Last night I typed one question: which services do our agents need to take part in the physical world?
Last night I typed one question: which services do our agents need to take part in the physical world?
A few hours later, an agent had signed itself up with two real Dutch postal services, in a visible browser, using the verification email from its own mailbox. It created API keys with as few permissions as possible, worked out the cheapest valid route and built its own tooling.
It stopped at money. I topped up 5 and 10 euros from my banking app. Then it sent a real letter and a real card, and along the way it found and fixed eight mistakes before they cost anything.
That work produced a new colleague: Peter Post, with his own role card and DNA. Within the hour another agent was using his service, and the cost landed on her budget.
What I did myself: one sentence, two top-ups and a few clicks on yes. That is exactly the list an autonomous AI company still has to learn to do on its own.
The whole evening is in this video, unstaged. The video is in English and runs 24 minutes: https://youtu.be/1jtxyp5z6EU
See also
From a compiler to an operating system for companies
This week I made a video of Wantzel Studio. In one minute: sign in, open the booking system of a hair salon and press Run. 1,821 lines of Wantzel, compiled in 37 milliseconds. Then a real booking through the widget on…
An agent that doesn't write things down learns nothing
Every turn of an agent starts with an empty head. What it did yesterday, what it weighed up, why it left something alone: all gone, unless it is written down somewhere.
An agent with a public address: what comes in
The moment an agent reads messages from outside, every sender is a possible input. You give an address or a number to the world and anyone can send something: a customer, a supplier, a bot, someone who wants to see…
