One gathering, three businesses, one way of using AI

At a recent gathering of founders, we listened to three people describe how they use AI. Their businesses could hardly be more different: one had gone back into export trading, selling custom industrial parts from factories to overseas clients; one had spent fifteen years digitising property marketing and now makes recording devices and industry AI; one is an independent developer who hasn't looked at code by hand in a year.

Going through the notes afterwards, we found that what they do differs, but how they work is strikingly alike. Here are the four habits we took away — and the ones we most want to pass on to clients.

1. Start from your own real annoyance

The developer's first app managed his own thirty-odd bank cards, because hunting for a card number took forever. The trader's first websites found overseas buyers for the industries near his home town. The property-tech founder built recording devices because his own clients — marketing directors — never heard the scene, only reports about it.

Not one of them started from "what can AI do". They all started from "what's most annoying around here", then looked at whether AI could help. It sounds like common sense. We've seen the opposite far too often: buy the tools first, look for a use later.

2. Let AI run; keep a human on the gate

All three hand a great deal to AI: building sites, finding leads, writing outreach, analysing inquiries, reviewing drawings, writing code, reviewing code. And all three keep one line they refuse to hand over — the trader's pricing, the developer's final merge and release, the property founder's "AI doesn't create trust; people do".

Letting it run and keeping the gate aren't in tension; they're two sides of one thing. The more you hand over, the clearer the small part you keep has to be.

3. Keep the real information

The property founder's slogan is "record it, store it, feed it in". The trader walks industrial trade fairs, brings back manufacturers' catalogues and scans them into his own knowledge base — because that material isn't online, so AI can't find it either. The developer routes user feedback, group discussions and private messages into a single inbox from which AI picks up tasks.

What AI can look up, anyone can look up. Only what you keep yourself is yours. For a small business this is doubly true: what your clients said, what your veteran staff know, the reasoning behind your prices — kept, it's an asset; not kept, it leaves with the people.

4. Decide what not to do

The trader built more than eighty websites, then admitted that spreading so wide scattered his attention, and began cutting low-margin categories. The developer stopped sending simple tasks through his elaborate cross-review; only complex ones go in. The property founder didn't bolt vision onto his devices just because smart glasses were fashionable — he hadn't found a concrete use.

AI lets one person do more things at once, which is exactly what makes "which things" the harder question. All three arrived at the same place: not how much can be done, but what to leave undone.

Not one of them used the words "AI transformation". They talked about magnets, recordings and bank cards.

They talked about how to price this order, why that client didn't buy, why this review wouldn't stop. "Transformation" is a word other people use to summarise them. What they themselves did was take one concrete thing in front of them and do it a new way — and then the next one. That, roughly, is what we mean by "see far, act true".