Level 0
Draft and hold
The agent prepares the work. A human releases it. This is the default for every new workload, and it is where most sensitive actions stay for good.
Nothing sensitive moves without your say so.
Marshal agents run in draft-and-approve mode by default. Anything that spends money, touches a customer, or changes a record you care about waits for a human. Autonomy is granted by you and every action leaves a receipt.
Founder-led teams use Marshal
What's the risk of anunintended action?
Low. But never zero. That's why Marshal agents cannot spend, send, or delete outside the scope you assign. When it reaches that boundary, Marshal stops, logs the pause, and hands the decision to a person.
The autonomy ladder
Three levels. Every workload starts at zero. Nothing climbs without your signature.
Level 1
Act within limits
The agent executes inside written limits: a spend cap, an approved list of recipients, the record types it may touch. It reports every action.
LEVEL 2
Autonomous action
The agent executes and files the receipt. You review the exceptions. Earned over weeks of clean runs. 1-click revocation.
You decide. We enforce.
Marshal reaches only the systems you open. It stays in the lane you outline. When a decision requires judgment, Marshal stops and waits.
See security policies →Marshal enforces
- Agents operate within the scope you set.
- Spend limits, per action type, in your numbers.
- The off-limits list: actions the agent never takes.
- A receipt for every action: the input, the steps, the action taken, the gate, the time.
- Credentials stay yours. The agent never owns a login.
You decide
- Agent goals and target outcomes.
- Scope and boundaries: what it can spend, who it can contact, what it can change.
- Rules, procedures, and the off-limits list.
- Approval authorities.
- The level of autonomy.
Work at machine speed withagents on the job.
Reimagine your business with Marshal on the team.