Skip to content

Human review for consequential AI actions

Let the agent help.
Keep the final say.

When Kole proposes a consequential action, inspect the operation, adjust its arguments, and approve or reject it before execution continues.

Inspect the proposal. Edit the arguments. Make the decision.

Applies to configured agent-proposed consequential actions, not every query or manual notebook action.

A closer look at Approval Workflow.

Product preview · Sample data
Illustrative workflow

Configured consequential action

Sample approval -- no live execution

No live execution
01Kole proposes

Save weekly revenue query

The agent proposes saving a query after inspecting the connection and schema context.

Proposed action

save_query

name:
weekly_revenue_by_region
connection:
warehouse / analytics

Illustrative proposal

02Queued for review

Execution paused

The action is held because save_query is configured as consequential. No execution until a person reviews.

Pending review

Action:
save_query
Source:
Kole / Quantum Lab
Status:
Awaiting decision

Illustrative status

03Review and decide

Approve or reject

Inspect the arguments, edit if needed, then approve to proceed or reject with feedback.

Approval card

Action:
save_query
Name:
weekly_revenue_by_region
ApproveReject

Controls disabled -- illustrative only

Why Approval Workflow

A clear decision before consequential action.

01

See what is proposed.

Configured consequential agent actions pause with the operation and arguments visible.

A checkpoint in Quantum Lab

02

Adjust the scope.

Edit arguments to refine the target or correct the proposal before approving it.

Inspect + edit

03

Choose what happens next.

Approve the reviewed action or reject it with feedback so Kole can continue from your decision.

Approve or reject

A proposal you can make a decision on

Kole can gather context through read-only discovery. When it proposes a configured consequential action, Quantum Lab presents the operation and its arguments for review.

Check the target and scope, edit the arguments if needed, then approve. If the proposal misses the intent, reject it with feedback. An approved action uses the final arguments; a rejected action does not execute.

Use Quantum Lab to keep that decision beside the analysis it belongs to.

Before you get started

A few useful answers.

Talk through your use case
What triggers an approval?

A proposal from an agent tool configured as consequential. Read-only discovery, such as inspecting schemas or listing connections, can run automatically. This is not an approval gate on every query or manual notebook action.

Can I change the proposal before approving?

Yes. Inspect the operation and arguments, narrow or correct the arguments, then approve. You can also reject with feedback.

What happens if I reject it?

The proposed action does not execute. Kole receives the rejection and any feedback so the conversation can continue.

Give the agent a clear point to pause.

Let Kole gather context, then review configured consequential actions before they run.