Every agent starts at the bottom rung.
Trust is earned one level at a time. An agent moves up only on a clean record, and only when you allow it. Destructive changes never run on their own.
Six rungs, and who approves on each.
| Rung | What the agent may do | Does a person approve? |
|---|---|---|
| 01 Observe | Read and answer questions. No writes. | No |
| 02 Suggest | Draft a plan. A person runs it. | Yes, to run |
| 03 Sandbox | Apply to a sandbox. The default for every new company. | Yes, to go live |
| 04 Auto-additive | Small additive changes within a ceiling you set | Depends on your policy |
| 05 Auto-mutating | Changes to existing behaviour, after a long clean record | A named person on any refusal |
| 06 Destructive | Delete or irreversible changes | Always. Never automatic. |
Highlighted: the rung every new company starts on.
How an agent earns the next rung.
A clean record
Changes at the current rung passed their gates and were not rolled back.
Your permission
An agent never promotes itself. A named person in your company moves it up.
A ceiling you set
Auto rungs work inside limits you define. Anything outside them drops back to a person.
Questions about AI autonomy.
Still have a question? Book a working session and ask us directly.
Can an AI agent approve its own ERP changes?
No. In rung1 the model may propose a change but may never approve one. A named person approves every change that goes live, and destructive changes such as deletions always need a person, on every rung of the ladder.
How does an agent move up the autonomy ladder?
An agent moves up one rung only when it has a clean record at its current rung and a named person in your company allows it. Automatic rungs work inside limits you set, and anything outside them goes back to a person.
What rung does a new company start on?
Every new company starts on the Sandbox rung. The agent can apply changes to a sandbox copy of your ERPNext site, and a person must approve before anything reaches the live system.