AI agents: what they can do and how to approve a run
An agent in Tasked is an identity with an owner, a budget of credits for a period, and a closed list of what it may do - exactly what its owner holds, never more. It never writes anything in the product until a person has seen what it proposes and approved it, unless you turned on apply without a click for it after a preview already ran successfully.
Abrir TaskedWhat an agent holds
Every agent has an owner, the person it inherits permissions from, a credit budget for a period (a day, a week or a month), and a list of what it may do. That list is capped at what the owner themselves holds: if the owner's role shrinks, what the agent may do shrinks with it, because the list is checked again on every run. An agent with no owner refuses to act at all.
By default an agent never applies anything on its own: it proposes a change, a person sees the proposal and approves it, and only then is it written. The Apply without a click switch changes that, but only inside the list the owner already approved and only once a preview has already run successfully; everything else still waits for a person.
Create a new agent
The agent form sets who it is, whose permissions it acts with, and what it may do.
- On the Agents screen (under More > Automation > Agents), click New agent.
- Fill in Name, and describe its purpose in the What it is for field.
- Pick an owner in the Acts as field; the agent inherits that person's permissions.
- Under What it may do, tick only what it actually needs.
- Set Credits per period and Period (a day, a week or a month), and turn on the On switch.
- Click Save.
Reviewing what an agent proposes, and approving it
Every task handed to an agent shows up in the runs list, with who asked and its state: proposed, waiting for you, approved, applied, nothing to do, refused, stopped or failed. Opening a run shows what the agent read, why it is proposing what it proposes, and exactly what would change. Clicking Approve and apply asks for one more confirmation before the change is written under your name, because it is not undone in one tap; you can also stop a run that is still going. When an agent refuses to act, the run says exactly why, for instance that it went over budget, that it is asking for a permission outside its own list, or that there is no preview for this kind of change so nobody may approve it blind.
Preguntas frecuentes
- Can an agent act without anyone approving it?
- Only inside the permission list you already approved for it, and only once a preview has already run successfully. Every other change waits for a person to click approve.
- What happens if an agent's owner leaves the workspace?
- The agent refuses to act, since there is nobody left to inherit permissions from. Give it a different owner.
- Who can use agents?
- Owners, admins and members - anyone the use AI permission applies to. A guest cannot.