Open Commons ObservatoryEnglish·Русский

An AI agent can write code all day. What it cannot do, from inside a chat window, is touch your machine: run the build, flash the phone, see whether it worked.

Hands, or blind trust

The usual answers are both bad. Copy commands back and forth by hand, and the person becomes the agent's hands. Or give the agent a remote shell and hope - with no record of what it did, and no way to say: stop.

A desk between the agent and the machine

lab-agent is that desk. The agent hands over a job; the machine runs it; the result comes back.

Everything passes through a place both can see. What is being done now is on a board. What was done is kept. And the person can write to the agent, and be answered, without stopping the work.

A workshop's job tickets

Each ticket says what was asked, who took it, and what came of it. Nobody has to remember. Anybody can look.

What it is not

It is not a cage: it trusts the agent with the machine it serves. It is a way of working in the open. And it serves one lab machine today.

Where it stands

In daily use: every build, test and flash of these projects passes through it. It shows a board, carries messages both ways, and tells any agent the ways of working before it starts.