How AI is changing the way voice agents are created, calls are automated, and phone customer support is built without technical configurations.
Two years ago, turning a software idea into something real meant one of two things: learning to code or paying someone who knew how. Today you go into Lovable, type "I want an app to manage my gym bookings," and after a while you have a working application. You haven't touched a line of code, you simply described what you wanted and the AI built it.
At Diga we have brought that same idea to voice agents. In this article we explain how voice agents were built until now, what changes when you create them by talking to a copilot instead of configuring them by hand, how far this approach goes, what issues are still open, and why we believe this is how next-generation voice agents will be built.
The underlying shift: no longer coding, but describing
Many of you will know about vibe coding, the idea of programming simply by sending prompts to an AI. In a matter of months it stopped being a Twitter meme and became the method that many people follow when they are creating or developing a product.
Lovable became one of the fastest-growing software startups in history, from zero to 100 million dollars in recurring revenue in eight months and a valuation of 6.6 billion by late 2025, because it tore down a requirement that had stood for decades: to make software, you had to be a programmer. Today, hundreds of thousands of projects are launched daily on its platform, many of them by people who have never written a line of code. And it is not the only example; we can find other similar platforms like Cursor, v0, or Replit.
But the important change is not Lovable, Cursor, or v0. It is the interface change. For fifty years, asking a computer for something new required an intermediate language: code. AI made that language unnecessary; now you describe the intent and the machine handles the execution.
Apps were only the first domain to fall; the next are voice agents.
How AI voice agents were built until now
Setting up a voice agent that answers your business phone has never been a single afternoon of work. You have to configure the telephony and connect a number. You have to write the prompt that defines how the agent speaks and what it does. You have to connect the integrations, since we want it to be able to edit the CRM or send an email. If we want it to have information about our business, we have to select the documents it needs to have in its context. And then test, listen to calls, adjust, and repeat.
None of this is impossible, but currently, to have a voice agent, you need a person who knows how to set up the prompts, the telephony, and the integrations.
Melo: you describe the agent, and it gets built
So, who is Melo? Melo is our copilot, and it is what happens when you apply to voice agents the same idea that Lovable applied to applications.
"I need an agent that answers calls for my clinic and schedules appointments." Just tell Melo this and he chooses the voice, writes the agent's instructions, sets up the call flow, and leaves it ready on a phone number. You simply describe, review, and confirm. What we described in the previous section is still necessary, it's just that now Melo takes care of everything.
What you can ask Melo to do
Creating the AI voice agent is just the beginning. From there, expanding it is just a matter of continued talk about the capabilities your agent needs.
"I want it to know our hours and relevant clinic information." You pass it the PDF, or drop the entire website, and from there the agent replies with your real information instead of improvising. This is possible thanks to RAG systems.
"I need it to look up the customer's order in our system during the call." Melo connects it to your CRM or your database, and the agent consults, creates, or updates data mid-conversation, and gives the customer their order number without anyone having to hang up and call back.
"At the end of the call, it should send me a transcript by email and log the information in an Excel sheet." Melo builds that automation with the tools you already use like Gmail, Slack, your spreadsheet, your calendar, or your CRM.
And when something doesn't go as expected, you can ask Melo: "Why did this call fail?" He will open the log of what happened, what the agent decided, what tool it used, where it got stuck, and what the transcript says.
Even the boring stuff is handled via chat: buying a number, seeing how many minutes you have left this month, inviting someone from your team. What in another tool would be five screens, here is just one question.
What Melo does not know how to do yet
Melo is still not able to do everything; as you increase the complexity of what you ask, you can see where his limits are.
Ask him for an agent that takes calls and books appointments, and you have it in minutes. Ask him for a three-level support flow, with escalation depending on the incident category, warranty validation against two different systems, and varying rules by country, and things change radically. Melo will build the skeleton, but you will need to fine-tune each branch of the conversation flow. Defining a brand's exact tone or managing delicate and complex conversations is still iterative work where you must run tests, listen to calls, and correct.
It is a similar story with integrations. Connecting Gmail, a calendar, or a well-known CRM is simple. Connecting a proprietary ERP still requires a person who knows that system from the inside. Melo prepares the ground; truly custom cases still demand expert hands.
There is a range of complex problems that Melo cannot solve on his own today. What he does do, however, is serve as a foundation to build agents in Diga.
Why we believe this is the way
The pattern is the same everywhere and there is no turning back: creating ceased to be a matter of knowing how to build and became a matter of knowing what you want to build. It happened with software, it is happening with voice agents, and the implication for a business is that the advantage is no longer "having someone who knows how to set it up" but rather "knowing what you want your agent to do."
That is why we built Melo. For the agency setting up agents for twenty clients without a technical team behind them. For the restaurant, the clinic, or the real estate office that knows perfectly how they want to answer the phone but has no reason to know about prompts or telephony. We prefer that part to be handled by a copilot, leaving you to focus on your business, which is what you actually know about.
The future of Melo
Today, you ask Melo and he builds. Where he is heading is toward doing more and more without you having to ask in such detail.
First, achieving what we mentioned earlier, meaning that tangled flows and custom integrations no longer require so much from you. Each version of the copilot tries to improve a bit in that regard.
Second, and what matters most to us, is that Melo does not stop at creating the agent but accompanies it. That he looks at how real calls are going and tells you, "this agent hangs up a lot during the payment step, shall we adjust it?" or "half of the people ask about the hours and you don't have it in the knowledge base." Moving from a copilot that builds when you ask to one that observes, proposes, and improves with you.
And ultimately, that more and more of the product fits into the conversation. That what are menus, tabs, and forms today becomes simply telling Melo what you want: creating, publishing, reviewing, and improving an agent, all by talking.
There is still a long way to go for all of that, and we are on our way. But the fundamental shift has already happened. Before, to have a voice agent, you needed someone who knew how to set it up. Today, you only need to know what you want it to do.
