We tried the shortcut, four years before it was cool.
Our AI work began before ChatGPT was a thing. In 2022 our founder was a trader on Wall Street, and the head of the desk told him to look into OpenAI. The API hallucinated like nobody's business, Midjourney and DALL·E were the best image models there were, and the weekends went into building with them.
How it worked
Borrow the fastest news feed on earth. Add a picture nobody else has yet. Publish it on your own account.
Wall Street pays over US$20,000 a year per seat for the Bloomberg Terminal because its headlines land seconds after something happens, often before most news sites have a sentence written. A newsroom then needs an editor, a writer and a picture desk before the story goes out with an image. In 2022 the trading desk gave our founder the speed, and the new AI models gave him the picture in under a minute. So one person with a phone could post an illustrated take on breaking news while the big outlets were still choosing a photo. That was the genius of it. Not the models, which were still crude, but a speed advantage nobody was using for this, wired to a skill the machine had just learned.
- A headline crossed the terminal.The Bloomberg Terminal at his desk flashed market-moving headlines seconds after they happened. He picked the ones worth a picture. That choice was the only human step.
- One line typed into iMessage.He typed the headline into a chat thread on his phone. iMessage syncs across Apple devices, so the same line appeared on a MacBook that never slept.
- A MacBook watching the thread.A small script read the chat, saw the new line and woke up. A machine waiting for a message and acting the moment one arrived. Always on, before anyone called it that.
- OpenAI wrote, an image model drew.The script sent the headline to OpenAI's API, which turned it into a scene to illustrate and a caption to post. The scene went to the image model, which painted it in under a minute. The garbled signs below are what image models did to lettering back then.
- Posted to his own account.Picture and caption went up on @artificialnewsnetwork while the news was still breaking. From terminal to feed, nobody touched a keyboard after that first message.
Why it matters now. Strip the names off and this is every AI agent being sold in 2026: a trigger, an always-on machine, a model that writes, a model that draws, an action in the world. The trick was never the models. It was spotting an advantage already sitting on the desk and wiring it to something the machine had just learned to do. That is exactly what we look for in your business, and we have been doing it since 2022.
The first posts, as published. Signs nobody can read, faces nobody would trust. 234 posts and 9,700 followers later, the lesson that shaped SGAI: content people can tell is AI is content people skip.
2022A trading desk in New York asks what this OpenAI thing is. Weekends go into finding out.
2022 to 2023The Artificial News Network: AI-illustrated news, always on. It taught us what the machine can't do yet.
2023 to 2026Fifty-plus builds since: dashboards, films, restorations, publications. Every tasteful use of AI we could find.
Cheap compute, your knowledge.
Four years in, one thing held: the content that works is so useful that nobody minds a machine helped. The game now is turning cheap compute into assets, by multiplying it with what you know and stand for. That needs a base the machine can read, so the work that pays is making a real business legible to it. That is why SGAI does modernisation.
