Teaching you toprompt it yourself!
Menu

Four hours. 3K leads.

A sales channel of his own, built in an afternoon: around three thousand organisations that commission work like his, a scoring survey on his own site, and a domain bought live to run it from.

Where it startedIstvan has projected light onto buildings for twenty years, and the work always found him. He came on for sales, the thing his studio had never had to do.

He asks AI for ten designs per question. Istvan has projected light onto buildings on four continents since 2006, and the work always came to him. He came on to turn that around. Twenty years in, his studio had never done sales.

The first thing you see him do is the move he found on the day: ask for ten versions of a thing, pick one by number, ask again. You would never ask a designer for ten goes. The agent does not mind, and the asking costs nothing.

Finding new leads. Who commissions projection mapping? He did not know, and neither did I. So he gave the agent his website and one plain sentence, as the owner of the company, asking it to review the site and suggest how to reach his customers better. No technique, no structure.

It came back with the shape of his market, the competitors, and the kinds of organisation that buy this: festivals, cities, brands, venues. From there it built a library of around three thousand of them, with a line each. When it answered from memory instead of looking, he told it to go and look. That corrective worked every time.

A survey that filters. The contact form on his site asked nothing and got nothing. He typed one line, that he would like a score app like a survey on his website, and gave it the address. It built a survey that branches: what you have in mind, where, when, and a budget band at the end, so a small good lead is not frightened off by a big-budget question first.

His verdict mattered more than the tech. Someone arriving on the site finally understands what they are looking for, and it gives a professional image. By the end it was deployed behind an email login, with a status and comments on each lead, so a sales team could work in it.

Rebuilding the website. An hour before the end he bought a domain, card off screen, and the agent rebuilt the site on it from what it had learned about his customers: light and dark, one page, a first paint that no longer took eight seconds, checked on air.

He said the mock-up built in an afternoon was close to the one his agency spent a year on. Uncomfortable, and the most honest thing anyone has said on the show about what this costs.

He had already had a go before the show. In the days before, he let an AI agent take over his computer, and some moments of it were scary. Not long ago he was the one refusing it, afraid it would destroy what is human. He still wants it nowhere near his artists. Sales was another matter.

Within the first hour he had two agents working side by side, one reading his market and one building a survey. Later they started helpers of their own, and he said he had a team now. Again and again I called its work lazy and sent it back. He was still amazed: this was research he could never have done on his own.

The turn came late in the session. I asked the agent for ten designs for every question in his survey.

I was laughing when you said give me 10 variations of each. I could never ask a human designer for 10 variations. They would kill me. Are you crazy? Three maximum, but two would be better.

Istvan

Then he noticed something about himself: he treats the people he works with as a great team, and here he was ordering a machine about like a rude man in an old Brazilian soap opera.

At the end he was the one who corrected me. It was four hours, not three. And he could tell his friends he was actually coding.

Where it ended3,000 sourced leads sorted by when the client chooses a supplier, a survey that scores who asks, and his site rebuilt on a domain bought on air.