same documents you would hand a new hire on day one.

Tools give it the ability to act

Tools are the integration layer. They are what let the agent do something rather than just say something — reach into a system, take an action, and come back. Knowledge base without tools gives you an agent that can explain. Tools without a knowledge base gives you an agent that can act but has nothing accurate to act on. You want both.

Step four: customise the voice, and pick the brain

Two settings, and they are genuinely independent of each other.

Voice, language and accent

You customise the agent's voice, with multiple language and accent options available. For a business serving more than one region this is not cosmetic — an agent that answers in the caller's language and a familiar accent gets treated as staff rather than as an automated system, and it is a dropdown, not a project.

Choosing the LLM

Separately, you pick the LLM model behind the agent: OpenAI, Claude or Google. This is the reasoning layer — how the agent thinks about what it has been asked, as distinct from how it sounds saying the answer. The fact that it is a swappable choice rather than a fixed dependency is one of the better things about building an AI voice agent on a platform like this: you are not locked to one provider's model behaviour.

Step five: publish and deploy

Finally you publish, and deploy on channels — WhatsApp, Slack, or a telephone line.

That list is the part worth sitting with. Deployment is not "we built an app, now please download it." The agent goes where the conversation already happens. A clinic's patients already have WhatsApp. A company's staff are already in Slack. Everyone, everywhere, already has a phone. There is nothing to install and no new habit to teach anyone, which removes the single most common reason internal tools go unused.

What happened when I tested my own agent

I built a career guidance agent and then tried to break it. I asked it about courses. I asked it about fees. I asked it about admission.

It answered all of it — cleanly, in context, the way a real human agent would have. That test is the reason I stopped treating this as an interesting demo. The gap between "impressive" and "usable" in AI tooling is usually enormous; here, for a well-scoped advisory role with a good knowledge base behind it, there wasn't one.