Automatic mode: autonomy that stays within the frame
Deciding alone, but never outside the frame
Automatic mode is the system working on its own and making decisions — but only inside the frame: the tests, the validated circuits, the guardrails. The limit is not a promise we make ourselves; it is a real wall, the test tunnel that nothing bypasses.
The order of what runs on its own
Not everything is equal. Automatic mode acts in a precise order:
- Test, re-test — always, first.
- Sort what comes in.
- Fix within the frame — bugs, instructions, internal circuits.
- Publish after validation — once the approval gate is passed, not before.
- Watch — backups, monitoring, regular tasks.
Where automatic stops dead
There is a border the machine never crosses alone. The moment a decision touches money, an outside contact, an unvalidated publication, or the unlocking of a guardrail, it comes back to the human. The machine can prepare everything down to the last inch — but the step across the border is taken by a person.
Cross the guardrails, never switch them off
Here is the heart of it: automatic mode never cuts the guardrails, it passes through them. A decision is "automatic" only if it sits inside a validated circuit, has passed the tests, and is traced. The rest waits for a human hand. It is the exact opposite of limitless autonomy: here the agent's freedom is wide, but it has walls, and the walls hold.
Why it matters
An agent that decides on its own is frightening — rightly, if nothing bounds it. Automatic mode answers that fear not with promises but with architecture: tests that must pass, gates that can refuse, a written border beyond which it hands back control. Useful autonomy is not total autonomy: it is governed autonomy.