Start safe
New agents draft and a person sends, so nothing goes out unchecked.
Autonomy levels
Freedom an agent earns, one level at a time.
Every agent starts with tight limits. As it proves itself on real work, you let it do more on its own.
- Draft only A person sends every result
- Approve each step The agent acts after a yes
- Approve risky steps Routine steps run on their own
- Run routine cases Sensitive actions still wait
Payments above threshold Always needs approval
Trust built on real work, not promises.
Open up what is proven
Routine cases that keep getting approved can run on their own.
Sensitive actions stay gated
Payments, contracts and external messages keep their approvals at every level.
Clear levels, clearly shown
Every agent shows its current level, from draft only to running routine cases, so everyone knows what it can do alone.
- Draft only A person sends every result
- Approve each step The agent acts after a yes
- Approve risky steps Routine steps run on their own
- Run routine cases Sensitive actions still wait
Payments above threshold Always needs approval
Move up on evidence
Approval and rework history shows how an agent performs, so the decision to give it more room is based on its record.
184 Runs
37m Turnaround
92% Approval clarity
$428 Model cost
Gates that never go away
Whatever the level, actions you mark as sensitive still pause for a person.
Approval required
Send renewal recommendation to vendor owner
Includes business context, evidence, risk flags, and action history.
Customer-impactingExternal communicationFinancial threshold
Approve Edit Escalate
How it works
From request to recorded result.
-
Start in draft
A person sends every result.
-
Approve each step
The agent acts after a yes.
-
Approve risky steps
Routine steps run on their own.
-
Run routine cases
Sensitive actions still wait.
Every feature runs inside the same six controls.
Identity, access, approvals, supervision, audit and policy apply to every agent and every run, whichever part of the workspace it happens in.
See governance and audit- Identity
- Access
- Approvals
- Supervision
- Audit
- Policy
FAQ
Autonomy levels: questions and answers
Short answers to what teams ask before they put this part of SoftworkerAI to work.
Contact us directlyWho decides when an agent moves up?
The agent owner and admins, based on its approval and rework history.
Can an agent move back down?
Yes. You can tighten an agent at any time.
Do sensitive actions ever skip approval?
No. Actions you mark as sensitive keep their approval gate at every level.
The multiplayer workspace for humans and AI agents is how teams move beyond prompting and safely hand AI real work. It gives people one place to plan, approve, and oversee execution, so they can run work with AI agents they can govern and trust.