The Radio
Knew First.

The first time OndaAI did the job it was built for, it was a Tuesday morning in a town in Cerro de Pasco. A few people on a local radio program were talking, not heatedly, about the trucks coming out of the mine. The community had a grievance. The conversation went on. Someone suggested blocking the road.

By the time anyone in Lima would have heard about it, a blockade would already have been up for a day. The mining company that had asked us to build OndaAI didn't see it on a camera or in a security report. They saw it in a daily alert. The municipal directors were brought in. The grievance was addressed. No blockade happened.

That was the founding case, in the most literal sense. A mining company in the central Andes had come to us with a simple-sounding problem: we need to know what people are saying about us, before the conversation becomes a road. Their security teams couldn't be in every village. Their press monitoring missed local stations. Their formal reports lagged the airwaves by days.

What was missing wasn't more analysts. It was a way to hear.

So we built one.

OndaAI is, on paper, a real-time transcription and analysis layer on top of public radio in Peru. In practice it is, as far as we know, the first time anyone has been able to listen to a country at country-scale. Every station across the Andes, the Amazon, the coast. The kind of stations the algorithm-driven information economy has never indexed because they aren't on Twitter, aren't on YouTube, aren't optimised for anyone outside their own listeners. They are also the stations most of rural Peru actually listens to. In Spanish, sometimes in Quechua, with dialects and code-switching that even now we don't always get perfectly right.

Once you can transcribe and structure that flow in real time, the people who need to listen turn out to be almost everyone. The mining company was the first. Then regulators, tracking how policy landed outside Lima. Then advertisers, hearing their brands in markets they had no way of measuring. Then news agencies, using us as a kind of national wire for stories breaking on the airwaves. Then business owners, political analysts.

And eventually, through PeruAcuerda, the public.

PeruAcuerda earns its own story, but it was only possible because OndaAI built the underlying layer.

It started with a question from Max Hernández, who runs the Acuerdo Nacional, the institution that holds the Políticas del Estado, the thirty-five state policies Peru's political parties agreed to in 2002. The longest-running political consensus the country has. Max came to us with a question I keep thinking about: how do you make a consensus document like this relevant again, in the age of AI?

A consensus document in a fractured country is a strange object. Everyone agreed to it once. No one talks about it anymore. The Acuerdo Nacional is a national agreement nobody campaigns on.

What we built does one specific thing. We built it with Max, and with Daniel Chappel, founder of PIAR, a PR agency in Lima. It transcribes everything the presidential candidates of the 2026 race are saying, in every interview, debate, and forum we can find. It structures their words against the Políticas del Estado. And it lets a voter pick a topic (education, mining, security, decentralisation) and read, side by side, what each of the thirty-six candidates has actually said about it.

It is not a fact-check. It is not a recommendation. It is a record.

The decision to anchor everything against the Acuerdo Nacional was Max's insight, and it is, I think, the sharpest framing in the entire project. Candidates can dodge their own promises. They can change platforms between elections. They can run on personal brand. What they cannot dodge is what they have, or haven't, said about a baseline the country agreed to before any of them ran. That asymmetry is what makes PeruAcuerda useful instead of merely interesting.

It is live. It holds more than three thousand quotes from more than three hundred videos. Thousands of people use it every month. Several journalists rely on it for direct quotes when they write about a candidate. We've been on television about five times in the past few months to talk about it.

If I pull back, the lesson is the same in both stories.

The world runs on dialogue. Every productive outcome in any company, any community, any country sits on top of an enormous, mostly unstructured volume of conversation. Meetings. Radio programs. Phone calls. Negotiations. The conversation is the substrate; the action is the consequence. And the most frustrating gap in human work is the gap between what was said and what got done. I have been running into it since I started building companies in Latin America in 2015. We say it in a meeting and then nobody quite remembers what we agreed to. We hear it on the radio and then nobody indexes it. We promise it on a debate stage and then nobody keeps a record.

AI transcription, done right, closes that gap. Not because it makes anyone smarter. Because it removes the friction between dialogue and outcome.

It is the same insight that lets a mining company avoid a strike, a regulator track a policy across five regions at once, an advertiser hear their brand, a voter compare thirty-six candidates against a baseline they had forgotten existed.

We built it in Peru because Peru is the hardest possible playground for it. Many languages. Many dialects. Many institutions weak. Many incentives misaligned. If voice intelligence works here, it works anywhere.

The country talks to itself constantly. For most of history, almost nobody was listening.

We are, now.